From 5cf50930dee214fb641bb95bbf8639b8f76930d9 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Mon, 28 Sep 2026 11:12:10 -0300 Subject: Normal mode's s and S search as addr does, through one shared regexp.zig The sam-style calling convention for mvzr (a line per haystack, . made [^\\n] only when a pattern names \\n) lived in addr.zig; normal mode's s and S called mvzr over the raw selection, where ^ meant the selection's start and . crossed lines. src/regexp.zig now holds the one Regex (compile, find) both call, and fs.zig and pardes.zig drop an unused mvzr import. hxdiff and hxparity stay at their known 17 and 6 mismatches, none of them regex cases. Co-Authored-By: Claude Opus 5.5 --- src/regexp.zig | 104 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 104 insertions(+) create mode 100644 src/regexp.zig (limited to 'src/regexp.zig') diff --git a/src/regexp.zig b/src/regexp.zig new file mode 100644 index 00000000..665e51d5 --- /dev/null +++ b/src/regexp.zig @@ -0,0 +1,104 @@ +//! How pardes runs a regular expression over text: mvzr's, searched the way +//! sam searches (editors/acme/regx.c). `addr` (src/ninep/addr.zig) and +//! normal mode's `s` and `S` (src/normal.zig) both call it. +const std = @import("std"); +const mvzr = @import("mvzr"); + +/// A compiled pattern and how to run it. +/// +/// sam searches the text as lines: `^` and `$` at any line's start and end, +/// and `.` never a newline. mvzr has no such mode (its `^` and `$` are the +/// haystack's ends, its `.` any byte), so each line is its own haystack, and +/// a pattern that names a newline (`\n`) runs over the whole text with its +/// `.`s made `[^\n]`. +/// ponytail: mvzr takes the first alternative that matches, not sam's +/// longest (`gam|gamma` finds `gam`); a search from the middle of a line lets +/// `^` match there unless the pattern starts with it; across lines, `^`, `$` +/// and `[^...]` keep mvzr's meaning. A regex engine of sam's own would lift +/// these; the user chose not to have one. +pub const Regex = struct { + re: mvzr.Regex, + /// The pattern names a newline: it runs over the whole text. + spans: bool, + /// The pattern starts with `^`: a search begun mid-line skips the line. + bol: bool, + + pub fn compile(pat: []const u8) ?Regex { + if (pat.len == 0) return null; + const spans = std.mem.indexOf(u8, pat, "\\n") != null; + var buf: [256]u8 = undefined; + var len: usize = 0; + var i: usize = 0; + var in_class = false; + while (i < pat.len) : (i += 1) { + const c = pat[i]; + const piece: []const u8 = if (c == '\\') piece: { + if (i + 1 >= pat.len) return null; + i += 1; + break :piece pat[i - 1 .. i + 1]; + } else if (c == '.' and spans and !in_class) "[^\\n]" else pat[i .. i + 1]; + if (len + piece.len > buf.len) return null; + if (c == '[') in_class = true; + if (c == ']') in_class = false; + @memcpy(buf[len..][0..piece.len], piece); + len += piece.len; + } + return .{ .re = mvzr.compile(buf[0..len]) orelse return null, .spans = spans, .bol = pat[0] == '^' }; + } + + /// The first match that starts in `from..=last` and ends by `hi`, as + /// offsets into `text`. The text before `from` still says where lines + /// begin. + pub fn find(rx: *const Regex, text: []const u8, from: usize, last: usize, hi: usize) ?struct { start: usize, end: usize } { + if (rx.spans) { + const m = rx.re.matchPos(from, text[0..hi]) orelse return null; + return if (m.start <= last) .{ .start = m.start, .end = m.end } else null; + } + var start = if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; + var at = from - start; + // `^` cannot match in the middle of a line. + if (rx.bol and at > 0) { + start = (std.mem.indexOfScalarPos(u8, text[0..hi], from, '\n') orelse return null) + 1; + at = 0; + } + while (start <= hi and start <= last) { + const end = std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; + const line = text[start..end]; + // matchPos finds nothing at a line's very end, where `$` or an + // empty match still can. + const hit: ?[2]usize = if (at < line.len) + (if (rx.re.matchPos(at, line)) |m| .{ m.start, m.end } else null) + else if (at == line.len and rx.re.isMatch(line[at..])) .{ at, at } else null; + if (hit) |h| { + if (start + h[0] > last) return null; + return .{ .start = start + h[0], .end = start + h[1] }; + } + if (end == hi) return null; + start = end + 1; + at = 0; + } + return null; + } +}; + +test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans lines" { + const text = "alpha beta\nbeta gamma\ngamma\n"; + const Case = struct { pat: []const u8, from: usize, start: usize, end: usize }; + for ([_]Case{ + .{ .pat = "^beta", .from = 0, .start = 11, .end = 15 }, + .{ .pat = "beta$", .from = 0, .start = 6, .end = 10 }, + .{ .pat = "a.*", .from = 0, .start = 0, .end = 10 }, + .{ .pat = "a\\nbeta", .from = 0, .start = 9, .end = 15 }, + .{ .pat = "t.\\nbeta", .from = 0, .start = 8, .end = 15 }, + .{ .pat = "^", .from = 1, .start = 11, .end = 11 }, + }) |c| { + const rx = Regex.compile(c.pat).?; + const m = rx.find(text, c.from, text.len, text.len).?; + try std.testing.expectEqual(c.start, m.start); + try std.testing.expectEqual(c.end, m.end); + } + try std.testing.expect(Regex.compile("a.*a\\nq") != null); + try std.testing.expect((Regex.compile("zzz").?).find(text, 0, text.len, text.len) == null); + try std.testing.expect(Regex.compile("") == null); + try std.testing.expect(Regex.compile("a\\") == null); +} -- cgit v1.3 From 78c60342022bc44307ad789aaef163ab96ebd08b Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Mon, 28 Sep 2026 11:48:17 -0300 Subject: A regular expression search comes back: costly patterns refused, long lines windowed, 300 ms deadline mvzr backtracks with no bound on its work (a*a*a*a*x over a hundred a's takes a second, each repeat multiplying by the haystack length), and a search runs holding the editor's turn, so one pasted pattern froze the editor. pardes does not write or vendor a regex engine, so regexp.zig bounds what it hands mvzr: more than four repeats is refused, a line longer than a window sized from measured worst cases is searched in half-overlapping windows, and a deadline stops the search. addr names each failure; normal s/S keeps what it found. The prescan also stops reading an escaped backslash before n as a newline, and ends a class where mvzr does. Co-Authored-By: Claude Opus 5.5 --- .agents/skills/pardes-9p/SKILL.md | 3 +- docs/fs.md | 12 ++- docs/helix-keys.md | 5 +- src/ninep/addr.zig | 22 ++++- src/normal.zig | 5 +- src/regexp.zig | 190 +++++++++++++++++++++++++++++--------- 6 files changed, 184 insertions(+), 53 deletions(-) (limited to 'src/regexp.zig') diff --git a/.agents/skills/pardes-9p/SKILL.md b/.agents/skills/pardes-9p/SKILL.md index a9cecaf8..330f029d 100644 --- a/.agents/skills/pardes-9p/SKILL.md +++ b/.agents/skills/pardes-9p/SKILL.md @@ -140,7 +140,8 @@ is why copying one onto another is all that acme's `addr=dot`, `dot=addr` and `/pattern/`, `2+1`), whose regexps are mvzr's searched as sam searches: `^`/`$` match at any line's start and end, `.` and `[^...]` never match a newline, the leftmost match wins (the first alternative there, not the -longest), and `/re/` wraps unless `limit` is set. A failed address says +longest), `/re/` wraps unless `limit` is set, and since mvzr backtracks a +pattern with more than four `*`/`+`/`{}` is refused. A failed address says why (`no match for regexp`, `address out of range`) and leaves no address: `data` refuses until the next good one, so a missed target is never written at the old one. Moving `dot` scrolls the pane to it. `limit` bounds a diff --git a/docs/fs.md b/docs/fs.md index 3b49105d..2c565ade 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -280,13 +280,21 @@ the alternatives at that place mvzr takes the first that matches where sam takes the longest (`/gam|gamma/` finds `gam`); in a search begun in the middle of a line, `^` inside an alternation can match there; and in a pattern that spans lines, `^`, `$` and `[^...]` keep mvzr's own meaning. +mvzr backtracks without bound (`a*a*a*a*x` over a hundred `a`s takes a +second), and a search holds the editor, so a pattern with more than four +repeats (`*`, `+`, `{m,n}`) is refused as `regular expression has more than +four repeats`, a long line is searched in overlapping windows (64 KB for +a pattern with at most one repeat, 256, 96 and 40 bytes for two, three and +four), where a match longer than half a window can be missed, and a +search still running after 300 ms fails with `regular expression search +took too long`. pardes has no regex engine of its own on purpose; these are its limits. Normal mode's `s` and `S` search a selection the same way (src/regexp.zig is the one place both call), so `^` there also means a line's start. An address that does not evaluate fails the write with why: `bad address -syntax`, `no match for regexp`, `address out of range` or `bad regular -expression`. A failed write to `addr` leaves no address at all, where acme +syntax`, `no match for regexp`, `address out of range`, `bad regular +expression` or one of the two costs above. A failed write to `addr` leaves no address at all, where acme keeps the old one: until an address is written or `addr` is truncated, reading `addr`, and reading, writing or truncating `data` and `xdata`, fail with `no address: the last one written to addr failed`, so a script that diff --git a/docs/helix-keys.md b/docs/helix-keys.md index f2855b3a..087b83d0 100644 --- a/docs/helix-keys.md +++ b/docs/helix-keys.md @@ -249,7 +249,10 @@ compiles a runtime pattern with no allocator, called the way sam searches (`src/regexp.zig`, shared with a pane's `addr` file): each line is its own haystack, so `^` and `$` match at every line's start and end and `.` never crosses a newline, and a pattern naming `\n` runs over the whole selection. -Where helix's `^` is only the selection's start, this is sam's. +Where helix's `^` is only the selection's start, this is sam's. mvzr +backtracks without bound, so the caps docs/fs.md gives for `addr` hold +here too: more than four repeats will not compile (the selection is left +alone) and a search past 300 ms keeps the matches it found so far. | Key | Behavior | Notes | Status | | --- | --- | --- | --- | diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 347a2d11..69e6dc77 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -14,6 +14,8 @@ fn clip(n: usize) u32 { pub const e_no_match = "no match for regexp"; pub const e_range = "address out of range"; pub const e_regexp = "bad regular expression"; +pub const e_costly = "regular expression has more than four repeats"; +pub const e_slow = "regular expression search took too long"; pub const e_syntax = "bad address syntax"; pub const Addr = struct { @@ -196,8 +198,8 @@ pub const Addr = struct { a.err = "no previous regular expression"; return null; } - const rx = regexp_.Regex.compile(pat) orelse { - a.err = e_regexp; + const rx = regexp_.Regex.compile(pat) catch |err| { + a.err = if (err == error.TooCostly) e_costly else e_regexp; return null; }; const found = if (back) found: { @@ -205,7 +207,10 @@ pub const Addr = struct { var before: ?Range = null; var at: usize = 0; while (at <= a.text.len) { - const m = rx.find(a.text, at, a.text.len, a.text.len) orelse break; + const m = (rx.find(a.text, at, a.text.len, a.text.len) catch { + a.err = e_slow; + return null; + }) orelse break; const found_range: Range = .{ .q0 = clip(m.start), .q1 = clip(m.end) }; if (found_range.q1 <= r.q0) before = found_range; last = found_range; @@ -215,7 +220,14 @@ pub const Addr = struct { } else found: { const hi = if (a.lim) |l| @min(@as(usize, l.q1), a.text.len) else a.text.len; const from = @min(@as(usize, r.q1), hi); - const m = rx.find(a.text, from, hi, hi) orelse (if (a.lim != null or from == 0) null else rx.find(a.text, 0, from - 1, hi)) orelse break :found null; + const ahead = rx.find(a.text, from, hi, hi) catch { + a.err = e_slow; + return null; + }; + const m = ahead orelse (if (a.lim != null or from == 0) null else rx.find(a.text, 0, from - 1, hi) catch { + a.err = e_slow; + return null; + }) orelse break :found null; break :found Range{ .q0 = clip(m.start), .q1 = clip(m.end) }; }; return found orelse { @@ -366,7 +378,7 @@ test "the address language, form by form" { for ([_][2][]const u8{ .{ "zzz", e_syntax }, .{ "/nomatch/", e_no_match }, .{ "99", e_range }, .{ "#999", e_range }, .{ "/a[/", e_regexp }, .{ "/(a/", e_regexp }, - .{ "/*a/", e_regexp }, + .{ "/*a/", e_regexp }, .{ "/a*b*c*d*e*/", e_costly }, }) |c| { _ = th.wr(p, addr, "#0"); try testing.expectEqualStrings(c[1], th.wr(p, addr, c[0]).reply.ename); diff --git a/src/normal.zig b/src/normal.zig index 8c5f232d..09e06b13 100644 --- a/src/normal.zig +++ b/src/normal.zig @@ -110,7 +110,8 @@ pub fn applySelRegex(p: *Pardes, pane: *Pane, t: *Text, pat: []const u8, split: var at = from; var piece = from; // split: where the next piece begins while (at < to and m < Text.max_selections) { - const hit_at = re.find(hay_all, at, to, to) orelse break; + // A search too slow to finish keeps what it found so far. + const hit_at = (re.find(hay_all, at, to, to) catch break) orelse break; if (split) { out[m] = .{ .anchor = piece, .head = hit_at.start }; m += 1; @@ -127,7 +128,7 @@ pub fn applySelRegex(p: *Pardes, pane: *Pane, t: *Text, pat: []const u8, split: m += 1; } } - } + } else |_| {} if (m == 0) { // the text CAN move under an armed prompt (a tag chord runs a // builtin), and these are raw offsets into the surface as it was diff --git a/src/regexp.zig b/src/regexp.zig index 665e51d5..c758ee81 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -3,6 +3,7 @@ //! normal mode's `s` and `S` (src/normal.zig) both call it. const std = @import("std"); const mvzr = @import("mvzr"); +const hosted = @import("pardes.zig").hosted; /// A compiled pattern and how to run it. /// @@ -11,67 +12,140 @@ const mvzr = @import("mvzr"); /// haystack's ends, its `.` any byte), so each line is its own haystack, and /// a pattern that names a newline (`\n`) runs over the whole text with its /// `.`s made `[^\n]`. +/// +/// mvzr backtracks, and has no bound on its work: `a*a*a*a*x` over a line of +/// a hundred `a`s takes a second, and each repeat multiplies the cost by the +/// haystack's length. The search runs with the editor's turn, so it must +/// come back: a pattern with more than four repeats (`*`, `+`, `{}`) is +/// refused, the more repeats the shorter the stretch of a line one mvzr +/// call is given (a line longer than that is searched in overlapping +/// windows), and a search that has run for 300 ms stops with `TooSlow`. +/// /// ponytail: mvzr takes the first alternative that matches, not sam's /// longest (`gam|gamma` finds `gam`); a search from the middle of a line lets /// `^` match there unless the pattern starts with it; across lines, `^`, `$` -/// and `[^...]` keep mvzr's meaning. A regex engine of sam's own would lift -/// these; the user chose not to have one. +/// and `[^...]` keep mvzr's meaning; in a windowed line a match longer than +/// half a window may be missed or `^` match at a window's start. A regex +/// engine of sam's own would lift these; the user chose not to have one. pub const Regex = struct { re: mvzr.Regex, /// The pattern names a newline: it runs over the whole text. spans: bool, /// The pattern starts with `^`: a search begun mid-line skips the line. bol: bool, + /// The most bytes one mvzr call searches. + window: usize, + /// When searches with this pattern give up, in milliseconds. + deadline: i64, - pub fn compile(pat: []const u8) ?Regex { - if (pat.len == 0) return null; - const spans = std.mem.indexOf(u8, pat, "\\n") != null; + pub const Error = error{ Bad, TooCostly }; + + pub fn compile(pat: []const u8) Error!Regex { + if (pat.len == 0) return error.Bad; var buf: [256]u8 = undefined; var len: usize = 0; - var i: usize = 0; - var in_class = false; - while (i < pat.len) : (i += 1) { - const c = pat[i]; - const piece: []const u8 = if (c == '\\') piece: { - if (i + 1 >= pat.len) return null; - i += 1; - break :piece pat[i - 1 .. i + 1]; - } else if (c == '.' and spans and !in_class) "[^\\n]" else pat[i .. i + 1]; - if (len + piece.len > buf.len) return null; - if (c == '[') in_class = true; - if (c == ']') in_class = false; - @memcpy(buf[len..][0..piece.len], piece); - len += piece.len; + var spans = false; + var repeats: usize = 0; + // Twice over the pattern: the first pass learns whether it names a + // newline, which the second needs to rewrite its `.`s. + for ([2]bool{ false, true }) |emit| { + var i: usize = 0; + // mvzr ends a class at its first unescaped `]`, even one that + // comes first (`[]a]` is an empty class, then `a]`); a `]` member + // is written `\]`. + var in_class = false; + var quantified = false; + while (i < pat.len) : (i += 1) { + const c = pat[i]; + var piece: []const u8 = pat[i .. i + 1]; + if (c == '\\') { + if (i + 1 >= pat.len) return error.Bad; + i += 1; + piece = pat[i - 1 .. i + 1]; + if (pat[i] == 'n') spans = true; + quantified = false; + } else if (in_class) { + in_class = c != ']'; + } else if (c == '[') { + in_class = true; + quantified = false; + } else if (c == '*' or c == '+' or c == '{' or (c == '?' and !quantified)) { + // A lazy or possessive mark after a repeat is the same repeat. + if (!emit and !quantified and c != '?') repeats += 1; + quantified = c != '?'; + } else { + if (c == '.' and spans) piece = "[^\\n]"; + quantified = false; + } + if (!emit) continue; + if (len + piece.len > buf.len) return error.Bad; + @memcpy(buf[len..][0..piece.len], piece); + len += piece.len; + } + if (repeats > 4) return error.TooCostly; } - return .{ .re = mvzr.compile(buf[0..len]) orelse return null, .spans = spans, .bol = pat[0] == '^' }; + // Sizes from timing mvzr: one call over a window of the worst text + // for its pattern (`a*a*a*a*x` over `a`s) takes under 40 ms, so the + // deadline is never overrun by much. + const window: usize = switch (repeats) { + 0, 1 => 64 * 1024, + 2 => 256, + 3 => 96, + else => 40, + }; + return .{ + .re = mvzr.compile(buf[0..len]) orelse return error.Bad, + .spans = spans, + .bol = pat[0] == '^', + .window = window, + .deadline = nowMs() + 300, + }; } + fn nowMs() i64 { + if (comptime !hosted) return 0; + var ts: std.c.timespec = undefined; + _ = std.c.clock_gettime(.MONOTONIC, &ts); + return @as(i64, ts.sec) * 1000 + @divTrunc(@as(i64, ts.nsec), 1_000_000); + } + + pub const Match = struct { start: usize, end: usize }; + /// The first match that starts in `from..=last` and ends by `hi`, as /// offsets into `text`. The text before `from` still says where lines /// begin. - pub fn find(rx: *const Regex, text: []const u8, from: usize, last: usize, hi: usize) ?struct { start: usize, end: usize } { - if (rx.spans) { - const m = rx.re.matchPos(from, text[0..hi]) orelse return null; - return if (m.start <= last) .{ .start = m.start, .end = m.end } else null; - } - var start = if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; + pub fn find(rx: *const Regex, text: []const u8, from: usize, last: usize, hi: usize) error{TooSlow}!?Match { + var start: usize = if (rx.spans) 0 else if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; var at = from - start; // `^` cannot match in the middle of a line. - if (rx.bol and at > 0) { + if (!rx.spans and rx.bol and at > 0) { start = (std.mem.indexOfScalarPos(u8, text[0..hi], from, '\n') orelse return null) + 1; at = 0; } while (start <= hi and start <= last) { - const end = std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; + const end = if (rx.spans) hi else std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; const line = text[start..end]; - // matchPos finds nothing at a line's very end, where `$` or an - // empty match still can. - const hit: ?[2]usize = if (at < line.len) - (if (rx.re.matchPos(at, line)) |m| .{ m.start, m.end } else null) - else if (at == line.len and rx.re.isMatch(line[at..])) .{ at, at } else null; - if (hit) |h| { - if (start + h[0] > last) return null; - return .{ .start = start + h[0], .end = start + h[1] }; + // A line longer than a window is searched a window at a time, + // each overlapping the last by half; a match ending at a window's + // edge that is not the line's is left to the next window, whose + // `$` is not a false line end. + var lo = at -| at % @max(1, rx.window / 2); + while (true) { + if (comptime hosted) if (nowMs() > rx.deadline) return error.TooSlow; + const top = @min(line.len, lo + rx.window); + const hay = line[lo..top]; + const pos = @max(at, lo) - lo; + // matchPos finds nothing at a haystack's very end, where `$` + // or an empty match still can. + const hit: ?[2]usize = if (pos < hay.len) + (if (rx.re.matchPos(pos, hay)) |m| .{ lo + m.start, lo + m.end } else null) + else if (pos == hay.len and top == line.len and rx.re.isMatch(hay[pos..])) .{ lo + pos, lo + pos } else null; + if (hit) |h| if (top == line.len or h[1] < top) { + if (start + h[0] > last) return null; + return .{ .start = start + h[0], .end = start + h[1] }; + }; + if (top == line.len or (rx.bol and !rx.spans)) break; + lo += rx.window / 2; } if (end == hi) return null; start = end + 1; @@ -92,13 +166,45 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li .{ .pat = "t.\\nbeta", .from = 0, .start = 8, .end = 15 }, .{ .pat = "^", .from = 1, .start = 11, .end = 11 }, }) |c| { - const rx = Regex.compile(c.pat).?; - const m = rx.find(text, c.from, text.len, text.len).?; + const rx = try Regex.compile(c.pat); + const m = (try rx.find(text, c.from, text.len, text.len)).?; try std.testing.expectEqual(c.start, m.start); try std.testing.expectEqual(c.end, m.end); } - try std.testing.expect(Regex.compile("a.*a\\nq") != null); - try std.testing.expect((Regex.compile("zzz").?).find(text, 0, text.len, text.len) == null); - try std.testing.expect(Regex.compile("") == null); - try std.testing.expect(Regex.compile("a\\") == null); + _ = try Regex.compile("a.*a\\nq"); + try std.testing.expect(try (try Regex.compile("zzz")).find(text, 0, text.len, text.len) == null); + try std.testing.expectError(error.Bad, Regex.compile("")); + try std.testing.expectError(error.Bad, Regex.compile("a\\")); +} + +test "a quoted backslash before n is no newline, and a class ends where mvzr ends it" { + // `\\n` is a backslash then an n: the pattern stays on one line, so its + // `.` is not made [^\n] and does match within the line. + const slash = try Regex.compile("a\\\\n."); + try std.testing.expect(!slash.spans); + try std.testing.expectEqual(@as(usize, 0), (try slash.find("a\\nX", 0, 4, 4)).?.start); + // `[\]x]` is a class of `]` and `x`: the escaped `]` does not end it, + // so the `*` after it is a repeat and the `.` a newline's rewrite. + const class = try Regex.compile("[\\]x]*y.\\n"); + try std.testing.expect(class.spans); + try std.testing.expectEqual(@as(usize, 1), (try class.find("-]x]y!\nz", 0, 8, 8)).?.start); + try std.testing.expect(try class.find("-]x]y\n\n", 0, 7, 7) == null); + // mvzr reads `[]x]` as an empty class then `x]`, which nothing matches. + try std.testing.expect(try (try Regex.compile("[]x]")).find("x]", 0, 2, 2) == null); +} + +test "a pattern that would backtrack without end comes back promptly, refused or cut short" { + try std.testing.expectError(error.TooCostly, Regex.compile("a*a*a*a*a*x")); + // Four repeats over a long line of what they match: windows keep each + // mvzr call small, and the deadline ends the whole search. + var text: [20000]u8 = @splat('a'); + for (0..20) |i| text[i * 1000 + 999] = '\n'; + for ([_][]const u8{ "a*a*a*a*x", ".*.*.*x", ".*.*x", "(a*)*x", "(a|a)*x" }) |pat| { + const rx = try Regex.compile(pat); + const before = Regex.nowMs(); + if (rx.find(&text, 0, text.len, text.len)) |m| { + try std.testing.expect(m == null); + } else |err| try std.testing.expectEqual(error.TooSlow, err); + if (comptime hosted) try std.testing.expect(Regex.nowMs() - before < 2000); + } } -- cgit v1.3 From c8a3ecc3bcb36f0259f4d8ad2e0cdfcdc3183c9f Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Mon, 28 Sep 2026 12:39:43 -0300 Subject: A regular expression search is bounded by a step budget patched into mvzr, not windows and a repeat cap The windows returned wrong matches: a candidate reaching a window's edge was left to the next window, half a window on, which could answer a match starting mid-token rather than the leftmost, and addr then pointed data's next write at the wrong bytes. The repeat cap missed mvzr's own worst case, a chain of a?, and alternation under a repeat, each exponential inside one mvzr call the deadline could not interrupt; and it refused ordinary s/S patterns. build.zig now patches the fetched mvzr at build time with a step counter on its backtracking recursion (matchPattern), so a fresh fetch keeps it and a moved anchor stops the build; regexp.zig gives each compiled pattern a budget, about 300 ms here, and a search that spends it fails as taking too long. Windows, the cap and their special cases are gone, and matches are exact again. Co-Authored-By: Claude Opus 5.5 --- .agents/skills/pardes-9p/SKILL.md | 5 +- build.zig | 32 +++++++- docs/fs.md | 17 ++--- docs/helix-keys.md | 6 +- src/ninep/addr.zig | 7 +- src/normal.zig | 3 +- src/regexp.zig | 151 ++++++++++++++++---------------------- 7 files changed, 114 insertions(+), 107 deletions(-) (limited to 'src/regexp.zig') diff --git a/.agents/skills/pardes-9p/SKILL.md b/.agents/skills/pardes-9p/SKILL.md index 77acade3..393b0b0f 100644 --- a/.agents/skills/pardes-9p/SKILL.md +++ b/.agents/skills/pardes-9p/SKILL.md @@ -147,8 +147,9 @@ is why copying one onto another is all that acme's `addr=dot`, `dot=addr` and `/pattern/`, `2+1`), whose regexps are mvzr's searched as sam searches: `^`/`$` match at any line's start and end, `.` and `[^...]` never match a newline, the leftmost match wins (the first alternative there, not the -longest), `/re/` wraps unless `limit` is set, and since mvzr backtracks a -pattern with more than four `*`/`+`/`{}` is refused. A failed address says +longest), `/re/` wraps unless `limit` is set, and a search that backtracks +past a step budget (about 300 ms) fails with `regular expression search +took too long`. A failed address says why (`no match for regexp`, `address out of range`) and leaves no address: `data` refuses until the next good one, so a missed target is never written at the old one. Moving `dot` scrolls the pane to it. `limit` bounds a diff --git a/build.zig b/build.zig index f0675a70..666e792d 100644 --- a/build.zig +++ b/build.zig @@ -418,7 +418,37 @@ pub fn build(b: *std.Build) void { .root_source_file = b.path("src/web/zstbi.zig"), }); - const mvzr_mod = b.dependency("mvzr", .{ .target = target, .optimize = optimize }).module("mvzr"); + // mvzr backtracks with no bound on its work (its own test "do not make + // this test any longer": `a?` n times is 2^n steps), and pardes searches + // with it holding the editor. Its matcher gets a step budget, patched into + // the fetched source here at build time so that a fresh fetch keeps it: + // each entry to matchPattern, which its backtracking recurses through, + // spends a step, and with none left the match fails and says so + // (src/regexp.zig). A bump that moves the anchor stops the build. + const mvzr_mod = mvzr: { + const dep = b.dependency("mvzr", .{ .target = target, .optimize = optimize }); + const src = std.Io.Dir.cwd().readFileAlloc(io, dep.path("src/mvzr.zig").getPath(b), b.allocator, .limited(1 << 20)) catch @panic("read mvzr"); + const anchor = "fn matchPattern(patt: []const RegOp, sets: []const CharSet, haystack: []const u8, i_in: usize) ?OpMatch {\n"; + const at = (std.mem.indexOf(u8, src, anchor) orelse @panic("mvzr's matchPattern changed: redo its step budget in build.zig")) + anchor.len; + const spend = + \\ if (steps_left == 0) { + \\ exhausted = true; + \\ return null; + \\ } + \\ steps_left -= 1; + \\ + ; + const budget = + \\ + \\/// pardes's patch (its build.zig): matchPattern entries left before a + \\/// match gives up, and whether one did. Set by the caller per search. + \\pub threadlocal var steps_left: u64 = std.math.maxInt(u64); + \\pub threadlocal var exhausted: bool = false; + \\ + ; + const patched = b.fmt("{s}{s}{s}{s}", .{ src[0..at], spend, src[at..], budget }); + break :mvzr b.createModule(.{ .root_source_file = b.addWriteFiles().add("mvzr.zig", patched), .target = target, .optimize = optimize }); + }; // The C heap every C library with an allocator hook is handed: the core's // tree-sitter and fonts, and pdf.zig's MuPDF, which is a module of its own. diff --git a/docs/fs.md b/docs/fs.md index c9b268b0..a2d8a9b2 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -288,21 +288,20 @@ the alternatives at that place mvzr takes the first that matches where sam takes the longest (`/gam|gamma/` finds `gam`); in a search begun in the middle of a line, `^` inside an alternation can match there; and in a pattern that spans lines, `^`, `$` and `[^...]` keep mvzr's own meaning. -mvzr backtracks without bound (`a*a*a*a*x` over a hundred `a`s takes a -second), and a search holds the editor, so a pattern with more than four -repeats (`*`, `+`, `{m,n}`) is refused as `regular expression has more than -four repeats`, a long line is searched in overlapping windows (64 KB for -a pattern with at most one repeat, 256, 96 and 40 bytes for two, three and -four), where a match longer than half a window can be missed, and a -search still running after 300 ms fails with `regular expression search -took too long`. +mvzr backtracks without bound of its own (`a?` twenty times then twenty +`a`s is 2^20 steps from each place it tries), and a search holds the editor, so pardes patches a +step budget into mvzr's matcher (build.zig): a search that spends it, +about 300 ms, fails with `regular expression search took too long` rather +than answer a match it is not sure of. Ordinary patterns spend a few +thousand steps; what runs out is exponential backtracking, and a quadratic +pattern over a very long line (`\s*(\w+)\s*=` over 20 KB of letters). pardes has no regex engine of its own on purpose; these are its limits. Normal mode's `s` and `S` search a selection the same way (src/regexp.zig is the one place both call), so `^` there also means a line's start. An address that does not evaluate fails the write with why: `bad address syntax`, `no match for regexp`, `address out of range`, `bad regular -expression` or one of the two costs above. A failed write to `addr` leaves no address at all, where acme +expression` or `regular expression search took too long`. A failed write to `addr` leaves no address at all, where acme keeps the old one: until an address is written or `addr` is truncated, reading `addr`, and reading, writing or truncating `data` and `xdata`, fail with `no address: the last one written to addr failed`, so a script that diff --git a/docs/helix-keys.md b/docs/helix-keys.md index 087b83d0..41f9b364 100644 --- a/docs/helix-keys.md +++ b/docs/helix-keys.md @@ -250,9 +250,9 @@ compiles a runtime pattern with no allocator, called the way sam searches haystack, so `^` and `$` match at every line's start and end and `.` never crosses a newline, and a pattern naming `\n` runs over the whole selection. Where helix's `^` is only the selection's start, this is sam's. mvzr -backtracks without bound, so the caps docs/fs.md gives for `addr` hold -here too: more than four repeats will not compile (the selection is left -alone) and a search past 300 ms keeps the matches it found so far. +backtracks without bound of its own, so the step budget docs/fs.md gives +for `addr` holds here too: a search that runs out of it (about 300 ms) +keeps the matches it found so far. | Key | Behavior | Notes | Status | | --- | --- | --- | --- | diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 69e6dc77..cc7ca632 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -14,7 +14,6 @@ fn clip(n: usize) u32 { pub const e_no_match = "no match for regexp"; pub const e_range = "address out of range"; pub const e_regexp = "bad regular expression"; -pub const e_costly = "regular expression has more than four repeats"; pub const e_slow = "regular expression search took too long"; pub const e_syntax = "bad address syntax"; @@ -198,8 +197,8 @@ pub const Addr = struct { a.err = "no previous regular expression"; return null; } - const rx = regexp_.Regex.compile(pat) catch |err| { - a.err = if (err == error.TooCostly) e_costly else e_regexp; + var rx = regexp_.Regex.compile(pat) catch { + a.err = e_regexp; return null; }; const found = if (back) found: { @@ -378,7 +377,7 @@ test "the address language, form by form" { for ([_][2][]const u8{ .{ "zzz", e_syntax }, .{ "/nomatch/", e_no_match }, .{ "99", e_range }, .{ "#999", e_range }, .{ "/a[/", e_regexp }, .{ "/(a/", e_regexp }, - .{ "/*a/", e_regexp }, .{ "/a*b*c*d*e*/", e_costly }, + .{ "/*a/", e_regexp }, }) |c| { _ = th.wr(p, addr, "#0"); try testing.expectEqualStrings(c[1], th.wr(p, addr, c[0]).reply.ename); diff --git a/src/normal.zig b/src/normal.zig index 09e06b13..da707e26 100644 --- a/src/normal.zig +++ b/src/normal.zig @@ -92,7 +92,8 @@ pub fn applySelRegex(p: *Pardes, pane: *Pane, t: *Text, pat: []const u8, split: const snap = pane.sel_snap[0..pane.nsel_snap]; var out: [Text.max_selections]modal.Selection = undefined; var m: usize = 0; - if (regexp.Regex.compile(pat)) |re| { + if (regexp.Regex.compile(pat)) |compiled| { + var re = compiled; var hay_all = text; if (for (pat) |c| { if (std.ascii.isUpper(c)) break false; diff --git a/src/regexp.zig b/src/regexp.zig index c758ee81..fe68585d 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -2,8 +2,8 @@ //! sam searches (editors/acme/regx.c). `addr` (src/ninep/addr.zig) and //! normal mode's `s` and `S` (src/normal.zig) both call it. const std = @import("std"); +const builtin = @import("builtin"); const mvzr = @import("mvzr"); -const hosted = @import("pardes.zig").hosted; /// A compiled pattern and how to run it. /// @@ -13,39 +13,39 @@ const hosted = @import("pardes.zig").hosted; /// a pattern that names a newline (`\n`) runs over the whole text with its /// `.`s made `[^\n]`. /// -/// mvzr backtracks, and has no bound on its work: `a*a*a*a*x` over a line of -/// a hundred `a`s takes a second, and each repeat multiplies the cost by the -/// haystack's length. The search runs with the editor's turn, so it must -/// come back: a pattern with more than four repeats (`*`, `+`, `{}`) is -/// refused, the more repeats the shorter the stretch of a line one mvzr -/// call is given (a line longer than that is searched in overlapping -/// windows), and a search that has run for 300 ms stops with `TooSlow`. +/// mvzr backtracks, and has no bound on its work of its own: `a?` twenty +/// times then twenty `a`s is 2^20 steps from each start, `a*a*a*a*x` over a line of a +/// hundred `a`s a second. The search runs with the editor's turn, so it must +/// come back: build.zig patches a step budget into mvzr's matcher, and a +/// search that spends `budget` steps (about 300 ms here) stops with +/// `TooSlow`, found nothing rather than something wrong. /// /// ponytail: mvzr takes the first alternative that matches, not sam's /// longest (`gam|gamma` finds `gam`); a search from the middle of a line lets /// `^` match there unless the pattern starts with it; across lines, `^`, `$` -/// and `[^...]` keep mvzr's meaning; in a windowed line a match longer than -/// half a window may be missed or `^` match at a window's start. A regex -/// engine of sam's own would lift these; the user chose not to have one. +/// and `[^...]` keep mvzr's meaning; and a quadratic pattern over a long +/// enough line (`\s*(\w+)\s*=` over 20 KB of letters) runs out of budget. A +/// regex engine of sam's own would lift these; the user chose not to have +/// one. pub const Regex = struct { re: mvzr.Regex, /// The pattern names a newline: it runs over the whole text. spans: bool, /// The pattern starts with `^`: a search begun mid-line skips the line. bol: bool, - /// The most bytes one mvzr call searches. - window: usize, - /// When searches with this pattern give up, in milliseconds. - deadline: i64, + /// mvzr steps left to every search with this compiled pattern together, + /// so that addr's backward scan and s/S's many calls share one bound. + steps: u64 = budget, - pub const Error = error{ Bad, TooCostly }; + /// Timed on this machine's ReleaseSafe build: 10M steps of the worst + /// patterns take 45-80 ms, and a Debug build is ten times slower. + pub const budget: u64 = if (builtin.mode == .Debug) 4_000_000 else 32_000_000; - pub fn compile(pat: []const u8) Error!Regex { + pub fn compile(pat: []const u8) error{Bad}!Regex { if (pat.len == 0) return error.Bad; var buf: [256]u8 = undefined; var len: usize = 0; var spans = false; - var repeats: usize = 0; // Twice over the pattern: the first pass learns whether it names a // newline, which the second needs to rewrite its `.`s. for ([2]bool{ false, true }) |emit| { @@ -54,7 +54,6 @@ pub const Regex = struct { // comes first (`[]a]` is an empty class, then `a]`); a `]` member // is written `\]`. var in_class = false; - var quantified = false; while (i < pat.len) : (i += 1) { const c = pat[i]; var piece: []const u8 = pat[i .. i + 1]; @@ -63,58 +62,35 @@ pub const Regex = struct { i += 1; piece = pat[i - 1 .. i + 1]; if (pat[i] == 'n') spans = true; - quantified = false; } else if (in_class) { in_class = c != ']'; } else if (c == '[') { in_class = true; - quantified = false; - } else if (c == '*' or c == '+' or c == '{' or (c == '?' and !quantified)) { - // A lazy or possessive mark after a repeat is the same repeat. - if (!emit and !quantified and c != '?') repeats += 1; - quantified = c != '?'; - } else { - if (c == '.' and spans) piece = "[^\\n]"; - quantified = false; + } else if (c == '.' and spans) { + piece = "[^\\n]"; } if (!emit) continue; if (len + piece.len > buf.len) return error.Bad; @memcpy(buf[len..][0..piece.len], piece); len += piece.len; } - if (repeats > 4) return error.TooCostly; } - // Sizes from timing mvzr: one call over a window of the worst text - // for its pattern (`a*a*a*a*x` over `a`s) takes under 40 ms, so the - // deadline is never overrun by much. - const window: usize = switch (repeats) { - 0, 1 => 64 * 1024, - 2 => 256, - 3 => 96, - else => 40, - }; return .{ .re = mvzr.compile(buf[0..len]) orelse return error.Bad, .spans = spans, .bol = pat[0] == '^', - .window = window, - .deadline = nowMs() + 300, }; } - fn nowMs() i64 { - if (comptime !hosted) return 0; - var ts: std.c.timespec = undefined; - _ = std.c.clock_gettime(.MONOTONIC, &ts); - return @as(i64, ts.sec) * 1000 + @divTrunc(@as(i64, ts.nsec), 1_000_000); - } - pub const Match = struct { start: usize, end: usize }; /// The first match that starts in `from..=last` and ends by `hi`, as /// offsets into `text`. The text before `from` still says where lines /// begin. - pub fn find(rx: *const Regex, text: []const u8, from: usize, last: usize, hi: usize) error{TooSlow}!?Match { + pub fn find(rx: *Regex, text: []const u8, from: usize, last: usize, hi: usize) error{TooSlow}!?Match { + mvzr.steps_left = rx.steps; + mvzr.exhausted = false; + defer rx.steps = mvzr.steps_left; var start: usize = if (rx.spans) 0 else if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; var at = from - start; // `^` cannot match in the middle of a line. @@ -125,27 +101,15 @@ pub const Regex = struct { while (start <= hi and start <= last) { const end = if (rx.spans) hi else std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; const line = text[start..end]; - // A line longer than a window is searched a window at a time, - // each overlapping the last by half; a match ending at a window's - // edge that is not the line's is left to the next window, whose - // `$` is not a false line end. - var lo = at -| at % @max(1, rx.window / 2); - while (true) { - if (comptime hosted) if (nowMs() > rx.deadline) return error.TooSlow; - const top = @min(line.len, lo + rx.window); - const hay = line[lo..top]; - const pos = @max(at, lo) - lo; - // matchPos finds nothing at a haystack's very end, where `$` - // or an empty match still can. - const hit: ?[2]usize = if (pos < hay.len) - (if (rx.re.matchPos(pos, hay)) |m| .{ lo + m.start, lo + m.end } else null) - else if (pos == hay.len and top == line.len and rx.re.isMatch(hay[pos..])) .{ lo + pos, lo + pos } else null; - if (hit) |h| if (top == line.len or h[1] < top) { - if (start + h[0] > last) return null; - return .{ .start = start + h[0], .end = start + h[1] }; - }; - if (top == line.len or (rx.bol and !rx.spans)) break; - lo += rx.window / 2; + // matchPos finds nothing at a haystack's very end, where `$` or + // an empty match still can. + const hit: ?[2]usize = if (at < line.len) + (if (rx.re.matchPos(at, line)) |m| .{ m.start, m.end } else null) + else if (at == line.len and rx.re.isMatch(line[at..])) .{ at, at } else null; + if (mvzr.exhausted) return error.TooSlow; + if (hit) |h| { + if (start + h[0] > last) return null; + return .{ .start = start + h[0], .end = start + h[1] }; } if (end == hi) return null; start = end + 1; @@ -166,13 +130,14 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li .{ .pat = "t.\\nbeta", .from = 0, .start = 8, .end = 15 }, .{ .pat = "^", .from = 1, .start = 11, .end = 11 }, }) |c| { - const rx = try Regex.compile(c.pat); + var rx = try Regex.compile(c.pat); const m = (try rx.find(text, c.from, text.len, text.len)).?; try std.testing.expectEqual(c.start, m.start); try std.testing.expectEqual(c.end, m.end); } _ = try Regex.compile("a.*a\\nq"); - try std.testing.expect(try (try Regex.compile("zzz")).find(text, 0, text.len, text.len) == null); + var none = try Regex.compile("zzz"); + try std.testing.expect(try none.find(text, 0, text.len, text.len) == null); try std.testing.expectError(error.Bad, Regex.compile("")); try std.testing.expectError(error.Bad, Regex.compile("a\\")); } @@ -180,31 +145,43 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li test "a quoted backslash before n is no newline, and a class ends where mvzr ends it" { // `\\n` is a backslash then an n: the pattern stays on one line, so its // `.` is not made [^\n] and does match within the line. - const slash = try Regex.compile("a\\\\n."); + var slash = try Regex.compile("a\\\\n."); try std.testing.expect(!slash.spans); try std.testing.expectEqual(@as(usize, 0), (try slash.find("a\\nX", 0, 4, 4)).?.start); // `[\]x]` is a class of `]` and `x`: the escaped `]` does not end it, // so the `*` after it is a repeat and the `.` a newline's rewrite. - const class = try Regex.compile("[\\]x]*y.\\n"); + var class = try Regex.compile("[\\]x]*y.\\n"); try std.testing.expect(class.spans); try std.testing.expectEqual(@as(usize, 1), (try class.find("-]x]y!\nz", 0, 8, 8)).?.start); try std.testing.expect(try class.find("-]x]y\n\n", 0, 7, 7) == null); // mvzr reads `[]x]` as an empty class then `x]`, which nothing matches. - try std.testing.expect(try (try Regex.compile("[]x]")).find("x]", 0, 2, 2) == null); + var empty = try Regex.compile("[]x]"); + try std.testing.expect(try empty.find("x]", 0, 2, 2) == null); } -test "a pattern that would backtrack without end comes back promptly, refused or cut short" { - try std.testing.expectError(error.TooCostly, Regex.compile("a*a*a*a*a*x")); - // Four repeats over a long line of what they match: windows keep each - // mvzr call small, and the deadline ends the whole search. - var text: [20000]u8 = @splat('a'); - for (0..20) |i| text[i * 1000 + 999] = '\n'; - for ([_][]const u8{ "a*a*a*a*x", ".*.*.*x", ".*.*x", "(a*)*x", "(a|a)*x" }) |pat| { - const rx = try Regex.compile(pat); - const before = Regex.nowMs(); - if (rx.find(&text, 0, text.len, text.len)) |m| { - try std.testing.expect(m == null); - } else |err| try std.testing.expectEqual(error.TooSlow, err); - if (comptime hosted) try std.testing.expect(Regex.nowMs() - before < 2000); +test "a search that would backtrack without end runs out of budget, promptly" { + var line: [20000]u8 = @splat('a'); + for ([_][]const u8{ "(a|ab)*c", "a*a*a*a*a*x", ".*.*.*x" }) |pat| { + var rx = try Regex.compile(pat); + try std.testing.expectError(error.TooSlow, rx.find(&line, 0, line.len, line.len)); } + // As long a chain as mvzr compiles, over runs one `a` short of it: 2^20 + // steps from every start. + for (0..line.len / 20) |i| line[i * 20 + 19] = 'b'; + var chain = try Regex.compile("a?" ** 20 ++ "a" ** 20); + try std.testing.expectError(error.TooSlow, chain.find(&line, 0, line.len, line.len)); +} + +test "the match is the leftmost, however long the line" { + // A window cut into the identifier would find `x... =` from its middle. + const text = "let " ++ "x" ** 100 ++ " = 1"; + var assign = try Regex.compile("\\s*(\\w+)\\s*="); + const m = (try assign.find(text, 0, text.len, text.len)).?; + try std.testing.expectEqual(@as(usize, 3), m.start); + try std.testing.expectEqual(@as(usize, 106), m.end); + // Repeats are not counted or capped. + var five = try Regex.compile("a*b*c*d*e*f"); + const f = (try five.find("xxaabbf", 0, 7, 7)).?; + try std.testing.expectEqual(@as(usize, 2), f.start); + try std.testing.expectEqual(@as(usize, 7), f.end); } -- cgit v1.3 From ff867666106fa85bd91341beb9e8121b2f99bc5f Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Mon, 28 Sep 2026 18:58:24 -0300 Subject: ^ in a pattern with \n matches at every line start, and a ^ or $ that cannot work is refused A pattern naming a newline runs over the whole text, where mvzr's ^ is only the text's start: /^def .*\n/ from #0 failed, and Edit ,x/^def .*\n/.../ silently did nothing. A leading ^ is now tried at each line start from the search's own; $ just before \n is dropped, as it changes nothing; any other ^ or $ in such a pattern is refused with why, since mvzr would read it as the text's ends. ^ inside an alternation holds only where a search starts, an mvzr limit, documented. Co-Authored-By: Claude Opus 5.5 --- .agents/skills/pardes-9p/SKILL.md | 5 ++++- docs/fs.md | 10 +++++++++- src/ninep/addr.zig | 4 ++-- src/regexp.zig | 41 ++++++++++++++++++++++++++++++++++++++- src/sam_edit.zig | 8 +++++++- 5 files changed, 62 insertions(+), 6 deletions(-) (limited to 'src/regexp.zig') diff --git a/.agents/skills/pardes-9p/SKILL.md b/.agents/skills/pardes-9p/SKILL.md index 757ce64a..57ebd6cf 100644 --- a/.agents/skills/pardes-9p/SKILL.md +++ b/.agents/skills/pardes-9p/SKILL.md @@ -182,7 +182,10 @@ is why copying one onto another is all that acme's `addr=dot`, `dot=addr` and `12:5,14:1`), whose regexps are mvzr's searched as sam searches: `^`/`$` match at any line's start and end, `.` and `[^...]` never match a newline, the leftmost match wins (the first alternative there, not the -longest). An expression is evaluated from the current address (the last one +longest). In a pattern with `\n`, `^` works only first (`^def .*\n` finds +every def line) and `$` only just before a `\n`; anywhere else the pattern is +refused, not silently unmatched. `^` inside an alternation (`^def|^ `) holds +only where the search starts: fine in `x` over lines or `g`, not mid-line. An expression is evaluated from the current address (the last one written, or just past the last `data` write): `.` is that address, not the selection, `/re/` searches on from its end and wraps unless `limit` is set, `?re?` or `-/re/` searches back, `#100,#50` fails `addresses out of order`, diff --git a/docs/fs.md b/docs/fs.md index 98a8638d..d60b4e73 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -401,7 +401,15 @@ lazy `*?` included), searched the way sam searches (editors/acme/regx.c): as lines, so `^` and `$` match at the start and end of any line, `.` and a negated class never match a newline, and `$` also matches at the end of a text with no final newline. A pattern that names a newline (`\n`) runs over -the whole text instead, its `.` kept to one line. An expression is +the whole text instead, its `.` kept to one line; there a leading `^` still +matches at every line start, and `$` may stand just before a `\n` (where it +changes nothing). Any other `^` or `$` in such a pattern, `(^|\n)def` or +`a\nb$`, is refused with `in a pattern with \n, ^ can only come first and $ +only just before a \n`, since mvzr would read it as the start or end of the +whole text: never a search that silently finds nothing. Within a line, `^` +inside an alternation (`^def|^ `) matches only where the search starts, so it +works from a line's start (an `x` over lines, a `g` on one) and not from its +middle: an mvzr limit. An expression is evaluated from the current address, the range last written to `addr` (or left by the last `data` write, just past it), as acme evaluates it from `w->addr` (xfid.c:446): `.` is that address, not the selection (`dot` is diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index f57c62bd..1e47183b 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -222,8 +222,8 @@ pub const Addr = struct { a.err = "no previous regular expression"; return null; } - var rx = regexp_.Regex.compile(pat) catch { - a.err = e_regexp; + var rx = regexp_.Regex.compile(pat) catch |err| { + a.err = if (err == error.Anchor) regexp_.Regex.e_anchor else e_regexp; return null; }; const found = if (back) found: { diff --git a/src/regexp.zig b/src/regexp.zig index fe68585d..9c34ae2c 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -41,7 +41,12 @@ pub const Regex = struct { /// patterns take 45-80 ms, and a Debug build is ten times slower. pub const budget: u64 = if (builtin.mode == .Debug) 4_000_000 else 32_000_000; - pub fn compile(pat: []const u8) error{Bad}!Regex { + pub const e_anchor = "in a pattern with \\n, ^ can only come first and $ only just before a \\n"; + + /// `Anchor`: a pattern that names a newline has `^` other than first, + /// or `$` other than just before a `\n`, which mvzr would read as the + /// ends of the whole text and so never match where sam would. + pub fn compile(pat: []const u8) error{ Bad, Anchor }!Regex { if (pat.len == 0) return error.Bad; var buf: [256]u8 = undefined; var len: usize = 0; @@ -68,6 +73,12 @@ pub const Regex = struct { in_class = true; } else if (c == '.' and spans) { piece = "[^\\n]"; + } else if (spans and c == '^' and i != 0) { + return error.Anchor; + } else if (spans and c == '$') { + // `x$\n` is `x\n`; any other `$` would be the text's end. + if (std.mem.startsWith(u8, pat[i + 1 ..], "\\n")) continue; + return error.Anchor; } if (!emit) continue; if (len + piece.len > buf.len) return error.Bad; @@ -91,6 +102,20 @@ pub const Regex = struct { mvzr.steps_left = rx.steps; mvzr.exhausted = false; defer rx.steps = mvzr.steps_left; + // A `^` pattern that spans lines: mvzr's `^` is its haystack's start, + // so each line start from `from` on is tried as one. + // ponytail: a search per line start, each to `hi`; the step budget + // bounds it. + if (rx.spans and rx.bol) { + var s = if (from == 0 or text[from - 1] == '\n') from else (std.mem.indexOfScalarPos(u8, text[0..hi], from, '\n') orelse return null) + 1; + while (s <= last and s <= hi) { + const hit = rx.re.match(text[s..hi]); + if (mvzr.exhausted) return error.TooSlow; + if (hit) |m| if (m.start == 0) return .{ .start = s, .end = s + m.end }; + s = (std.mem.indexOfScalarPos(u8, text[0..hi], s, '\n') orelse return null) + 1; + } + return null; + } var start: usize = if (rx.spans) 0 else if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; var at = from - start; // `^` cannot match in the middle of a line. @@ -136,6 +161,20 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li try std.testing.expectEqual(c.end, m.end); } _ = try Regex.compile("a.*a\\nq"); + // `^` at every line start and `$` before a newline, when a pattern + // spans lines; anywhere else they are refused, never silently wrong. + const defs = "x = 1\ndef a\n\ndef b\n"; + var def = try Regex.compile("^def .*\\n"); + const d = (try def.find(defs, 0, defs.len, defs.len)).?; + try std.testing.expectEqual(@as(usize, 6), d.start); + try std.testing.expectEqual(@as(usize, 12), d.end); + try std.testing.expectEqual(@as(usize, 13), (try def.find(defs, 7, defs.len, defs.len)).?.start); + var blank = try Regex.compile("^\\n"); + try std.testing.expectEqual(@as(usize, 12), (try blank.find(defs, 0, defs.len, defs.len)).?.start); + var dollar = try Regex.compile("1$\\n"); + try std.testing.expectEqual(@as(usize, 4), (try dollar.find(defs, 0, defs.len, defs.len)).?.start); + try std.testing.expectError(error.Anchor, Regex.compile("(^|\\n)def")); + try std.testing.expectError(error.Anchor, Regex.compile("a$\\nb$")); var none = try Regex.compile("zzz"); try std.testing.expect(try none.find(text, 0, text.len, text.len) == null); try std.testing.expectError(error.Bad, Regex.compile("")); diff --git a/src/sam_edit.zig b/src/sam_edit.zig index c08c6cdd..40221628 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -380,7 +380,10 @@ const Exec = struct { } fn compile(ex: *Exec, pat: []const u8, c: u8) Failure!regexp.Regex { - return regexp.Regex.compile(pat) catch fail(ex.why, "bad regexp in {c} command", .{c}); + return regexp.Regex.compile(pat) catch |err| switch (err) { + error.Anchor => fail(ex.why, "{s}", .{regexp.Regex.e_anchor}), + error.Bad => fail(ex.why, "bad regexp in {c} command", .{c}), + }; } fn find(ex: *Exec, rx: *regexp.Regex, from: usize, hi: usize) Failure!?regexp.Regex.Match { @@ -637,6 +640,8 @@ test "sam's classic commands" { try expectEdit("one\ntwo\n", "1 m $", "two\none\n"); try expectEdit("one\ntwo\n", "1 t $", "one\ntwo\none\n"); try expectEdit("one\ntwo\nthree\n", "2 d", "one\nthree\n"); + // ^ at every line start in a pattern that spans lines (dogfood round 5). + try expectEdit("x = 1\ndef a\ndef b\n", ",x/^def .*\\n/a/X\\n/", "x = 1\ndef a\nX\ndef b\nX\n"); // pardes's line:col. try expectEdit("one\ntwo\n", "2:2 i/-/", "one\nt-wo\n"); } @@ -653,6 +658,7 @@ test "an Edit that fails halfway changes nothing, and says why in acme's words" .{ "w /tmp/x", "w is not supported in pardes" }, .{ ",s/(a)/\\1/", "no \\1: mvzr keeps no submatches" }, .{ "1 m 1,2", "move overlaps itself" }, + .{ ",x/(^|\\n)foo/d", "in a pattern with \\n, ^ can only come first and $ only just before a \\n" }, }) |c| { var why: Why = .{}; try std.testing.expectError(error.Edit, run(arena_state.allocator(), "foo a\nfoo\n", .{}, "t", c[0], &why)); -- cgit v1.3 From 197fe7396e7f63e9471bb317f1341228a0690ce6 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Mon, 28 Sep 2026 19:52:50 -0300 Subject: A search pattern with a short \x escape is refused, where mvzr panicked on it Co-Authored-By: Claude Opus 5.5 --- src/regexp.zig | 11 +++++++++++ 1 file changed, 11 insertions(+) (limited to 'src/regexp.zig') diff --git a/src/regexp.zig b/src/regexp.zig index fe68585d..f6d540c4 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -62,6 +62,10 @@ pub const Regex = struct { i += 1; piece = pat[i - 1 .. i + 1]; if (pat[i] == 'n') spans = true; + // mvzr slices two hex digits after `\x` without looking, + // so a short one panics it: refuse it here. + if (pat[i] == 'x' and (i + 2 >= pat.len or + !std.ascii.isHex(pat[i + 1]) or !std.ascii.isHex(pat[i + 2]))) return error.Bad; } else if (in_class) { in_class = c != ']'; } else if (c == '[') { @@ -185,3 +189,10 @@ test "the match is the leftmost, however long the line" { try std.testing.expectEqual(@as(usize, 2), f.start); try std.testing.expectEqual(@as(usize, 7), f.end); } + +test "a \\x without two hex digits is refused, not handed to mvzr to panic on" { + for ([_][]const u8{ "\\x", "a\\x1", "\\x1b[\\x", "[\\x]", "\\xg1" }) |pat| + try std.testing.expectError(error.Bad, Regex.compile(pat)); + var rx = try Regex.compile("\\x41"); + try std.testing.expectEqual(@as(usize, 1), (try rx.find("xA", 0, 1, 2)).?.start); +} -- cgit v1.3 From fad6bf0cc7cc2cb579d9ad03147a7014906f6b92 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 00:36:25 -0300 Subject: A repeat inside a regexp group gives back to what follows the group On aab, /a+ab/ matched but /(a+)ab/ and /(a*)ab/ missed, and so did (.+)_area and ([a-z_]+)_area, in addr, Edit and look alike. mvzr's hasAlt counts a group's own ) as an alternative, so every group with pattern after it takes matchGroup's alternatives branch, which never tries a repeat's shorter matches. No newer mvzr fixes it (trunk is the pinned commit), so the build patches the fetched source as it does the step budget, anchor-checked: when the rest fails after a group, the group's shorter matches are tried, longest first. docs/mvzr-group-backtrack.md is a report for upstream. Co-Authored-By: Claude Opus 5.5 --- build.zig | 35 +++++++++++++++++++++++- docs/mvzr-group-backtrack.md | 64 ++++++++++++++++++++++++++++++++++++++++++++ src/regexp.zig | 23 ++++++++++++++++ 3 files changed, 121 insertions(+), 1 deletion(-) create mode 100644 docs/mvzr-group-backtrack.md (limited to 'src/regexp.zig') diff --git a/build.zig b/build.zig index a78cd442..53cbcda5 100644 --- a/build.zig +++ b/build.zig @@ -454,7 +454,40 @@ pub fn build(b: *std.Build) void { \\pub threadlocal var exhausted: bool = false; \\ ; - const patched = b.fmt("{s}{s}{s}{s}", .{ src[0..at], spend, src[at..], budget }); + const budgeted = b.fmt("{s}{s}{s}{s}", .{ src[0..at], spend, src[at..], budget }); + // A group's greedy match never gave back to what follows the group: + // on `aab`, `(a+)ab` missed where `a+ab` matched. hasAlt counts the + // group's own `)` as an alternative, so every group with pattern + // after it takes matchGroup's alternatives branch, which tries each + // alternative's one (greedy) match and gives up. When the rest fails + // after it, try that alternative's shorter matches, longest first + // (docs/mvzr-group-backtrack.md is the report for upstream). + const group_anchor = + \\ const next_match = matchPattern(next_patt, sets, haystack, m1.i); + \\ if (next_match) |m2| { + \\ // Whole group matches, and we can use m2.j + \\ return m2; + \\ } else { // Try our next pattern, if any. + \\ + ; + const group_backoff = + \\ const next_match = matchPattern(next_patt, sets, haystack, m1.i); + \\ if (next_match) |m2| { + \\ // Whole group matches, and we can use m2.j + \\ return m2; + \\ } else { // Try our next pattern, if any. + \\ // pardes's patch (its build.zig): back the group off. + \\ var e = m1.i; + \\ while (e > i) { + \\ e -= 1; + \\ const shorter = matchAlt(remaining_patt, sets, haystack[0..e], i) orelse continue; + \\ if (shorter.i != e) continue; + \\ if (matchPattern(next_patt, sets, haystack, e)) |whole| return whole; + \\ } + \\ + ; + if (std.mem.count(u8, budgeted, group_anchor) != 1) @panic("mvzr's matchGroup changed: redo its group backtracking in build.zig"); + const patched = std.mem.replaceOwned(u8, b.allocator, budgeted, group_anchor, group_backoff) catch @panic("OOM"); break :mvzr b.createModule(.{ .root_source_file = b.addWriteFiles().add("mvzr.zig", patched), .target = target, .optimize = optimize }); }; diff --git a/docs/mvzr-group-backtrack.md b/docs/mvzr-group-backtrack.md new file mode 100644 index 00000000..d074af62 --- /dev/null +++ b/docs/mvzr-group-backtrack.md @@ -0,0 +1,64 @@ +# mvzr: a repeat inside a group never gives back to what follows the group + +A report for upstream (https://github.com/mnemnion/mvzr), against trunk +3efcfe3 (0.3.9). pardes patches its fetched copy at build time (build.zig) +until it is fixed there. + +## What happens + +```zig +const mvzr = @import("mvzr"); +// a+ab matches aab; the same with the repeat in a group does not +std.debug.assert(mvzr.compile("a+ab").?.match("aab") != null); +std.debug.assert(mvzr.compile("(a+)ab").?.match("aab") == null); // expected a match +``` + +The same for `(a*)ab` on `aab`, `(.+)_area` and `([a-z_]+)_area` on +`the_total_area`. `\w+_area` and `(\w)+_area` match, as does +`([a-z]+)_area` (whose class cannot eat the `_`, so no giving back is +needed). + +## Why + +`hasAlt` returns true when its scan reaches the group's own closing +`.right` at depth 0: + +```zig +.right => { + if (pump == 0) + return true // the group's own `)`, not an alternative + else + pump -= 1; +}, +``` + +so every group with more pattern after it goes down `matchGroup`'s +alternatives branch. That branch takes each alternative's single match (a +greedy repeat's longest) and, when the pattern after the group fails from +there, moves on to the next alternative; it never tries a shorter match of +the same alternative. The alt-free branch has the same shape. + +## A fix + +Where the pattern after the group fails from an alternative's match, try +that alternative's shorter matches, longest first, before the next +alternative. pardes's patch, in `matchGroup`'s loop: + +```zig +} else { // Try our next pattern, if any. + var e = m1.i; + while (e > i) { + e -= 1; + const shorter = matchAlt(remaining_patt, sets, haystack[0..e], i) orelse continue; + if (shorter.i != e) continue; + if (matchPattern(next_patt, sets, haystack, e)) |whole| return whole; + } + remaining_patt = m1.j; +} +``` + +Matching the alternative against `haystack[0..e]` makes `$` and `\b` see a +string end at `e`, so a group ending in either can accept a cut the full +string would not; a continuation-passing matcher (each repeat trying the +rest of the whole pattern) would be the full fix. `hasAlt` returning false +at the group's own `)` is worth fixing too. diff --git a/src/regexp.zig b/src/regexp.zig index 249763c2..97e3dc91 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -235,3 +235,26 @@ test "a \\x without two hex digits is refused, not handed to mvzr to panic on" { var rx = try Regex.compile("\\x41"); try std.testing.expectEqual(@as(usize, 1), (try rx.find("xA", 0, 1, 2)).?.start); } + +test "a repeat inside a group gives back to what follows the group" { + const Case = struct { pat: []const u8, text: []const u8, start: usize, end: usize }; + for ([_]Case{ + .{ .pat = "a+ab", .text = "aab", .start = 0, .end = 3 }, + .{ .pat = "(a+)ab", .text = "aab", .start = 0, .end = 3 }, + .{ .pat = "(a*)ab", .text = "aab", .start = 0, .end = 3 }, + .{ .pat = "(.+)_area", .text = "the_total_area x", .start = 0, .end = 14 }, + .{ .pat = "([a-z_]+)_area", .text = "the_total_area x", .start = 0, .end = 14 }, + .{ .pat = "\\w+_area", .text = "the_total_area x", .start = 0, .end = 14 }, + .{ .pat = "(\\w)+_area", .text = "the_total_area x", .start = 0, .end = 14 }, + .{ .pat = "([a-z]+)_area", .text = "the_total_area x", .start = 4, .end = 14 }, + .{ .pat = "x(ab|a)bc", .text = "xabc", .start = 0, .end = 4 }, + }) |c| { + var rx = try Regex.compile(c.pat); + const m = (try rx.find(c.text, 0, c.text.len, c.text.len)) orelse { + std.debug.print("no match for {s}\n", .{c.pat}); + return error.NoMatch; + }; + try std.testing.expectEqual(c.start, m.start); + try std.testing.expectEqual(c.end, m.end); + } +} -- cgit v1.3 From 0423e315dfeaa5af48d2186a8f0997e1904f3da9 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 01:41:06 -0300 Subject: $ may end any alternative in a pattern, not only the last mvzr refuses $ anywhere but at the very end, so /h$|zz/ and Edit ,x/r$|pi/ were bad patterns while /zz|h$/ worked. Its matcher already ends an alternative at .end, so a build-time patch, anchor-checked like the group one, lets $ stand before | and ). The upstream report is in docs/mvzr-group-backtrack.md. Co-Authored-By: Claude Opus 5.5 --- build.zig | 19 ++++++++++++++++++- docs/mvzr-group-backtrack.md | 22 ++++++++++++++++++++++ src/regexp.zig | 17 +++++++++++++++++ 3 files changed, 57 insertions(+), 1 deletion(-) (limited to 'src/regexp.zig') diff --git a/build.zig b/build.zig index e63b7f4b..fb17afd8 100644 --- a/build.zig +++ b/build.zig @@ -500,7 +500,24 @@ pub fn build(b: *std.Build) void { \\ ; if (std.mem.count(u8, budgeted, group_anchor) != 1) @panic("mvzr's matchGroup changed: redo its group backtracking in build.zig"); - const patched = std.mem.replaceOwned(u8, b.allocator, budgeted, group_anchor, group_backoff) catch @panic("OOM"); + const backed = std.mem.replaceOwned(u8, b.allocator, budgeted, group_anchor, group_backoff) catch @panic("OOM"); + // `$` is refused anywhere but last, so `h$|zz` was a bad pattern + // while `zz|h$` was fine. An alternative or a group ends at `|` and + // `)`, and the matcher's `.end` ends the alternative it is in, so + // `$` may stand before either (docs/mvzr-group-backtrack.md). + const end_anchor = + \\ '$' => { + \\ if (i + 1 < in.len) { + \\ + ; + const end_before_alt = + \\ '$' => { + \\ // pardes's patch (its build.zig): `$` may end an alternative. + \\ if (i + 1 < in.len and in[i + 1] != '|' and in[i + 1] != ')') { + \\ + ; + if (std.mem.count(u8, backed, end_anchor) != 1) @panic("mvzr's parser changed: redo its `$` before `|` in build.zig"); + const patched = std.mem.replaceOwned(u8, b.allocator, backed, end_anchor, end_before_alt) catch @panic("OOM"); break :mvzr b.createModule(.{ .root_source_file = b.addWriteFiles().add("mvzr.zig", patched), .target = target, .optimize = optimize }); }; diff --git a/docs/mvzr-group-backtrack.md b/docs/mvzr-group-backtrack.md index d074af62..867336a3 100644 --- a/docs/mvzr-group-backtrack.md +++ b/docs/mvzr-group-backtrack.md @@ -62,3 +62,25 @@ string end at `e`, so a group ending in either can accept a cut the full string would not; a continuation-passing matcher (each repeat trying the rest of the whole pattern) would be the full fix. `hasAlt` returning false at the group's own `)` is worth fixing too. + +# mvzr: `$` before `|` is a bad pattern + +A second report, same trunk. `compile` refuses `$` anywhere but at the +very end of the pattern, so an anchored alternative must come last: + +```zig +std.debug.assert(mvzr.compile("zz|h$") != null); +std.debug.assert(mvzr.compile("h$|zz") == null); // expected a regex +std.debug.assert(mvzr.compile("(h$|zz)") == null); // likewise +``` + +The parser's check is `if (i + 1 < in.len) bad_string = true`. The +matcher already handles `.end` inside an alternative: `matchAlt` and +`matchGroup` hand `matchPattern` each alternative as its own slice, and +`.end` returns an empty rest, which ends that slice. So `$` can stand +before `|` or `)` as well as last; pardes's patch: + +```zig +'$' => { + if (i + 1 < in.len and in[i + 1] != '|' and in[i + 1] != ')') { +``` diff --git a/src/regexp.zig b/src/regexp.zig index 97e3dc91..64e1d75a 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -258,3 +258,20 @@ test "a repeat inside a group gives back to what follows the group" { try std.testing.expectEqual(c.end, m.end); } } + +test "`$` may end any alternative, not only the last" { + const Case = struct { pat: []const u8, text: []const u8, start: usize, end: usize }; + for ([_]Case{ + .{ .pat = "h$|zz", .text = "xh", .start = 1, .end = 2 }, + .{ .pat = "zz|h$", .text = "xh", .start = 1, .end = 2 }, + .{ .pat = "h$|zz", .text = "hx zz", .start = 3, .end = 5 }, + .{ .pat = "(h$|zz)", .text = "xh", .start = 1, .end = 2 }, + }) |c| { + var rx = try Regex.compile(c.pat); + const m = (try rx.find(c.text, 0, c.text.len, c.text.len)).?; + try std.testing.expectEqual(c.start, m.start); + try std.testing.expectEqual(c.end, m.end); + } + var only = try Regex.compile("h$|zz"); + try std.testing.expectEqual(null, try only.find("hx", 0, 2, 2)); +} -- cgit v1.3 From 2d9f40667197ac775d5215ffd164c800f83cab4c Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 12:35:40 -0300 Subject: A pattern with a newline and an anchor mvzr would misread is refused EINVAL, as the doc says Its words named no malformed input, so 9ns read the refusal as EIO, and Edit's failed EIO directly. It now says bad regular expression: ..., and Edit's failure with those words is EINVAL too. Co-Authored-By: Claude Opus 5.5 --- src/ninep/addr.zig | 10 ++++++++++ src/ninep/ctl.zig | 5 ++++- src/regexp.zig | 2 +- src/sam_edit.zig | 2 +- 4 files changed, 16 insertions(+), 3 deletions(-) (limited to 'src/regexp.zig') diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 09ef2b66..f60c07b1 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -377,6 +377,16 @@ test "an address is evaluated from the current one, as acme's are" { try testing.expectEqualStrings(e_order, th.wr(q, qaddr, "/a/,/b/").reply.ename); } +test "a pattern with a newline and an anchor mvzr would misread is refused EINVAL, by addr and by Edit" { + const p = try th.withFile(testing.allocator, "a\nfoo\n"); + defer p.deinit(); + const r = th.wr(p, Node.of(th.serialOf(p), .addr), "/(^|\\n)foo/"); + try testing.expectEqual(E.INVAL, r.errno()); + try testing.expectEqualStrings(regexp_.Regex.e_anchor, r.reply.ename); + const e = th.wr(p, Node.of(th.serialOf(p), .ctl), "Edit ,x/(^|\\n)foo/d\n"); + try testing.expectEqual(E.INVAL, e.errno()); +} + test "a row's range, L:C-L2:C2 or L:C-C2, is an address through its last character" { const p = try th.withFile(testing.allocator, "one\ntwo three\nfour\n"); defer p.deinit(); diff --git a/src/ninep/ctl.zig b/src/ninep/ctl.zig index 978ce707..b79b1ab8 100644 --- a/src/ninep/ctl.zig +++ b/src/ninep/ctl.zig @@ -250,7 +250,10 @@ pub fn captured(p: *Pardes, req: Req, id: usize, exec: bool, text: []const u8, i /// there (`no such`, `not found`, the words 9ns reads as ENOENT too), else /// EIO. fn failureErrno(failure: []const u8) u16 { - return if (std.mem.indexOf(u8, failure, "no such") != null or std.mem.indexOf(u8, failure, "not found") != null) E.NOENT else E.IO; + if (std.mem.indexOf(u8, failure, "no such") != null or std.mem.indexOf(u8, failure, "not found") != null) return E.NOENT; + // A pattern refused (Edit's too) is malformed input, as 9ns reads it. + if (std.mem.indexOf(u8, failure, "bad regular expression") != null) return E.INVAL; + return E.IO; } pub fn resultsLen(p: *Pardes) u64 { diff --git a/src/regexp.zig b/src/regexp.zig index 64e1d75a..27b3c0ba 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -41,7 +41,7 @@ pub const Regex = struct { /// patterns take 45-80 ms, and a Debug build is ten times slower. pub const budget: u64 = if (builtin.mode == .Debug) 4_000_000 else 32_000_000; - pub const e_anchor = "in a pattern with \\n, ^ can only come first and $ only just before a \\n"; + pub const e_anchor = "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n"; /// `Anchor`: a pattern that names a newline has `^` other than first, /// or `$` other than just before a `\n`, which mvzr would read as the diff --git a/src/sam_edit.zig b/src/sam_edit.zig index cd2998ef..a88728aa 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -735,7 +735,7 @@ test "an Edit that fails halfway changes nothing, and says why in acme's words" .{ "w /tmp/x", "w is not supported in pardes" }, .{ ",s/(a)/\\1/", "no \\1: mvzr keeps no submatches" }, .{ "1 m 1,2", "move overlaps itself" }, - .{ ",x/(^|\\n)foo/d", "in a pattern with \\n, ^ can only come first and $ only just before a \\n" }, + .{ ",x/(^|\\n)foo/d", "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n" }, }) |c| { var why: Why = .{}; try std.testing.expectError(error.Edit, run(arena_state.allocator(), "foo a\nfoo\n", .{}, "t", c[0], &why)); -- cgit v1.3 From a945c035ce9e3d5fda81a05f95f3ea1788a19ece Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 13:43:52 -0300 Subject: A ^ after | anchors its branch: ^def|^ finds lines starting either way mvzr takes ^ only at its pattern's start, so in ^def|^ the second branch never matched mid-text. An alternation whose every top-level branch starts with ^ is now compiled as ^(def| ); a mix of anchored and unanchored branches is refused rather than silently wrong. Co-Authored-By: Claude Opus 5.5 --- .agents/skills/pardes-9p/SKILL.md | 4 +-- docs/fs.md | 11 ++++--- src/regexp.zig | 65 +++++++++++++++++++++++++++++++++++++++ 3 files changed, 73 insertions(+), 7 deletions(-) (limited to 'src/regexp.zig') diff --git a/.agents/skills/pardes-9p/SKILL.md b/.agents/skills/pardes-9p/SKILL.md index a91b4861..16e90dc9 100644 --- a/.agents/skills/pardes-9p/SKILL.md +++ b/.agents/skills/pardes-9p/SKILL.md @@ -289,8 +289,8 @@ through 14:2 inclusive), whose regexps are mvzr's searched as sam searches: newline, the leftmost match wins (the first alternative there, not the longest). In a pattern with `\n`, `^` works only first (`^def .*\n` finds every def line) and `$` only just before a `\n`; anywhere else the pattern is -refused, not silently unmatched. `^` inside an alternation (`^def|^ `) holds -only where the search starts: fine in `x` over lines or `g`, not mid-line. An expression is evaluated from the current address (the last one +refused, not silently unmatched. `^def|^ ` finds lines starting either way +(every branch anchored; a mix like `^def|x` is refused). An expression is evaluated from the current address (the last one written, or just past the last `data` write): `.` is that address, not the selection, `/re/` searches on from its end and wraps unless `limit` is set, `?re?` or `-/re/` searches back, `#100,#50` fails `addresses out of order`, diff --git a/docs/fs.md b/docs/fs.md index 088d1c92..339c14ba 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -703,10 +703,11 @@ matches at every line start, and `$` may stand just before a `\n` (where it changes nothing). Any other `^` or `$` in such a pattern, `(^|\n)def` or `a\nb$`, is refused, EINVAL, with `bad regular expression: in a pattern with \n, ^ can only come first and $ only just before a \n`, since mvzr would read it as the start or end of the -whole text: never a search that silently finds nothing. Within a line, `^` -inside an alternation (`^def|^ `) matches only where the search starts, so it -works from a line's start (an `x` over lines, a `g` on one) and not from its -middle: an mvzr limit. An expression is +whole text: never a search that silently finds nothing. An alternation +whose every branch starts with `^` (`^def|^ `) finds a line that starts +either way (it is taken as `^(def| )`); one that mixes anchored and +unanchored branches (`^def|x`) is refused, `bad regular expression`, since +mvzr keeps `^` only first. An expression is evaluated from the current address, the range last written to `addr` (or left by the last `data` write, just past it), as acme evaluates it from `w->addr` (xfid.c:446): `.` is that address, not the selection (`dot` is @@ -718,7 +719,7 @@ otherwise wraps to the start of the text; `?re?` and `-/re/` find the last match ending before the range, wrapping to the text's last. The match is the leftmost, but of the alternatives at that place mvzr takes the first that matches where sam takes the longest (`/gam|gamma/` finds `gam`); in a search begun in the -middle of a line, `^` inside an alternation can match there; and in a +middle of a line, `^` inside a group can match there; and in a pattern that spans lines, `^`, `$` and `[^...]` keep mvzr's own meaning. mvzr backtracks without bound of its own (`a?` twenty times then twenty `a`s is 2^20 steps from each place it tries), and a search holds the editor, so pardes patches a diff --git a/src/regexp.zig b/src/regexp.zig index 27b3c0ba..ca396ff1 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -48,6 +48,11 @@ pub const Regex = struct { /// ends of the whole text and so never match where sam would. pub fn compile(pat: []const u8) error{ Bad, Anchor }!Regex { if (pat.len == 0) return error.Bad; + // mvzr takes `^` only at its pattern's start, so `^def|^ ` (a `^` + // after a `|`) is written `^(def| )` for it: the same lines. A mix, + // `^a|b`, has no such spelling and is refused rather than wrong. + var joined: [258]u8 = undefined; + if (try anchoredAlternation(pat, &joined)) |whole| return compile(whole); var buf: [256]u8 = undefined; var len: usize = 0; var spans = false; @@ -97,6 +102,56 @@ pub const Regex = struct { }; } + /// `^a|^b` as `^(a|b)` in `out`, when the pattern is an alternation at + /// its top level and every branch starts with `^`; null when it is not + /// one, or no branch does. + fn anchoredAlternation(pat: []const u8, out: *[258]u8) error{Bad}!?[]const u8 { + var bars: [16]usize = undefined; + var n: usize = 0; + var depth: usize = 0; + var in_class = false; + var i: usize = 0; + while (i < pat.len) : (i += 1) { + const c = pat[i]; + if (c == '\\') { + i += 1; + } else if (in_class) { + in_class = c != ']'; + } else if (c == '[') { + in_class = true; + } else if (c == '(') { + depth += 1; + } else if (c == ')') { + depth -|= 1; + } else if (c == '|' and depth == 0) { + if (n == bars.len) return error.Bad; + bars[n] = i; + n += 1; + } + } + if (n == 0) return null; + var anchored: usize = 0; + var from: usize = 0; + for (0..n + 1) |k| { + const to = if (k < n) bars[k] else pat.len; + anchored += @intFromBool(to > from and pat[from] == '^'); + from = to + 1; + } + if (anchored == 0) return null; + if (anchored != n + 1) return error.Bad; + var w = std.Io.Writer.fixed(out); + w.writeAll("^(") catch return error.Bad; + from = 0; + for (0..n + 1) |k| { + const to = if (k < n) bars[k] else pat.len; + if (k > 0) w.writeByte('|') catch return error.Bad; + w.writeAll(pat[from + 1 .. to]) catch return error.Bad; + from = to + 1; + } + w.writeByte(')') catch return error.Bad; + return w.buffered(); + } + pub const Match = struct { start: usize, end: usize }; /// The first match that starts in `from..=last` and ends by `hi`, as @@ -185,6 +240,16 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li try std.testing.expectError(error.Bad, Regex.compile("a\\")); } +test "a ^ after | anchors that branch: ^def|^ finds a line that starts either way, and a mix is refused" { + const text = "x def\n a\ndef b\n"; + var both = try Regex.compile("^def|^ "); + try std.testing.expectEqual(@as(usize, 6), (try both.find(text, 0, text.len, text.len)).?.start); + try std.testing.expectEqual(@as(usize, 9), (try both.find(text, 7, text.len, text.len)).?.start); + try std.testing.expectError(error.Bad, Regex.compile("^def|x")); + var plain = try Regex.compile("a|b"); + try std.testing.expectEqual(@as(usize, 7), (try plain.find(text, 0, text.len, text.len)).?.start); +} + test "a quoted backslash before n is no newline, and a class ends where mvzr ends it" { // `\\n` is a backslash then an n: the pattern stays on one line, so its // `.` is not made [^\n] and does match within the line. -- cgit v1.3 From fa07370ea6cbd4be031bab2c14b24203ae329365 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 15:50:13 -0300 Subject: A pattern up to 512 characters compiles; a longer one is refused naming the limit mvzr's Regex holds 64 operations, so a pattern much past 64 bytes was refused as bad regular expression. pardes now compiles into mvzr's SizedRegex(512, 64), and a pattern that is too long (told apart from a malformed one by compiling it with room) says it is longer than mvzr's 512 operations. Co-Authored-By: Claude Opus 5.5 --- docs/fs.md | 5 ++++- src/ninep/addr.zig | 6 +++++- src/regexp.zig | 34 +++++++++++++++++++++++++++------- src/sam_edit.zig | 1 + 4 files changed, 37 insertions(+), 9 deletions(-) (limited to 'src/regexp.zig') diff --git a/docs/fs.md b/docs/fs.md index 7b1e47bf..9ca5f936 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -713,7 +713,10 @@ whole text: never a search that silently finds nothing. An alternation whose every branch starts with `^` (`^def|^ `) finds a line that starts either way (it is taken as `^(def| )`); one that mixes anchored and unanchored branches (`^def|x`) is refused, `bad regular expression`, since -mvzr keeps `^` only first. An expression is +mvzr keeps `^` only first. A pattern may be up to 512 of mvzr's operations, +about 512 characters (mvzr's own is 64; pardes builds it with more); a +longer one is refused, `bad regular expression: longer than mvzr's 512 +operations`. An expression is evaluated from the current address, the range last written to `addr` (or left by the last `data` write, just past it), as acme evaluates it from `w->addr` (xfid.c:446): `.` is that address, not the selection (`dot` is diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index f60c07b1..c052a665 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -299,7 +299,11 @@ pub const Addr = struct { return null; } var rx = regexp_.Regex.compile(pat) catch |err| { - a.err = if (err == error.Anchor) regexp_.Regex.e_anchor else e_regexp; + a.err = switch (err) { + error.Anchor => regexp_.Regex.e_anchor, + error.TooLong => regexp_.Regex.e_long, + error.Bad => e_regexp, + }; return null; }; // sam's nextmatch (editors/sam/address.c:97-119): an empty match diff --git a/src/regexp.zig b/src/regexp.zig index ca396ff1..dd2e7157 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -27,8 +27,14 @@ const mvzr = @import("mvzr"); /// enough line (`\s*(\w+)\s*=` over 20 KB of letters) runs out of budget. A /// regex engine of sam's own would lift these; the user chose not to have /// one. +/// mvzr's own `Regex` holds 64 operations, some 64 pattern characters; a +/// search pattern is often longer. Past these a pattern is refused naming +/// the limit (`e_long`). +pub const max_ops = 512; +const Compiled = mvzr.SizedRegex(max_ops, 64); + pub const Regex = struct { - re: mvzr.Regex, + re: Compiled, /// The pattern names a newline: it runs over the whole text. spans: bool, /// The pattern starts with `^`: a search begun mid-line skips the line. @@ -41,19 +47,21 @@ pub const Regex = struct { /// patterns take 45-80 ms, and a Debug build is ten times slower. pub const budget: u64 = if (builtin.mode == .Debug) 4_000_000 else 32_000_000; + pub const e_long = std.fmt.comptimePrint("bad regular expression: longer than mvzr's {d} operations (about {d} pattern characters)", .{ max_ops, max_ops }); pub const e_anchor = "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n"; /// `Anchor`: a pattern that names a newline has `^` other than first, /// or `$` other than just before a `\n`, which mvzr would read as the /// ends of the whole text and so never match where sam would. - pub fn compile(pat: []const u8) error{ Bad, Anchor }!Regex { + pub fn compile(pat: []const u8) error{ Bad, Anchor, TooLong }!Regex { if (pat.len == 0) return error.Bad; // mvzr takes `^` only at its pattern's start, so `^def|^ ` (a `^` // after a `|`) is written `^(def| )` for it: the same lines. A mix, // `^a|b`, has no such spelling and is refused rather than wrong. - var joined: [258]u8 = undefined; + var joined: [5 * max_ops + 2]u8 = undefined; if (try anchoredAlternation(pat, &joined)) |whole| return compile(whole); - var buf: [256]u8 = undefined; + // `.` may become `[^\n]`: five bytes for one. + var buf: [5 * max_ops]u8 = undefined; var len: usize = 0; var spans = false; // Twice over the pattern: the first pass learns whether it names a @@ -90,13 +98,17 @@ pub const Regex = struct { return error.Anchor; } if (!emit) continue; - if (len + piece.len > buf.len) return error.Bad; + if (len + piece.len > buf.len) return error.TooLong; @memcpy(buf[len..][0..piece.len], piece); len += piece.len; } } return .{ - .re = mvzr.compile(buf[0..len]) orelse return error.Bad, + .re = Compiled.compile(buf[0..len]) orelse { + // Too long, or malformed: told apart by trying it with room. + if (mvzr.SizedRegex(4 * max_ops, 256).compile(buf[0..len]) != null) return error.TooLong; + return error.Bad; + }, .spans = spans, .bol = pat[0] == '^', }; @@ -105,7 +117,7 @@ pub const Regex = struct { /// `^a|^b` as `^(a|b)` in `out`, when the pattern is an alternation at /// its top level and every branch starts with `^`; null when it is not /// one, or no branch does. - fn anchoredAlternation(pat: []const u8, out: *[258]u8) error{Bad}!?[]const u8 { + fn anchoredAlternation(pat: []const u8, out: *[5 * max_ops + 2]u8) error{Bad}!?[]const u8 { var bars: [16]usize = undefined; var n: usize = 0; var depth: usize = 0; @@ -240,6 +252,14 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li try std.testing.expectError(error.Bad, Regex.compile("a\\")); } +test "a pattern past 64 characters compiles, and one past the limit says so" { + var long = try Regex.compile("a" ** 200); + const text = "x" ++ "a" ** 200 ++ "\n"; + try std.testing.expectEqual(@as(usize, 1), (try long.find(text, 0, text.len, text.len)).?.start); + try std.testing.expectError(error.TooLong, Regex.compile("a" ** (max_ops + 8))); + try std.testing.expectError(error.Bad, Regex.compile("a[b")); +} + test "a ^ after | anchors that branch: ^def|^ finds a line that starts either way, and a mix is refused" { const text = "x def\n a\ndef b\n"; var both = try Regex.compile("^def|^ "); diff --git a/src/sam_edit.zig b/src/sam_edit.zig index a88728aa..113ad14f 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -445,6 +445,7 @@ const Exec = struct { fn compile(ex: *Exec, pat: []const u8, c: u8) Failure!regexp.Regex { return regexp.Regex.compile(pat) catch |err| switch (err) { error.Anchor => fail(ex.why, "{s}", .{regexp.Regex.e_anchor}), + error.TooLong => fail(ex.why, "{s}", .{regexp.Regex.e_long}), error.Bad => fail(ex.why, "bad regexp in {c} command", .{c}), }; } -- cgit v1.3 From a76971c9a4cfb84258203f01c7ba5f4927790201 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 17:11:13 -0300 Subject: A search whose pattern opens with a literal jumps to where that literal is, not through every line Finding /line 049999/ in 1171 KB took 175.5 ms (Debug), each line run through mvzr's matcher in turn. The literal prefix (none when the pattern has a | anywhere) is found with indexOfPos first and matching starts on its line: 4.44 ms. Co-Authored-By: Claude Opus 5.5 --- src/regexp.zig | 59 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 59 insertions(+) (limited to 'src/regexp.zig') diff --git a/src/regexp.zig b/src/regexp.zig index dd2e7157..695f2b87 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -35,6 +35,11 @@ const Compiled = mvzr.SizedRegex(max_ops, 64); pub const Regex = struct { re: Compiled, + /// The literal every match starts with, when the pattern opens with one + /// (`line 049`, `foo` in `foo.*bar`): a search goes straight to where it + /// occurs rather than trying mvzr at every line. + lit: [32]u8 = undefined, + lit_len: u8 = 0, /// The pattern names a newline: it runs over the whole text. spans: bool, /// The pattern starts with `^`: a search begun mid-line skips the line. @@ -53,6 +58,22 @@ pub const Regex = struct { /// `Anchor`: a pattern that names a newline has `^` other than first, /// or `$` other than just before a `\n`, which mvzr would read as the /// ends of the whole text and so never match where sam would. + /// The plain characters a pattern opens with, each of which every match + /// must start with: up to the first metacharacter, less the last one + /// when a quantifier makes it optional or repeated. + fn literalPrefix(pat: []const u8, out: *[32]u8) u8 { + // ponytail: any `|` and there is none (a match may start with + // another branch); a top-level-only check would keep `a(b|c)`'s. + if (std.mem.indexOfScalar(u8, pat, '|') != null) return 0; + var n: usize = 0; + while (n < pat.len and n < out.len) : (n += 1) { + if (std.mem.indexOfScalar(u8, "\\^$.[]()|*+?{}", pat[n]) != null) break; + out[n] = pat[n]; + } + if (n < pat.len and std.mem.indexOfScalar(u8, "*?{", pat[n]) != null) n -|= 1; + return @intCast(n); + } + pub fn compile(pat: []const u8) error{ Bad, Anchor, TooLong }!Regex { if (pat.len == 0) return error.Bad; // mvzr takes `^` only at its pattern's start, so `^def|^ ` (a `^` @@ -103,7 +124,11 @@ pub const Regex = struct { len += piece.len; } } + var lit: [32]u8 = undefined; + const lit_len = if (spans) 0 else literalPrefix(pat, &lit); return .{ + .lit = lit, + .lit_len = lit_len, .re = Compiled.compile(buf[0..len]) orelse { // Too long, or malformed: told apart by trying it with room. if (mvzr.SizedRegex(4 * max_ops, 256).compile(buf[0..len]) != null) return error.TooLong; @@ -195,6 +220,24 @@ pub const Regex = struct { at = 0; } while (start <= hi and start <= last) { + // A pattern that opens with a literal matches only where it + // occurs: on to the line where it next does. + if (rx.lit_len > 0 and !rx.spans) { + const q = std.mem.indexOfPos(u8, text[0..hi], start + at, rx.lit[0..rx.lit_len]) orelse return null; + if (q > last) return null; + if (q > start + at) { + const line_start = if (std.mem.lastIndexOfScalar(u8, text[0..q], '\n')) |nl| nl + 1 else 0; + if (line_start > start) { + start = line_start; + at = q - line_start; + } else at = q - start; + if (rx.bol and at > 0) { + start = (std.mem.indexOfScalarPos(u8, text[0..hi], q, '\n') orelse return null) + 1; + at = 0; + continue; + } + } + } const end = if (rx.spans) hi else std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; const line = text[start..end]; // matchPos finds nothing at a haystack's very end, where `$` or @@ -252,6 +295,22 @@ test "lines are haystacks: ^ and $ at each line, . never a newline, \\n spans li try std.testing.expectError(error.Bad, Regex.compile("a\\")); } +test "a pattern opening with a literal finds what a search from each line finds" { + const text = "alpha beta\nbeta gamma\ngamma alpha\nfoo line 049999 x\n"; + for ([_][]const u8{ "beta", "gam+a", "line 049999", "alph?a", "a.*a", "^gamma", "o+", "h$|zz", "beta|x" }) |pat| { + var fast = try Regex.compile(pat); + var slow = try Regex.compile(pat); + slow.lit_len = 0; + var from: usize = 0; + while (from < text.len) : (from += 1) { + const a = try fast.find(text, from, text.len, text.len); + const b = try slow.find(text, from, text.len, text.len); + try std.testing.expectEqual(b == null, a == null); + if (a) |m| try std.testing.expectEqual(b.?.start, m.start); + } + } +} + test "a pattern past 64 characters compiles, and one past the limit says so" { var long = try Regex.compile("a" ** 200); const text = "x" ++ "a" ** 200 ++ "\n"; -- cgit v1.3 From e3e605bd33fc5be56cb7ec842449718bc5337a75 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Tue, 29 Sep 2026 17:46:11 -0300 Subject: A class with non-ASCII runes matches them: [éa-z] is taken as (é|[a-z]), a range up to 256 runes spelled out MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit mvzr's classes hold bytes, and it refused a multibyte member as a bad expression. Rewriting the class before compiling, as ^a|^b already is, needs no mvzr patch; runes that differ only in their last byte go as one alternative (\xc3[\xa0-\xbf] for [à-ÿ]) so a 256-rune range fits the 512-operation limit, which counts the pattern as rewritten. A wider range and [^é] are refused, naming why. Co-Authored-By: Claude Opus 5.5 --- .agents/skills/pardes-9p/SKILL.md | 4 +- docs/fs.md | 7 +- src/ninep/addr.zig | 2 + src/regexp.zig | 159 +++++++++++++++++++++++++++++++++++++- src/sam_edit.zig | 2 + 5 files changed, 171 insertions(+), 3 deletions(-) (limited to 'src/regexp.zig') diff --git a/.agents/skills/pardes-9p/SKILL.md b/.agents/skills/pardes-9p/SKILL.md index 94ad0de8..56123b89 100644 --- a/.agents/skills/pardes-9p/SKILL.md +++ b/.agents/skills/pardes-9p/SKILL.md @@ -293,7 +293,9 @@ newline, the leftmost match wins (the first alternative there, not the longest). In a pattern with `\n`, `^` works only first (`^def .*\n` finds every def line) and `$` only just before a `\n`; anywhere else the pattern is refused, not silently unmatched. `^def|^ ` finds lines starting either way -(every branch anchored; a mix like `^def|x` is refused). An expression is evaluated from the current address (the last one +(every branch anchored; a mix like `^def|x` is refused). `[éa-z]` and +`[à-ÿ]` match those runes (a range up to 256 runes); `[^é]` and a wider +range are refused. An expression is evaluated from the current address (the last one written, or just past the last `data` write): `.` is that address, not the selection, `/re/` searches on from its end and wraps unless `limit` is set, `?re?` or `-/re/` searches back, `#100,#50` fails `addresses out of order`, diff --git a/docs/fs.md b/docs/fs.md index 8464768b..af64fded 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -733,7 +733,12 @@ unanchored branches (`^def|x`) is refused, `bad regular expression`, since mvzr keeps `^` only first. A pattern may be up to 512 of mvzr's operations, about 512 characters (mvzr's own is 64; pardes builds it with more); a longer one is refused, `bad regular expression: longer than mvzr's 512 -operations`. An expression is +operations`. A class may hold non-ASCII runes (`[éa-z]`, `[à-ÿ]`): it is +taken as an alternation of them (`(é|[a-z])`), a range of up to 256 runes +spelled out, and the 512 counts the pattern as rewritten. A wider range +(`a range of runes wider than 256 in [...] is not supported`) and a negated +class with a non-ASCII rune (`[^é]`) are refused, since mvzr's classes hold +bytes. An expression is evaluated from the current address, the range last written to `addr` (or left by the last `data` write, just past it), as acme evaluates it from `w->addr` (xfid.c:446): `.` is that address, not the selection (`dot` is diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 005ad593..33a3f646 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -302,6 +302,8 @@ pub const Addr = struct { a.err = switch (err) { error.Anchor => regexp_.Regex.e_anchor, error.TooLong => regexp_.Regex.e_long, + error.WideRange => regexp_.Regex.e_wide, + error.NegatedRunes => regexp_.Regex.e_negated, error.Bad => e_regexp, }; return null; diff --git a/src/regexp.zig b/src/regexp.zig index 695f2b87..ddb5eb28 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -54,6 +54,11 @@ pub const Regex = struct { pub const e_long = std.fmt.comptimePrint("bad regular expression: longer than mvzr's {d} operations (about {d} pattern characters)", .{ max_ops, max_ops }); pub const e_anchor = "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n"; + pub const e_wide = std.fmt.comptimePrint("bad regular expression: a range of runes wider than {d} in [...] is not supported", .{max_range}); + pub const e_negated = "bad regular expression: a [^...] with non-ASCII runes is not supported (mvzr's classes hold bytes)"; + + /// The most runes a non-ASCII range in a class is spelled out as. + pub const max_range = 256; /// `Anchor`: a pattern that names a newline has `^` other than first, /// or `$` other than just before a `\n`, which mvzr would read as the @@ -74,8 +79,14 @@ pub const Regex = struct { return @intCast(n); } - pub fn compile(pat: []const u8) error{ Bad, Anchor, TooLong }!Regex { + pub const Error = error{ Bad, Anchor, TooLong, WideRange, NegatedRunes }; + + pub fn compile(pat: []const u8) Error!Regex { if (pat.len == 0) return error.Bad; + // mvzr's classes hold bytes: `[éa-z]` is written `(é|[a-z])` for it. + // The limits apply to the pattern as rewritten. + var runes: [5 * max_ops + 2]u8 = undefined; + if (try runeClasses(pat, &runes)) |whole| return compile(whole); // mvzr takes `^` only at its pattern's start, so `^def|^ ` (a `^` // after a `|`) is written `^(def| )` for it: the same lines. A mix, // `^a|b`, has no such spelling and is refused rather than wrong. @@ -139,6 +150,126 @@ pub const Regex = struct { }; } + /// `pat` with each class that holds a non-ASCII rune written as an + /// alternation of its runes and a class of the rest (`[éa-z]` is + /// `(é|[a-z])`), a range spelled out rune by rune; null when no class + /// holds one. + fn runeClasses(pat: []const u8, out: *[5 * max_ops + 2]u8) Error!?[]const u8 { + var w = std.Io.Writer.fixed(out); + var any = false; + var i: usize = 0; + while (i < pat.len) : (i += 1) { + const c = pat[i]; + if (c == '\\') { + w.writeAll(pat[i..@min(i + 2, pat.len)]) catch return error.TooLong; + i += 1; + continue; + } + if (c != '[') { + w.writeByte(c) catch return error.TooLong; + continue; + } + // The class's end, as mvzr finds it: the first unescaped `]`. + var end = i + 1; + while (end < pat.len and pat[end] != ']') : (end += 1) { + if (pat[end] == '\\') end += 1; + } + if (end >= pat.len) return error.Bad; + const body = pat[i + 1 .. end]; + if (for (body) |b| { + if (b >= 0x80) break false; + } else true) { + w.writeAll(pat[i .. end + 1]) catch return error.TooLong; + i = end; + continue; + } + if (body[0] == '^') return error.NegatedRunes; + any = true; + try runeClass(body, &w); + i = end; + } + return if (any) w.buffered() else null; + } + + fn runeClass(body: []const u8, w: *std.Io.Writer) Error!void { + var ascii: [5 * max_ops]u8 = undefined; + var n: usize = 0; + w.writeByte('(') catch return error.TooLong; + var alts: usize = 0; + var j: usize = 0; + while (j < body.len) { + if (body[j] == '\\') { + const len: usize = if (j + 1 < body.len and body[j + 1] == 'x') 4 else 2; + if (j + len > body.len) return error.Bad; + // An escape at one end of a range whose other end is a rune. + if (j + len + 1 < body.len and body[j + len] == '-' and body[j + len + 1] >= 0x80) return error.Bad; + if (n + len > ascii.len) return error.TooLong; + @memcpy(ascii[n..][0..len], body[j..][0..len]); + n += len; + j += len; + continue; + } + const lo, const lo_len = try rune(body[j..]); + j += lo_len; + var hi = lo; + if (j + 1 < body.len and body[j] == '-') { + if (body[j + 1] == '\\' and lo >= 0x80) return error.Bad; + if (body[j + 1] != '\\') { + hi, const hi_len = try rune(body[j + 1 ..]); + j += 1 + hi_len; + } + } + if (hi < lo) return error.Bad; + if (hi - lo + 1 > max_range) return error.WideRange; + // Runes that differ only in their last byte go as one + // alternative, `\xc3[\xa0-\xbf]` for `[à-ÿ]`: 256 runes one a + // time would pass `max_ops`. + var run: [4]u8 = undefined; + var run_len: usize = 0; + var run_hi: u8 = 0; + var cp = lo; + while (cp <= hi + 1) : (cp += 1) { + var enc: [4]u8 = undefined; + const len = if (cp > hi) 0 else std.unicode.utf8Encode(cp, &enc) catch continue; // a surrogate + if (run_len > 0 and (len != run_len or enc[len - 1] != run_hi + 1 or + !std.mem.eql(u8, enc[0 .. len - 1], run[0 .. len - 1]))) + { + if (alts > 0) w.writeByte('|') catch return error.TooLong; + w.writeAll(run[0 .. run_len - 1]) catch return error.TooLong; + if (run_hi == run[run_len - 1]) + w.writeByte(run_hi) catch return error.TooLong + else + w.print("[\\x{x:0>2}-\\x{x:0>2}]", .{ run[run_len - 1], run_hi }) catch return error.TooLong; + alts += 1; + run_len = 0; + } + if (cp > hi) break; + if (cp < 0x80) { + // `\xHH` rather than the byte: `]`, `^`, `-` and `\` + // would otherwise mean something in the class. + if (n + 4 > ascii.len) return error.TooLong; + _ = std.fmt.bufPrint(ascii[n..][0..4], "\\x{x:0>2}", .{cp}) catch unreachable; + n += 4; + continue; + } + if (run_len == 0) { + run = enc; + run_len = len; + } + run_hi = enc[len - 1]; + } + } + if (n > 0) w.print("|[{s}]", .{ascii[0..n]}) catch return error.TooLong; + w.writeByte(')') catch return error.TooLong; + } + + /// The rune `s` opens with and its length in bytes. + fn rune(s: []const u8) error{Bad}!struct { u21, usize } { + const len = std.unicode.utf8ByteSequenceLength(s[0]) catch return error.Bad; + if (len > s.len) return error.Bad; + return .{ std.unicode.utf8Decode(s[0..len]) catch return error.Bad, len }; + } + /// `^a|^b` as `^(a|b)` in `out`, when the pattern is an alternation at /// its top level and every branch starts with `^`; null when it is not /// one, or no branch does. @@ -311,6 +442,32 @@ test "a pattern opening with a literal finds what a search from each line finds" } } +test "a class with non-ASCII runes matches those runes: [éa-z], a range of them, and refuses a wide range or [^é]" { + const text = "1 é 2 b 3 ü 4 ñ\n"; + var mixed = try Regex.compile("[éa-z]+"); + try std.testing.expectEqual(@as(usize, 2), (try mixed.find(text, 0, text.len, text.len)).?.start); + const m = (try mixed.find(text, 5, text.len, text.len)).?; + try std.testing.expectEqualStrings("b", text[m.start..m.end]); + var range = try Regex.compile("3 [à-ÿ]"); + const r = (try range.find(text, 0, text.len, text.len)).?; + try std.testing.expectEqualStrings("3 ü", text[r.start..r.end]); + // An ASCII member that means something in a class stays a member. + var odd = try Regex.compile("[ñ\\]^-]"); + try std.testing.expectEqual(@as(usize, 16), (try odd.find(text, 0, text.len, text.len)).?.start); + var none = try Regex.compile("[ö]"); + try std.testing.expect(try none.find(text, 0, text.len, text.len) == null); + try std.testing.expectError(error.WideRange, Regex.compile("[ā-ӿ]")); + try std.testing.expectError(error.NegatedRunes, Regex.compile("[^é]")); + try std.testing.expectError(error.Bad, Regex.compile("[é")); + try std.testing.expectError(error.Bad, Regex.compile("[ÿ-à]")); + // A full 256-rune range fits the limit as rewritten. + var wide = try Regex.compile("x[Ā-ǿ]"); + try std.testing.expectEqual(@as(usize, 2), (try wide.find("ǿxǿ", 0, 5, 5)).?.start); + var kana = try Regex.compile("[ぁ-ゟ]"); // 3 bytes, across a last-byte wrap + try std.testing.expectEqual(@as(usize, 1), (try kana.find("aゞ", 0, 4, 4)).?.start); + try std.testing.expectError(error.WideRange, Regex.compile("[Ā-Ȁ]")); +} + test "a pattern past 64 characters compiles, and one past the limit says so" { var long = try Regex.compile("a" ** 200); const text = "x" ++ "a" ** 200 ++ "\n"; diff --git a/src/sam_edit.zig b/src/sam_edit.zig index ea1105bd..0f317a83 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -461,6 +461,8 @@ const Exec = struct { return regexp.Regex.compile(pat) catch |err| switch (err) { error.Anchor => fail(ex.why, "{s}", .{regexp.Regex.e_anchor}), error.TooLong => fail(ex.why, "{s}", .{regexp.Regex.e_long}), + error.WideRange => fail(ex.why, "{s}", .{regexp.Regex.e_wide}), + error.NegatedRunes => fail(ex.why, "{s}", .{regexp.Regex.e_negated}), error.Bad => fail(ex.why, "bad regexp in {c} command", .{c}), }; } -- cgit v1.3 From 7c329baddb0d45ce7937cf29941dde861f9ea27f Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Wed, 30 Sep 2026 11:50:52 -0300 Subject: A pattern anchoring some alternatives with ^ and not others is refused with why, not a bare bad regular expression MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit `^def|x` and `foo$|^zèbre` were refused with a bare "bad regular expression", the same words as a syntax error, so a script could not tell what to change. They now say "an alternation anchors every branch with ^ or none (^a|^b, not ^a|b)", through addr and Edit alike. fs.md states the rule and that a `$` does not count toward it (`foo$|bar` is fine). Co-Authored-By: Claude Opus 5.5 --- docs/fs.md | 5 +++-- src/ninep/addr.zig | 13 +++++++++++++ src/regexp.zig | 10 ++++++---- src/sam_edit.zig | 1 + 4 files changed, 23 insertions(+), 6 deletions(-) (limited to 'src/regexp.zig') diff --git a/docs/fs.md b/docs/fs.md index 2a618173..58ba0539 100644 --- a/docs/fs.md +++ b/docs/fs.md @@ -459,8 +459,9 @@ The same code (`src/regexp.zig`) serves addresses, Edit, and normal mode's submatches, so Edit's `s` has no `\1`-`\9`. - A pattern holding `\n` runs over the whole text: there `^` may only come first and `$` only just before a `\n`, else it is refused. -- An alternation must anchor every branch or none (`^def|^ ` works, - `^def|x` is refused). +- An alternation must anchor every branch with `^` or none (`^def|^ ` + works, `^def|x` is refused: `an alternation anchors every branch with ^ + or none`). A `$` does not count: `foo$|bar` is fine. - A class may hold non-ASCII runes (`[éa-z]`, a range up to 256 runes); a wider range or a negated class with one (`[^é]`) is refused. - At most 512 operations (about 512 characters, counted after that diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 33a3f646..9674a323 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -304,6 +304,7 @@ pub const Addr = struct { error.TooLong => regexp_.Regex.e_long, error.WideRange => regexp_.Regex.e_wide, error.NegatedRunes => regexp_.Regex.e_negated, + error.Mixed => regexp_.Regex.e_mixed, error.Bad => e_regexp, }; return null; @@ -383,6 +384,18 @@ test "an address is evaluated from the current one, as acme's are" { try testing.expectEqualStrings(e_order, th.wr(q, qaddr, "/a/,/b/").reply.ename); } +test "an alternation anchoring some branches and not others is refused with why, a $ not counting" { + const p = try th.withFile(testing.allocator, "def\nfoo\n"); + defer p.deinit(); + const node = Node.of(th.serialOf(p), .addr); + for ([_][]const u8{ "/^def|x/", "/foo$|^zèbre/" }) |pat| { + const r = th.wr(p, node, pat); + try testing.expectEqual(E.INVAL, r.errno()); + try testing.expectEqualStrings(regexp_.Regex.e_mixed, r.reply.ename); + } + try testing.expectEqual(Status.ok, th.wr(p, node, "/foo$|def/").reply.status); +} + test "a pattern with a newline and an anchor mvzr would misread is refused EINVAL, by addr and by Edit" { const p = try th.withFile(testing.allocator, "a\nfoo\n"); defer p.deinit(); diff --git a/src/regexp.zig b/src/regexp.zig index ddb5eb28..575e6861 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -56,6 +56,7 @@ pub const Regex = struct { pub const e_anchor = "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n"; pub const e_wide = std.fmt.comptimePrint("bad regular expression: a range of runes wider than {d} in [...] is not supported", .{max_range}); pub const e_negated = "bad regular expression: a [^...] with non-ASCII runes is not supported (mvzr's classes hold bytes)"; + pub const e_mixed = "bad regular expression: an alternation anchors every branch with ^ or none (^a|^b, not ^a|b)"; /// The most runes a non-ASCII range in a class is spelled out as. pub const max_range = 256; @@ -79,7 +80,7 @@ pub const Regex = struct { return @intCast(n); } - pub const Error = error{ Bad, Anchor, TooLong, WideRange, NegatedRunes }; + pub const Error = error{ Bad, Anchor, TooLong, WideRange, NegatedRunes, Mixed }; pub fn compile(pat: []const u8) Error!Regex { if (pat.len == 0) return error.Bad; @@ -273,7 +274,7 @@ pub const Regex = struct { /// `^a|^b` as `^(a|b)` in `out`, when the pattern is an alternation at /// its top level and every branch starts with `^`; null when it is not /// one, or no branch does. - fn anchoredAlternation(pat: []const u8, out: *[5 * max_ops + 2]u8) error{Bad}!?[]const u8 { + fn anchoredAlternation(pat: []const u8, out: *[5 * max_ops + 2]u8) error{ Bad, Mixed }!?[]const u8 { var bars: [16]usize = undefined; var n: usize = 0; var depth: usize = 0; @@ -306,7 +307,8 @@ pub const Regex = struct { from = to + 1; } if (anchored == 0) return null; - if (anchored != n + 1) return error.Bad; + // Only `^` counts: `$` is mvzr's own at a branch's end (`a$|b`). + if (anchored != n + 1) return error.Mixed; var w = std.Io.Writer.fixed(out); w.writeAll("^(") catch return error.Bad; from = 0; @@ -481,7 +483,7 @@ test "a ^ after | anchors that branch: ^def|^ finds a line that starts either w var both = try Regex.compile("^def|^ "); try std.testing.expectEqual(@as(usize, 6), (try both.find(text, 0, text.len, text.len)).?.start); try std.testing.expectEqual(@as(usize, 9), (try both.find(text, 7, text.len, text.len)).?.start); - try std.testing.expectError(error.Bad, Regex.compile("^def|x")); + try std.testing.expectError(error.Mixed, Regex.compile("^def|x")); var plain = try Regex.compile("a|b"); try std.testing.expectEqual(@as(usize, 7), (try plain.find(text, 0, text.len, text.len)).?.start); } diff --git a/src/sam_edit.zig b/src/sam_edit.zig index 0f317a83..94cf6fe0 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -463,6 +463,7 @@ const Exec = struct { error.TooLong => fail(ex.why, "{s}", .{regexp.Regex.e_long}), error.WideRange => fail(ex.why, "{s}", .{regexp.Regex.e_wide}), error.NegatedRunes => fail(ex.why, "{s}", .{regexp.Regex.e_negated}), + error.Mixed => fail(ex.why, "{s}", .{regexp.Regex.e_mixed}), error.Bad => fail(ex.why, "bad regexp in {c} command", .{c}), }; } -- cgit v1.3 From 4020d55aca63f2ba0b3d9bf580f36b0f4ee66f43 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Wed, 30 Sep 2026 12:56:16 -0300 Subject: A backward search along one long line takes time in the line's length, not its square: ?a? over 117 KB goes from 6.4 s to 0.03 s The 9P fuzzer (seed 124) hung a session for over 5 s on `?a?` in a 117 KB file of one line. A backward search is every forward match up to the limit, so Regex.find ran once per match. Each call looked back from its start to the line's start, and forward to the line's end, over the whole line each time. That is quadratic, and the step budget, which counts only mvzr's steps, never saw it. find now keeps where it last learned a line starts and ends in the same text, and looks back only to there. The literal prefix's look back stops at the known part of the line. Co-Authored-By: Claude Opus 5.5 --- src/ninep/addr.zig | 15 +++++++++++++++ src/regexp.zig | 38 +++++++++++++++++++++++++++++++++++--- 2 files changed, 50 insertions(+), 3 deletions(-) (limited to 'src/regexp.zig') diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 9674a323..df2cfaf0 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -384,6 +384,21 @@ test "an address is evaluated from the current one, as acme's are" { try testing.expectEqualStrings(e_order, th.wr(q, qaddr, "/a/,/b/").reply.ename); } +test "a backward search along one long line takes time in its length, not its square" { + const gpa = testing.allocator; + const line = try gpa.alloc(u8, 7 * 16800); + defer gpa.free(line); + for (0..16800) |i| @memcpy(line[i * 7 ..][0..7], "answer "); + const p = try th.withFile(gpa, line); + defer p.deinit(); + const t0 = std.Io.Timestamp.now(testing.io, .awake); + try testing.expectEqual(Status.ok, th.wr(p, Node.of(th.serialOf(p), .addr), "?a?").reply.status); + // Quadratic, it took seconds (6.4 s through 9P in Debug); linear, a + // few milliseconds. + const took = t0.durationTo(std.Io.Timestamp.now(testing.io, .awake)).toNanoseconds(); + try testing.expect(took < std.time.ns_per_s); +} + test "an alternation anchoring some branches and not others is refused with why, a $ not counting" { const p = try th.withFile(testing.allocator, "def\nfoo\n"); defer p.deinit(); diff --git a/src/regexp.zig b/src/regexp.zig index 575e6861..86132f60 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -47,6 +47,14 @@ pub const Regex = struct { /// mvzr steps left to every search with this compiled pattern together, /// so that addr's backward scan and s/S's many calls share one bound. steps: u64 = budget, + /// Where `find` last learned a line's start in a text (by its address + /// and length): the line holding `at` starts at `start`. A backward + /// search asks again at each match along a line, and looking back to + /// the line's start from the top each time was quadratic on a long one. + line_hint: struct { ptr: usize = 0, len: usize = 0, at: usize = 0, start: usize = 0 } = .{}, + /// ...and where a line learned to end: the line from `start` ends at + /// `end` (its newline, or `hi`), asked again at each match along it. + end_hint: struct { ptr: usize = 0, len: usize = 0, start: usize = 0, hi: usize = 0, end: usize = 0 } = .{}, /// Timed on this machine's ReleaseSafe build: 10M steps of the worst /// patterns take 45-80 ms, and a Debug build is ten times slower. @@ -324,6 +332,28 @@ pub const Regex = struct { pub const Match = struct { start: usize, end: usize }; + /// Where the line from `start` ends (its newline, else `hi`), `at` on + /// it: known when asked before of this line of this text. + fn lineEnd(rx: *Regex, text: []const u8, start: usize, at: usize, hi: usize) usize { + const h = rx.end_hint; + if (h.ptr == @intFromPtr(text.ptr) and h.len == text.len and h.start == start and h.hi == hi and at <= h.end) return h.end; + // No newline between the line's start and `at`: the look starts there. + const end = std.mem.indexOfScalarPos(u8, text[0..hi], at, '\n') orelse hi; + rx.end_hint = .{ .ptr = @intFromPtr(text.ptr), .len = text.len, .start = start, .hi = hi, .end = end }; + return end; + } + + /// Where the line holding `from` starts, looking back only to where + /// the last call learned one in this text. + fn lineStart(rx: *Regex, text: []const u8, from: usize) usize { + const h = rx.line_hint; + const known = h.ptr == @intFromPtr(text.ptr) and h.len == text.len and h.at <= from; + const lo = if (known) h.at else 0; + const s = if (std.mem.lastIndexOfScalar(u8, text[lo..from], '\n')) |nl| lo + nl + 1 else if (known) h.start else 0; + rx.line_hint = .{ .ptr = @intFromPtr(text.ptr), .len = text.len, .at = from, .start = s }; + return s; + } + /// The first match that starts in `from..=last` and ends by `hi`, as /// offsets into `text`. The text before `from` still says where lines /// begin. @@ -345,7 +375,7 @@ pub const Regex = struct { } return null; } - var start: usize = if (rx.spans) 0 else if (std.mem.lastIndexOfScalar(u8, text[0..from], '\n')) |nl| nl + 1 else 0; + var start: usize = if (rx.spans) 0 else rx.lineStart(text, from); var at = from - start; // `^` cannot match in the middle of a line. if (!rx.spans and rx.bol and at > 0) { @@ -359,7 +389,9 @@ pub const Regex = struct { const q = std.mem.indexOfPos(u8, text[0..hi], start + at, rx.lit[0..rx.lit_len]) orelse return null; if (q > last) return null; if (q > start + at) { - const line_start = if (std.mem.lastIndexOfScalar(u8, text[0..q], '\n')) |nl| nl + 1 else 0; + // `start` begins a line and `start + at` is on it: the + // look back ends there. + const line_start = if (std.mem.lastIndexOfScalar(u8, text[start + at .. q], '\n')) |nl| start + at + nl + 1 else start; if (line_start > start) { start = line_start; at = q - line_start; @@ -371,7 +403,7 @@ pub const Regex = struct { } } } - const end = if (rx.spans) hi else std.mem.indexOfScalarPos(u8, text[0..hi], start, '\n') orelse hi; + const end = if (rx.spans) hi else rx.lineEnd(text, start, start + at, hi); const line = text[start..end]; // matchPos finds nothing at a haystack's very end, where `$` or // an empty match still can. -- cgit v1.3 From bc3d18d27f9fc1155db7fac3946c4d38774f9e65 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Wed, 21 Oct 2026 14:35:15 -0300 Subject: Build with Zig 0.17 ghostty, libvaxis and uucode move to ghostty's 0.17 pins, cloud9 to its 0.17 main, SDL to castholm's release-3.4.x (3.4.18), zstbi and tree-sitter to forks with 0.17 build files (see build.zig.zon). A bare zig build copies the binaries into ~/.local/bin (-Dhome-install=false leaves it out): 0.17 keeps the prefix from build scripts. A terminal's edited copy goes when the program clears the screen. tinywin.snap is gone; perf baselines are re-recorded. Co-Authored-By: Claude Opus 5.5 --- README.md | 2 +- build.zig | 289 +++++++++++++++----------------- build.zig.zon | 45 ++--- build/snap.zig | 6 +- docs/typ/building.typ | 12 +- docs/typ/install.typ | 2 +- mupdf.zig | 6 +- src/9p_io.zig | 14 +- src/CHANGELOG.md | 8 + src/Layer.zig | 4 +- src/Messages.zig | 14 +- src/TerminalImages.zig | 82 ++++++++- src/builtins.zig | 26 +-- src/colors.zig | 70 ++++---- src/config.zig | 21 ++- src/detached/client.zig | 8 +- src/detached/wire.zig | 39 +++-- src/draw.zig | 22 +-- src/dump.zig | 24 +-- src/esp32p4.zig | 2 +- src/esp32p4/input_rescue.zig | 2 +- src/esp32p4/selftest.zig | 2 +- src/fs.zig | 8 +- src/gui/Post.zig | 14 +- src/gui/gui.zig | 20 +-- src/host_io.zig | 14 +- src/image.zig | 6 +- src/linux/v9fs.zig | 4 +- src/locations.zig | 24 +-- src/look.zig | 6 +- src/lsp/lsp.zig | 2 +- src/lsp/lsp_client.zig | 2 +- src/macos.zig | 26 +-- src/main.zig | 6 + src/memory.zig | 2 +- src/modal.zig | 2 +- src/ninep/addr.zig | 2 +- src/ninep/cols.zig | 10 +- src/ninep/ctl.zig | 14 +- src/ninep/events.zig | 10 +- src/ninep/pane.zig | 11 +- src/ninep/pty.zig | 6 +- src/ninep/screen.zig | 2 +- src/ninep/tree.zig | 16 +- src/normal.zig | 2 +- src/pager.zig | 2 +- src/panes.zig | 2 +- src/pardes.zig | 32 ++-- src/pdf.zig | 10 +- src/regexp.zig | 12 +- src/sam_edit.zig | 2 +- src/syntax.zig | 2 +- src/terminal.zig | 30 +++- src/tty/tty.zig | 22 +-- test/deck/replay.zig | 5 +- test/deck/schema.zig | 10 +- test/e2e_harness.zig | 2 +- test/fs_bench.zig | 4 +- test/history.zig | 10 +- test/hxdiff.zig | 14 +- test/image_harness.zig | 2 +- test/lspbench.zig | 2 +- test/monkey.zig | 2 +- test/panes.zig | 50 +++--- test/pdf.zig | 8 +- test/pdf_scroll_bench.zig | 2 +- test/pdf_sections_bench.zig | 2 +- test/perf-baseline-gui-Debug.json | 1 - test/perf-baseline-gui-debug.json | 1 + test/perf-baseline-tty-Debug.json | 1 - test/perf-baseline-tty-ReleaseFast.json | 1 - test/perf-baseline-tty-debug.json | 1 + test/perf-baseline-tty-fast.json | 1 + test/perf.zig | 15 +- test/snapshot.zig | 40 ++--- test/snapshots/exec.golden | 8 +- test/snapshots/tinywin.golden | 83 --------- test/snapshots/tinywin.snap | 31 ---- test/unit_profile_test.zig | 7 +- tools/gen_themes.zig | 15 +- 80 files changed, 664 insertions(+), 677 deletions(-) delete mode 100644 test/perf-baseline-gui-Debug.json create mode 100644 test/perf-baseline-gui-debug.json delete mode 100644 test/perf-baseline-tty-Debug.json delete mode 100644 test/perf-baseline-tty-ReleaseFast.json create mode 100644 test/perf-baseline-tty-debug.json create mode 100644 test/perf-baseline-tty-fast.json delete mode 100644 test/snapshots/tinywin.golden delete mode 100644 test/snapshots/tinywin.snap (limited to 'src/regexp.zig') diff --git a/README.md b/README.md index 3a566d2a..0a03bdba 100644 --- a/README.md +++ b/README.md @@ -6,7 +6,7 @@ looks; the keys are helix's; every session is a virtual filesystem, served over 9P, that any program can drive. Get the source with `git clone https://git.sr.ht/~gbrls/pardes && cd pardes` -(also at https://git.0x4200.cafe/pardes). Install with Zig 0.16: +(also at https://git.0x4200.cafe/pardes). Install with Zig 0.17: `zig build -Dplatform=tty --prefix ~/.local` for `pardes` in a terminal, or `-Dplatform=gui` for the `pardes-gui` window. The terminal should report the mouse; Shift-Esc needs the kitty keyboard diff --git a/build.zig b/build.zig index 0246d55f..ddd41a03 100644 --- a/build.zig +++ b/build.zig @@ -59,13 +59,14 @@ pub fn build(b: *std.Build) void { const requested_platform = b.option(Platform, "platform", "which shell to build (tty, gui, web, macos, esp32p4); absent builds the tty cli and the SDL gui together"); const platform = requested_platform orelse .tty; const also_gui = requested_platform == null; - const home_install = also_gui and prefixIsUntouched(b); - if (home_install) { - if (b.graph.environ_map.get("HOME")) |home| { - b.resolveInstallPrefix(b.pathJoin(&.{ home, ".local" }), .{}); - std.debug.print("pardes: installing into {s} (override with --prefix)\n", .{b.install_prefix}); - } - } + // A bare `zig build` also copies the binaries into ~/.local/bin. Zig + // 0.17 keeps the prefix in the maker process, out of this script's + // reach, so `--prefix` no longer turns it off: -Dhome-install=false does. + const home_bin: ?[]const u8 = if (also_gui and (b.option(bool, "home-install", "a bare `zig build` also copies the binaries into ~/.local/bin (default: true)") orelse true)) + if (b.graph.environ_map.get("HOME")) |home| b.pathJoin(&.{ home, ".local", "bin" }) else null + else + null; + const home_install = home_bin != null; const esp32p4_target: std.Target.Query = .{ .cpu_arch = .riscv32, .os_tag = .freestanding, @@ -114,18 +115,18 @@ pub fn build(b: *std.Build) void { const is_web_target = target.result.cpu.arch == .wasm32 and target.result.os.tag == .freestanding; const is_esp32p4_target = target.result.cpu.arch == .riscv32 and target.result.os.tag == .freestanding; const optimize = if (is_web) - .ReleaseSmall - else if (is_esp32p4_target and requested_optimize == .Debug) - .ReleaseFast + .small + else if (is_esp32p4_target and requested_optimize == .debug) + .fast else requested_optimize; const test_build = b.step("test-build", "compile unit-test programs without running them"); const doc_typ_opts = b.addOptions(); - doc_typ_opts.addOption([]const u8, "dir", b.pathFromRoot("docs/typ")); + doc_typ_opts.addOptionPathUntracked("dir", b.path("docs/typ")); // The docs (docs/typ, docs/site): each output is one typst run, and each // reads docs/typ/builtins-*.json; `builtins-json` refreshes this build's from the // registry first (the tool is wired below, once the core module is). - const typst_bin: ?[]const u8 = b.findProgram(&.{"typst"}, &.{}) catch null; + const typst_bin: ?[]const u8 = b.findProgram(.{ .names = &.{"typst"} }); var doc_renders: [3]?*std.Build.Step.Run = @splat(null); const cheatsheet_step = b.step("cheatsheet", "render docs/typ/cheatsheet-a4.typ to /share/doc/pardes/cheatsheet.pdf (needs typst)"); const book_step = b.step("book", "render docs/typ/book.typ to /share/doc/pardes/pardes.pdf (needs typst)"); @@ -139,11 +140,13 @@ pub fn build(b: *std.Build) void { sheet.addFileArg(b.path("docs/typ/cheatsheet-a4.typ")); sheet.has_side_effects = true; cheatsheet_step.dependOn(&b.addInstallFile(sheet.addOutputFileArg("cheatsheet.pdf"), "share/doc/pardes/cheatsheet.pdf").step); - const book = b.addSystemCommand(&.{ typst, "compile", "--ignore-system-fonts", "--root", b.pathFromRoot(".") }); + const book = b.addSystemCommand(&.{ typst, "compile", "--ignore-system-fonts", "--root", "." }); + book.setCwd(b.path(".")); book.addFileArg(b.path("docs/typ/book.typ")); book.has_side_effects = true; book_step.dependOn(&b.addInstallFile(book.addOutputFileArg("pardes.pdf"), "share/doc/pardes/pardes.pdf").step); - const site = b.addSystemCommand(&.{ typst, "compile", "--format", "bundle", "--features", "bundle,html", "--root", b.pathFromRoot(".") }); + const site = b.addSystemCommand(&.{ typst, "compile", "--format", "bundle", "--features", "bundle,html", "--root", "." }); + site.setCwd(b.path(".")); site.addFileArg(b.path("docs/site/site.typ")); const site_dir = site.addOutputDirectoryArg("site"); site.has_side_effects = true; @@ -168,7 +171,7 @@ pub fn build(b: *std.Build) void { .root_source_file = b.path("test/unit_profile.zig"), }), }); - const c_optimize: std.builtin.OptimizeMode = if (optimize == .Debug) .ReleaseFast else optimize; + const c_optimize: std.lang.Optimize = if (optimize == .debug) .fast else optimize; const default_grammars: TreeSitterGrammars = if (is_web) .zig else if (is_esp32p4) .disabled else .full; const requested_grammars = b.option(TreeSitterGrammars, "tree-sitter", "tree-sitter grammar set: disabled, zig, minimal (c/c++/zig), full"); const tree_sitter_grammars = requested_grammars orelse default_grammars; @@ -321,7 +324,7 @@ pub fn build(b: *std.Build) void { .name = "pardes-gen-themes", .root_module = b.createModule(.{ .target = b.graph.host, - .optimize = .Debug, + .optimize = .debug, .root_source_file = b.path("tools/gen_themes.zig"), }), }); @@ -353,7 +356,7 @@ pub fn build(b: *std.Build) void { // `pardes ` alone, like a tarball build; -Dstamp-commit=true // stamps one anyway. const stamp_commit = b.option(bool, "stamp-commit", "embed the git commit in --version and crash records (default: release builds and the ~/.local install)") orelse - (optimize != .Debug or home_install); + (optimize != .debug or home_install); const shell_cfg: ShellConfig = .{ .tree_sitter_grammars = tree_sitter_grammars, .tracy = tracy != null, @@ -391,9 +394,7 @@ pub fn build(b: *std.Build) void { if (tree_sitter_grammars != .disabled) { const tree_sitter_mod = b.dependency("tree_sitter", .{ .target = target, .optimize = c_optimize }).module("tree_sitter"); if (is_web) { - const web_libc_include: std.Build.LazyPath = .{ - .cwd_relative = b.path("src/web/libc/include").getPath(b), - }; + const web_libc_include = b.path("src/web/libc/include"); for (tree_sitter_mod.link_objects.items) |link_object| switch (link_object) { .other_step => |lib| { lib.root_module.link_libc = false; @@ -417,12 +418,11 @@ pub fn build(b: *std.Build) void { (tree_sitter_grammars == .zig and g.tier == .zig)) { const dep = b.dependency(g.dep, .{}); - const query_path = dep.path(g.query); - const query = std.Io.Dir.cwd().readFileAlloc(io, query_path.getPath(b), b.allocator, .limited(0x100000)) catch @panic("read " ++ g.name ++ " highlights"); + const query = configureRead(b, io, dep.builder.root, g.query, 0x100000) catch @panic("read " ++ g.name ++ " highlights"); ts_queries.addOption([]const u8, g.name ++ "_highlights", query); // helix's textobject queries, vendored (vendor/queries); a // grammar without one gets none - const objects = std.Io.Dir.cwd().readFileAlloc(io, b.pathFromRoot("vendor/queries/" ++ g.name ++ "/textobjects.scm"), b.allocator, .limited(0x100000)) catch ""; + const objects = configureRead(b, io, b.root, "vendor/queries/" ++ g.name ++ "/textobjects.scm", 0x100000) catch ""; ts_queries.addOption([]const u8, g.name ++ "_textobjects", objects); const lib = b.addLibrary(.{ @@ -438,8 +438,9 @@ pub fn build(b: *std.Build) void { const ts_cflags = [_][]const u8{ "-fno-sanitize=function", "-DTREE_SITTER_REUSE_ALLOCATOR", + // Relative to the build root, the compiler's working directory. "-include", - b.path("src/tree_sitter_heap.h").getPath(b), + "src/tree_sitter_heap.h", }; lib.root_module.addCSourceFile(.{ .file = dep.path(g.src ++ "/parser.c"), .flags = &ts_cflags }); if (g.scanner) lib.root_module.addCSourceFile(.{ .file = dep.path(g.src ++ "/scanner.c"), .flags = &ts_cflags }); @@ -467,7 +468,7 @@ pub fn build(b: *std.Build) void { // (src/regexp.zig). A bump that moves the anchor stops the build. const mvzr_mod = mvzr: { const dep = b.dependency("mvzr", .{ .target = target, .optimize = optimize }); - const src = std.Io.Dir.cwd().readFileAlloc(io, dep.path("src/mvzr.zig").getPath(b), b.allocator, .limited(1 << 20)) catch @panic("read mvzr"); + const src = configureRead(b, io, dep.builder.root, "src/mvzr.zig", 1 << 20) catch @panic("read mvzr"); const anchor = "fn matchPattern(patt: []const RegOp, sets: []const CharSet, haystack: []const u8, i_in: usize) ?OpMatch {\n"; const at = (std.mem.indexOf(u8, src, anchor) orelse @panic("mvzr's matchPattern changed: redo its step budget in build.zig")) + anchor.len; const spend = @@ -536,7 +537,17 @@ pub fn build(b: *std.Build) void { \\ ; if (std.mem.count(u8, backed, end_anchor) != 1) @panic("mvzr's parser changed: redo its `$` before `|` in build.zig"); - const patched = std.mem.replaceOwned(u8, b.allocator, backed, end_anchor, end_before_alt) catch @panic("OOM"); + var patched = std.mem.replaceOwned(u8, b.allocator, backed, end_anchor, end_before_alt) catch @panic("OOM"); + // Zig 0.17 removed `**`; mvzr still has it (its PR #16 replaces it upstream). + for ([_][3][]const u8{ + .{ "[1]RegOp{.{ .unused = {} }} ** ops", "@splat(.{ .unused = {} })", "1" }, + .{ "[1]CharSet{.{ .low = 0, .hi = 0 }} ** char_sets", "@splat(.{ .low = 0, .hi = 0 })", "1" }, + .{ "\"a\" ** 2048", "&@as([2048]u8, @splat('a'))", "2" }, + .{ "\"12345,\" ** 255", "std.mem.asBytes(&@as([255][6]u8, @splat(\"12345,\".*)))", "1" }, + }) |edit| { + if (std.mem.count(u8, patched, edit[0]) != std.fmt.parseInt(usize, edit[2], 10) catch unreachable) std.debug.panic("mvzr changed at `{s}`: redo its `**` patch in build.zig", .{edit[0]}); + patched = std.mem.replaceOwned(u8, b.allocator, patched, edit[0], edit[1]) catch @panic("OOM"); + } break :mvzr b.createModule(.{ .root_source_file = b.addWriteFiles().add("mvzr.zig", patched), .target = target, .optimize = optimize }); }; @@ -582,6 +593,7 @@ pub fn build(b: *std.Build) void { }); mupdf.linkTo(mupdf_mod); mupdf_mod.addImport("c_heap", c_heap_mod); + mupdf_mod.addImport("pdf_c", b.addTranslateC(.{ .root_source_file = b.path("src/pdf_bridge.h"), .target = target, .optimize = optimize }).createModule()); mupdf_mod.addIncludePath(b.path("src")); mupdf_mod.addCSourceFile(.{ .file = b.path("src/pdf_bridge.c"), @@ -599,12 +611,13 @@ pub fn build(b: *std.Build) void { const pdf_bench_mod = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("src/pdf.zig"), .link_libc = true, }); mupdf.linkTo(pdf_bench_mod); pdf_bench_mod.addImport("c_heap", c_heap_mod); + pdf_bench_mod.addImport("pdf_c", b.addTranslateC(.{ .root_source_file = b.path("src/pdf_bridge.h"), .target = target, .optimize = .fast }).createModule()); pdf_bench_mod.addIncludePath(b.path("src")); pdf_bench_mod.addCSourceFile(.{ .file = b.path("src/pdf_bridge.c"), @@ -614,14 +627,14 @@ pub fn build(b: *std.Build) void { .name = "pardes-pdf-bench", .root_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/pdf_bench.zig"), .link_libc = true, }), }); pdf_bench.root_module.addImport("mupdf", pdf_bench_mod); const run_pdf_bench = b.addRunArtifact(pdf_bench); - if (b.args) |args| run_pdf_bench.addArgs(args); + run_pdf_bench.addPassthruArgs(); run_pdf_bench.setCwd(b.path(".")); pdf_bench_step.dependOn(&run_pdf_bench.step); @@ -629,7 +642,7 @@ pub fn build(b: *std.Build) void { .name = "pardes-pdf-sections-bench", .root_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/pdf_sections_bench.zig"), .link_libc = true, }), @@ -637,10 +650,10 @@ pub fn build(b: *std.Build) void { pdf_sections_bench.root_module.addImport("pardes", core_module); pdf_sections_bench.root_module.addImport("mupdf", mupdf_mod); const sections_bench_opts = b.addOptions(); - sections_bench_opts.addOption(bool, "release_fast_core", optimize == .ReleaseFast); + sections_bench_opts.addOption(bool, "release_fast_core", optimize == .fast); pdf_sections_bench.root_module.addOptions("pdf_sections_bench_config", sections_bench_opts); const run_pdf_sections_bench = b.addRunArtifact(pdf_sections_bench); - if (b.args) |args| run_pdf_sections_bench.addArgs(args); + run_pdf_sections_bench.addPassthruArgs(); run_pdf_sections_bench.setCwd(b.path(".")); pdf_sections_bench_step.dependOn(&run_pdf_sections_bench.step); @@ -648,7 +661,7 @@ pub fn build(b: *std.Build) void { .name = "pardes-pdf-pointer-bench", .root_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/pdf_pointer_bench.zig"), .link_libc = true, }), @@ -656,7 +669,7 @@ pub fn build(b: *std.Build) void { pdf_pointer_bench.root_module.addImport("pardes", core_module); pdf_pointer_bench.root_module.addImport("mupdf", mupdf_mod); const run_pdf_pointer_bench = b.addRunArtifact(pdf_pointer_bench); - if (b.args) |args| run_pdf_pointer_bench.addArgs(args); + run_pdf_pointer_bench.addPassthruArgs(); run_pdf_pointer_bench.setCwd(b.path(".")); pdf_pointer_bench_step.dependOn(&run_pdf_pointer_bench.step); @@ -672,11 +685,11 @@ pub fn build(b: *std.Build) void { pdf_scroll_bench.root_module.addImport("pardes", core_module); pdf_scroll_bench.root_module.addImport("mupdf", mupdf_mod); const scroll_bench_opts = b.addOptions(); - scroll_bench_opts.addOption(bool, "release_fast_core", optimize == .ReleaseFast); + scroll_bench_opts.addOption(bool, "release_fast_core", optimize == .fast); scroll_bench_opts.addOption([]const u8, "core_optimize", @tagName(optimize)); pdf_scroll_bench.root_module.addOptions("pdf_scroll_bench_config", scroll_bench_opts); const run_pdf_scroll_bench = b.addRunArtifact(pdf_scroll_bench); - if (b.args) |args| run_pdf_scroll_bench.addArgs(args); + run_pdf_scroll_bench.addPassthruArgs(); run_pdf_scroll_bench.setCwd(b.path(".")); pdf_scroll_bench_step.dependOn(&run_pdf_scroll_bench.step); } @@ -712,15 +725,13 @@ pub fn build(b: *std.Build) void { // - An X10 mouse report and a kitty graphics reply cut short by a read // wait for their rest: the first became three bytes of text, the // second sliced past its input. - // - A malformed color or clipboard reply is dropped instead of failing - // the parse, which stopped input; so is a byte no UTF-8 starts with. - // - An OSC reply ends at its first terminator: one ended by BEL ran on to - // the next sequence's ESC and left what lay between as text. + // - A byte no UTF-8 starts with is dropped instead of failing the parse, + // which stopped input. (vaxis itself now drops a malformed color or + // clipboard reply and ends an OSC at its first terminator.) // - A device report's fields are sought in the sequence, not past it. // A bump that moves an anchor stops the build. { - const parser_path = vaxis_dep.path("src/Parser.zig").getPath(b); - var src = std.Io.Dir.cwd().readFileAlloc(io, parser_path, b.allocator, .limited(1 << 20)) catch @panic("read vaxis Parser.zig"); + var src = configureRead(b, io, vaxis_dep.builder.root, "src/Parser.zig", 1 << 20) catch @panic("read vaxis Parser.zig"); const edits = [_]struct { anchor: []const u8, with: []const u8, count: usize = 1 }{ .{ .anchor = @@ -731,7 +742,7 @@ pub fn build(b: *std.Build) void { \\ 0x5F => return parseApc(input), \\ // pardes's patch (its build.zig): ESC before a key's sequence is Alt on the key. \\ 0x1B => if (input.len > 2 and (input[2] == '[' or input[2] == 'O')) { - \\ var inner = if (input[2] == '[') parseCsi(input[1..], &self.buf) else parseSs3(input[1..]); + \\ var inner = if (input[2] == '[') parseCsi(input[1..], &self.buf, self.cursor_position_requests) else parseSs3(input[1..]); \\ if (inner.n == 0) return inner; \\ if (inner.event) |*event| switch (event.*) { \\ .key_press, .key_release => |*key| key.mods.alt = true, @@ -765,71 +776,17 @@ pub fn build(b: *std.Build) void { \\ switch (input[2]) { , }, - .{ - .anchor = - \\ const esc_result = skipUntilST(input); - \\ if (esc_result.n > 0) break :blk esc_result.n; - \\ - \\ // No escape, could be BEL terminated - \\ const bel = std.mem.indexOfScalarPos(u8, input, 2, 0x07) orelse return .{ - \\ .event = null, - \\ .n = 0, - \\ }; - \\ bel_terminated = true; - \\ break :blk bel + 1; - , - .with = - \\ // pardes's patch (its build.zig): whichever terminator comes first ends it. - \\ const esc = std.mem.indexOfScalarPos(u8, input, 2, 0x1b); - \\ const bel = std.mem.indexOfScalarPos(u8, input, 2, 0x07); - \\ if (bel != null and (esc == null or bel.? < esc.?)) { - \\ bel_terminated = true; - \\ break :blk bel.? + 1; - \\ } - \\ const esc_result = skipUntilST(input); - \\ if (esc_result.n > 0) break :blk esc_result.n; - \\ return .{ .event = null, .n = 0 }; - , - }, - .{ - .anchor = " const color = try Color.rgbFromSpec(color_spec);\n", - .with = " const color = Color.rgbFromSpec(color_spec) catch return null_event; // pardes's patch (its build.zig)\n", - .count = 2, - }, - .{ - .anchor = - \\ if (input[semicolon_idx + 1] != 'c') return null_event; - \\ const payload = if (bel_terminated) - \\ input[semicolon_idx + 3 .. sequence.len - 1] - \\ else - \\ input[semicolon_idx + 3 .. sequence.len - 2]; - \\ const decoder = std.base64.standard.Decoder; - \\ const text = try paste_allocator.?.alloc(u8, try decoder.calcSizeForSlice(payload)); - \\ try decoder.decode(text, payload); - , - .with = - \\ // pardes's patch (its build.zig): a malformed reply is dropped. - \\ const body_end = if (bel_terminated) sequence.len - 1 else sequence.len - 2; - \\ if (semicolon_idx + 3 > body_end or input[semicolon_idx + 1] != 'c') return null_event; - \\ const payload = input[semicolon_idx + 3 .. body_end]; - \\ const decoder = std.base64.standard.Decoder; - \\ const gpa = paste_allocator orelse return null_event; - \\ const text = gpa.alloc(u8, decoder.calcSizeForSlice(payload) catch return null_event) catch return null_event; - \\ decoder.decode(text, payload) catch { - \\ gpa.free(text); - \\ return null_event; - \\ }; - , - }, .{ .anchor = \\ 'A', 'B', 'C', 'D', 'E', 'F', 'H', 'P', 'Q', 'R', 'S' => { + \\ // Legacy keys \\ , .with = \\ 'A', 'B', 'C', 'D', 'E', 'F', 'H', 'P', 'Q', 'R', 'S' => { \\ // pardes's patch (its build.zig): a cursor position report is no key. \\ if (final == 'R' and cursorReport(sequence)) return null_event; + \\ // Legacy keys \\ , }, @@ -882,8 +839,7 @@ pub fn build(b: *std.Build) void { // (InactiveDim on a focus change) sent every blank of every pane it // touched again: 7 KB a click for three shell panes. The terminal // keeps the old ink there, and so does vaxis's last frame. - const render_path = vaxis_dep.path("src/Vaxis.zig").getPath(b); - var render_src = std.Io.Dir.cwd().readFileAlloc(io, render_path, b.allocator, .limited(1 << 20)) catch @panic("read vaxis Vaxis.zig"); + var render_src = configureRead(b, io, vaxis_dep.builder.root, "src/Vaxis.zig", 1 << 20) catch @panic("read vaxis Vaxis.zig"); const render_anchor = \\ if ((!self.refresh and \\ last.eql(cell) and @@ -941,7 +897,7 @@ pub fn build(b: *std.Build) void { var ghostty_vt_for_snap: ?*std.Build.Module = null; const ghostty_simd = !freestanding_core and (!target.result.os.tag.isDarwin() or b.graph.host.result.os.tag.isDarwin()); - const ghostty_optimize: std.builtin.OptimizeMode = if (optimize == .Debug) .ReleaseSafe else optimize; + const ghostty_optimize: std.lang.Optimize = if (optimize == .debug) .safe else optimize; const ghostty_dep = if (is_esp32p4) null else b.lazyDependency("ghostty", .{ .target = target, .optimize = ghostty_optimize, .simd = ghostty_simd, .@"app-runtime" = .none, .@"emit-xcframework" = false }); if (ghostty_dep) |dep| { const ghostty_vt = dep.module("ghostty-vt"); @@ -1017,6 +973,17 @@ pub fn build(b: *std.Build) void { const sdl_lib = dep.artifact("SDL3"); gm.linkLibrary(sdl_lib); gm.addIncludePath(dep.path("include")); + // gui.zig's `c`: SDL and the font shim, which @cImport gave it before Zig 0.17. + const header = b.addWriteFiles().add("gui_c.h", + \\#define SDL_DISABLE_OLD_NAMES 1 + \\#include + \\#include "font.h" + \\ + ); + const gui_c = b.addTranslateC(.{ .root_source_file = header, .target = target, .optimize = optimize }); + gui_c.addIncludePath(dep.path("include")); + gui_c.addIncludePath(b.path("src/gui")); + gm.addImport("gui_c", gui_c.createModule()); } gm.linkLibrary(freetype.?); // HarfBuzz from source, like FreeType: its FreeType integration @@ -1105,7 +1072,7 @@ pub fn build(b: *std.Build) void { }) |file| web_step.dependOn(&b.addInstallFileWithDir(b.path(file[0]), .{ .custom = "web" }, file[1]).step); const run_web_harness = b.addSystemCommand(&.{ "node", "test/web_harness.mjs" }); - run_web_harness.addArg(b.getInstallPath(.{ .custom = "web" }, "")); + run_web_harness.addDirectoryArg(installed("web")); run_web_harness.step.dependOn(web_step); b.step("web-harness", "run the dependency-free JS/WASM DOM harness").dependOn(&run_web_harness.step); b.getInstallStep().dependOn(web_step); @@ -1149,9 +1116,9 @@ pub fn build(b: *std.Build) void { const min_version = b.fmt("{d}.{d}", .{ macos_min_version.major, macos_min_version.minor }); const swift_mode: []const u8 = switch (optimize) { - .Debug => "-Onone", - .ReleaseSmall => "-Osize", - .ReleaseFast, .ReleaseSafe => "-O", + .debug => "-Onone", + .small => "-Osize", + .fast, .safe => "-O", }; const link = b.addSystemCommand(&.{ "swiftc", swift_mode, "-target", triple, "-import-objc-header" }); @@ -1218,7 +1185,8 @@ pub fn build(b: *std.Build) void { const sign = b.addSystemCommand(&.{ "codesign", "--force", "--sign", macos_identity }); if (!std.mem.eql(u8, macos_identity, "-")) sign.addArgs(&.{ "--options", "runtime", "--timestamp" }); - sign.addArg(b.getInstallPath(.prefix, "pardes.app")); + sign.setCwd(installed("")); + sign.addArg("pardes.app"); sign.has_side_effects = true; sign.step.dependOn(&install_app_bin.step); sign.step.dependOn(&install_plist.step); @@ -1227,8 +1195,8 @@ pub fn build(b: *std.Build) void { app_step.dependOn(&sign.step); const dmg = b.addSystemCommand(&.{ "hdiutil", "create", "-volname", "pardes", "-ov", "-format", "UDZO", "-srcfolder" }); - dmg.addArg(b.getInstallPath(.prefix, "pardes.app")); - dmg.addArg(b.getInstallPath(.prefix, "pardes.dmg")); + dmg.setCwd(installed("")); + dmg.addArgs(&.{ "pardes.app", "pardes.dmg" }); dmg.has_side_effects = true; dmg.step.dependOn(&sign.step); dmg_step.dependOn(&dmg.step); @@ -1267,7 +1235,7 @@ pub fn build(b: *std.Build) void { const run_e2e = std.Build.Step.Run.create(b, "run macos-e2e"); run_e2e.addFileArg(runnable); run_e2e.setCwd(b.path(".")); - if (b.args) |args| run_e2e.addArgs(args); + run_e2e.addPassthruArgs(); run_e2e.has_side_effects = true; e2e_step.dependOn(&run_e2e.step); } else e2e_step.dependOn(&b.addFail("macos-e2e needs a Darwin host and target").step); @@ -1290,7 +1258,7 @@ pub fn build(b: *std.Build) void { .linkage = if (static) .static else null, .root_module = root_mod, }); - b.installArtifact(exe); + installHome(b, exe, home_bin); if (gui_mod) |gm| { var gui_name: []const u8 = "pardes-gui"; if (target.result.os.tag != .linux or target.result.cpu.arch != .x86_64) @@ -1300,7 +1268,7 @@ pub fn build(b: *std.Build) void { .linkage = if (static) .static else null, .root_module = gm, }); - b.installArtifact(gui_exe); + installHome(b, gui_exe, home_bin); } const run = b.addRunArtifact(exe); b.step("run", "descend to the pardes").dependOn(&run.step); @@ -1312,7 +1280,7 @@ pub fn build(b: *std.Build) void { }); const run_iso = b.addRunArtifact(iso_exe); // Something to look at on arrival, read out of the binary itself. - if (b.args) |args| run_iso.addArgs(args) else run_iso.addArg("src/pardes.zig"); + run_iso.addPassthruArgs(); const iso_step = b.step("run-isolated", "descend to a pardes with no host but the terminal"); iso_step.dependOn(&run_iso.step); } @@ -1370,15 +1338,9 @@ pub fn build(b: *std.Build) void { hxdiff.root_module.addImport("pardes", core_module); const run_hxdiff = b.addRunArtifact(hxdiff); run_hxdiff.has_side_effects = true; - if (b.args) |args| run_hxdiff.addArgs(args) else { - run_hxdiff.addArgs(&.{ - "--strict", - "test/hxcases/cases.jsonl", - "test/hxcases/goldens.jsonl", - "test/hxcases/waivers.jsonl", - }); - run_hxdiff.setCwd(b.path(".")); - } + // Bare, it runs the checked-in suite (test/hxdiff.zig). + run_hxdiff.addPassthruArgs(); + run_hxdiff.setCwd(b.path(".")); const hxdiff_step = b.step("hxdiff", "compare Helix results (-- --strict cases reference [waivers]; cases alone emits)"); hxdiff_step.dependOn(&run_hxdiff.step); @@ -1403,16 +1365,8 @@ pub fn build(b: *std.Build) void { const run_hxparity = b.addRunArtifact(hxdiff); run_hxparity.has_side_effects = true; run_hxparity.addArg("--parity"); - if (b.args) |args| run_hxparity.addArgs(args) else { - run_hxparity.addArgs(&.{ - "--strict", - "--waivers", - "test/hxcases/parity-waivers.jsonl", - "test/hxcases/cases.jsonl", - "test/hxcases/parity.jsonl", - }); - run_hxparity.setCwd(b.path(".")); - } + run_hxparity.addPassthruArgs(); + run_hxparity.setCwd(b.path(".")); const hxparity_step = b.step("hxparity", "run the file-vs-pty editing parity suite (-- [--waivers w.jsonl] ...)"); hxparity_step.dependOn(&run_hxparity.step); @@ -1427,21 +1381,21 @@ pub fn build(b: *std.Build) void { .name = "pardes-lspbench", .root_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/lspbench.zig"), .link_libc = true, }), }); lspbench.root_module.addImport("pardes", core_module); const run_lspbench = b.addRunArtifact(lspbench); - if (b.args) |args| run_lspbench.addArgs(args); + run_lspbench.addPassthruArgs(); run_lspbench.setCwd(b.path(".")); run_lspbench.has_side_effects = true; const lspbench_step = b.step("lspbench", "language-backend latency + feature matrix (-- [--json] [--check] [repo-root]), or one query (-- probe : [arg] [--reps N])"); lspbench_step.dependOn(&run_lspbench.step); const run_lspbench_check = b.addRunArtifact(lspbench); run_lspbench_check.addArg("--check"); - if (b.args) |args| run_lspbench_check.addArgs(args); + run_lspbench_check.addPassthruArgs(); run_lspbench_check.setCwd(b.path(".")); run_lspbench_check.has_side_effects = true; const lspbench_check_step = b.step("lspbench-check", "require every configured language probe to complete correctly"); @@ -1449,12 +1403,12 @@ pub fn build(b: *std.Build) void { const locations_bench_step = b.step("locations-bench", "measure cold/warm result formatting (-Doptimize=ReleaseFast; CSV on stderr)"); - if (optimize != .ReleaseFast) { + if (optimize != .fast) { locations_bench_step.dependOn(&b.addFail("locations-bench requires -Doptimize=ReleaseFast").step); } else { const locations_bench_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/locations_bench.zig"), .link_libc = true, }); @@ -1465,7 +1419,7 @@ pub fn build(b: *std.Build) void { locations_bench_step.dependOn(&run_locations_bench.step); } - const perf_optimize: std.builtin.OptimizeMode = .ReleaseFast; + const perf_optimize: std.lang.Optimize = .fast; const perf_options = b.addOptions(); const perf_target = std.Target.Query.fromTarget(&target.result); perf_options.addOption(PerfBuild, "identity", .{ @@ -1499,7 +1453,7 @@ pub fn build(b: *std.Build) void { perf.root_module.addOptions("perf_options", perf_options); const run_perf = b.addRunArtifact(perf); run_perf.has_side_effects = true; - if (b.args) |args| run_perf.addArgs(args); + run_perf.addPassthruArgs(); run_perf.setCwd(b.path(".")); const perf_step = b.step("perf", "gesture latency or bounded terminal stress (-- [--json] [--reps N] [--base old.json] [--terminal-mib N])"); perf_step.dependOn(&run_perf.step); @@ -1536,7 +1490,7 @@ pub fn build(b: *std.Build) void { .name = "pardes-fs-bench", .root_module = b.createModule(.{ .target = target, - .optimize = .ReleaseFast, + .optimize = .fast, .root_source_file = b.path("test/fs_bench.zig"), .link_libc = true, }), @@ -1544,7 +1498,7 @@ pub fn build(b: *std.Build) void { fs_bench.root_module.addImport("pardes", core_module); const run_fs_bench = b.addRunArtifact(fs_bench); run_fs_bench.has_side_effects = true; - if (b.args) |args| run_fs_bench.addArgs(args); + run_fs_bench.addPassthruArgs(); run_fs_bench.setCwd(b.path(".")); const fs_bench_step = b.step("fs-bench", "9P filesystem request cost and allocation count (-- [--json] [--reps N])"); fs_bench_step.dependOn(&run_fs_bench.step); @@ -1581,11 +1535,11 @@ pub fn build(b: *std.Build) void { }); syntax_runner.root_module.addImport("syntax", syntax_module); const run_syntax = b.addRunArtifact(syntax_runner); - if (b.args) |args| run_syntax.addArgs(args); + run_syntax.addPassthruArgs(); b.step("syntax", "emit deterministic syntax snapshots").dependOn(&run_syntax.step); const run_syntax_bench = b.addRunArtifact(syntax_runner); run_syntax_bench.addArg("--bench"); - if (b.args) |args| run_syntax_bench.addArgs(args); + run_syntax_bench.addPassthruArgs(); run_syntax_bench.has_side_effects = true; b.step("syntax-bench", "measure highlighting latency and allocations").dependOn(&run_syntax_bench.step); const syntax_test = b.addTest(.{ @@ -1611,7 +1565,7 @@ pub fn build(b: *std.Build) void { // part of fs-test; the long run is manual. const run_monkey9p = b.addSystemCommand(&.{ "python3", "-B", "test/monkey9p.py" }); run_monkey9p.addArtifactArg(exe); - if (b.args) |args| run_monkey9p.addArgs(args); + run_monkey9p.addPassthruArgs(); run_monkey9p.setCwd(b.path(".")); run_monkey9p.has_side_effects = true; b.step("monkey-9p", "random 9P operations checking the documented rules (-- [--seed N]... [--steps M] [--replay F] [--shrink F] [--no-shrink])").dependOn(&run_monkey9p.step); @@ -1622,7 +1576,7 @@ pub fn build(b: *std.Build) void { if (platform == .gui) { const run_monkey_gui = b.addSystemCommand(&.{ "python3", "-B", "test/gui_monkey.py" }); run_monkey_gui.addArtifactArg(exe); - if (b.args) |args| run_monkey_gui.addArgs(args); + run_monkey_gui.addPassthruArgs(); run_monkey_gui.setCwd(b.path(".")); run_monkey_gui.has_side_effects = true; monkey_gui_step.dependOn(&run_monkey_gui.step); @@ -1653,7 +1607,7 @@ pub fn build(b: *std.Build) void { }), }); // Ordinary executable; each launch asks sudo inside its own PTY. - b.installArtifact(v9fs_helper); + installHome(b, v9fs_helper, home_bin); const run_v9fs = b.addSystemCommand(&.{ "python3", "-B", "test/v9fs.py" }); run_v9fs.addArtifactArg(exe); run_v9fs.addArtifactArg(v9fs_helper); @@ -1688,7 +1642,7 @@ pub fn build(b: *std.Build) void { }), }); const run_history = b.addRunArtifact(history); - if (b.args) |args| run_history.addArgs(args); + run_history.addPassthruArgs(); run_history.has_side_effects = true; b.step("history", "compare command output and test/build runtime across jj revisions").dependOn(&run_history.step); @@ -1843,7 +1797,7 @@ pub fn build(b: *std.Build) void { if (unit_match) |check| check.addArtifactArg(shell_test); const run_profile = b.addRunArtifact(unit_profile); run_profile.addArtifactArg(shell_test); - if (b.args) |args| run_profile.addArgs(args); + run_profile.addPassthruArgs(); run_profile.has_side_effects = true; b.step("unit-profile", "report test request time and process memory as JSONL (-- optional filter, --timeout-ms=N)").dependOn(&run_profile.step); const profile_test = b.addExecutable(.{ @@ -1876,19 +1830,38 @@ fn testMatch(b: *std.Build, profiler: *std.Build.Step.Compile, filters: []const return check; } -fn prefixIsUntouched(b: *std.Build) bool { - if (b.dest_dir != null) return false; - const zig_out = b.build_root.join(b.allocator, &.{"zig-out"}) catch @panic("OOM"); - if (!std.mem.eql(u8, b.install_prefix, zig_out)) return false; - return std.mem.eql(u8, b.exe_dir, b.pathJoin(&.{ b.install_path, "bin" })) and - std.mem.eql(u8, b.lib_dir, b.pathJoin(&.{ b.install_path, "lib" })) and - std.mem.eql(u8, b.h_dir, b.pathJoin(&.{ b.install_path, "include" })); +/// Installs `artifact` in the prefix and, given `home_bin`, copies it there +/// too (`install` unlinks first, so a running binary is replaced). +fn installHome(b: *std.Build, artifact: *std.Build.Step.Compile, home_bin: ?[]const u8) void { + b.installArtifact(artifact); + const dir = home_bin orelse return; + const copy = b.addSystemCommand(&.{ "install", "-Dm755" }); + copy.addArtifactArg(artifact); + copy.addArg(b.pathJoin(&.{ dir, artifact.name })); + copy.has_side_effects = true; + b.getInstallStep().dependOn(©.step); +} + +/// A file read while configuring, under a package's root. A dependency's is +/// hash-addressed and never changes under the same pin; this repo's are not +/// tracked one by one, since the configure runs again on every build anyway +/// (findProgram and gitCommit poison its cache). +// ponytail: untracked repo reads; b.dependOnFileContents each one if the configure cache ever stays pure. +fn configureRead(b: *std.Build, io: std.Io, root: std.Build.Cache.Path, sub: []const u8, limit: usize) ![]u8 { + return root.root_dir.handle.readFileAlloc(io, b.pathJoin(&.{ root.subPathOrDot(), sub }), b.allocator, .limited(limit)); +} + +/// `sub` under the install prefix, known only when the build runs. +fn installed(sub: []const u8) std.Build.LazyPath { + return .{ .relative = .{ .base = .install_prefix, .sub_path = sub } }; } fn gitCommit(b: *std.Build) ?[]const u8 { + // HEAD is not a file the configure cache can track. + b.graph.poisonCache(); var code: u8 = 0; const out = b.runAllowFail( - &.{ "git", "-C", b.build_root.path orelse ".", "rev-parse", "--short=12", "HEAD" }, + &.{ "git", "rev-parse", "--short=12", "HEAD" }, &code, .ignore, ) catch return null; @@ -1985,7 +1958,7 @@ fn compileGlsl(b: *std.Build, name: []const u8) std.Build.LazyPath { fn vendoredThemes(b: *std.Build, io: std.Io) []const []const u8 { var count: usize = 0; { - var dir = b.build_root.handle.openDir(io, "vendor/themes", .{ .iterate = true }) catch @panic("open vendor/themes"); + var dir = b.root.root_dir.handle.openDir(io, "vendor/themes", .{ .iterate = true }) catch @panic("open vendor/themes"); defer dir.close(io); var it = dir.iterate(); while (it.next(io) catch @panic("read vendor/themes")) |e| { @@ -1995,7 +1968,7 @@ fn vendoredThemes(b: *std.Build, io: std.Io) []const []const u8 { } const names = b.allocator.alloc([]const u8, count) catch @panic("OOM"); - var dir = b.build_root.handle.openDir(io, "vendor/themes", .{ .iterate = true }) catch @panic("open vendor/themes"); + var dir = b.root.root_dir.handle.openDir(io, "vendor/themes", .{ .iterate = true }) catch @panic("open vendor/themes"); defer dir.close(io); var it = dir.iterate(); var index: usize = 0; diff --git a/build.zig.zon b/build.zig.zon index b1f501fe..a2fc90dd 100644 --- a/build.zig.zon +++ b/build.zig.zon @@ -9,52 +9,59 @@ // read-only HTTPS URL is what a manifest can carry. Re-pin with // `zig fetch --save=cloud9 git+https://git.sr.ht/~gbrls/cloud9#`. .cloud9 = .{ - .url = "git+https://git.sr.ht/~gbrls/cloud9#fe9772b178575fcab93de09fc61c2f13db7db1a8", - .hash = "cloud9-0.1.0-yt86qh4qHgC6OpG4gZE9yoQaO8huc5mnQCaqGwuJ0OU1", + .url = "git+https://git.sr.ht/~gbrls/cloud9#d5c37d9122240fe95d308fbe15874e14dd786c5a", + .hash = "cloud9-0.1.0-yt86qnAvHgB1dEWGakukcJ54GZ7J3mv_lm_kApzAuEOY", }, // Pinned to the exact libvaxis commit ghostty is built against, so it // is compatible with ghostty's uucode configuration (the one we share // with it via `external_uucode` in build.zig). .vaxis = .{ - .url = "git+https://github.com/rockorager/libvaxis.git#1dbbe575dff4586fe51e3217aa5c3fecdcbb6089", - .hash = "vaxis-0.6.0-BWNV_CrbCQCscGpzsAlR402rYQ_tV3aAl081c2iRRkka", + .url = "git+https://github.com/rockorager/libvaxis.git#6fd944a27fb3d6f596e981076381a3131f2448b4", + .hash = "vaxis-0.6.0-BWNV_MafDAAEzkVUvrW9NFUqnnMUoeHqrs2mecnD6D7w", }, .ghostty = .{ - .url = "git+https://github.com/ghostty-org/ghostty.git#82e53e3f6e219e0cf0499fe29c82d925ea26cd5e", - .hash = "ghostty-1.3.2-dev-5UdBCzeJJQXw_vG5Hsfx0DYUmF7Exs_9yctijqOIfp7W", + .url = "git+https://github.com/ghostty-org/ghostty.git#521cf2c5d30bd48028ca61546c64e7913764073d", + .hash = "ghostty-1.3.2-dev-5UdBC6WZfwXgQ6HmA5DH6dGNAUqtDMLrhhrHiBX-99J0", }, // Same uucode commit ghostty and vaxis both depend on. We build it // ourselves (with uucode_config.zig) so ghostty-vt and vaxis can share // a single uucode module. See build.zig for the why. .uucode = .{ - .url = "git+https://github.com/jacobsandlund/uucode#2826a37a4562284fdacd8fa029d49509cc9bffcd", - .hash = "uucode-0.2.0-ZZjBPlK5VADj7fdoq7G8LIHzD5o6FSkcBXXrRWr4jnrA", + .url = "git+https://github.com/jacobsandlund/uucode#ea62149739404a73c202b48a33bf6dd2af4bd9b0", + .hash = "uucode-0.2.0-ZZjBPoGFVgCiflJNFc6Io8xzC67B6PA5bUkk4LT0kCxg", }, + // Pivok7's fork: zig-gamedev/zstbi PR #16, "Upgrade to zig 0.17.0" + // (two lines of build.zig). Back to upstream once it is merged. .zstbi = .{ - .url = "git+https://github.com/zig-gamedev/zstbi#3813f5f113b26f644cfa01a30155cc76b3526e91", - .hash = "zstbi-0.11.0-dev-L0Ea_3GWBwAh7m54XX055vwLKPTPij5hyqyeq92MXPE2", + .url = "git+https://github.com/Pivok7/zstbi#bc96f28d341f9c7fb9cbc5dc65aa4c5cf0c8f007", + .hash = "zstbi-0.11.0-dev-L0Ea_3GWBwCpNHVwGU4MPR_oDd0FoQOYNrGodwvJ6Z45", }, - // SDL3 (castholm/SDL zig bindings) — only fetched/compiled when `-Dsdl` - // is set. Lazy so the default vaxis build never touches it. The sdl + // SDL3 (castholm/SDL zig bindings, its release-3.4.x branch for Zig + // 0.17; david-vanderson's fork, pinned before, has no 0.17 build) — only + // fetched/compiled when `-Dsdl` is set. Lazy so the default vaxis build never touches it. The sdl // package resolves its own linux deps from its own build.zig.zon. .sdl = .{ - .url = "git+https://github.com/david-vanderson/SDL#5bbcf665494fe14acadf50601edf6d5f74ebf96a", - .hash = "sdl-0.4.2+3.4.4-SDL--tB1pQFLgwvCakiiOSoZIuVaz6a6-_gOZC66AGRD", + .url = "git+https://github.com/castholm/SDL#dbcb19aa3583c318556ecae3ecbeb29b102d37c5", + .hash = "sdl-0.6.0+3.4.18-SDL--p-gqAGWpnfTLlhs0dgjDtc4P4_CHeNLGY_SX_Fv", .lazy = true, }, // tree-sitter: Zig bindings + the C runtime, plus grammars (parser.c // we link, highlights.scm we read at build time into the ts_queries module). + // pierrelgol's forks of zig-tree-sitter (here) and tree-sitter-c + // (ts_c), which pin a fork of the tree-sitter core: upstream's + // build.zig files still use APIs Zig 0.17 removed. Their diffs touch + // only the build files and tests. Back to upstream once it moves. .tree_sitter = .{ - .url = "git+https://github.com/tree-sitter/zig-tree-sitter#b55d4977d61330390fc733944da06106ff972660", - .hash = "tree_sitter-0.26.0-8heIf3CaAQB4SSumoGzq3gLSz2rkvXxcKFyskRWqDGBz", + .url = "git+https://github.com/pierrelgol/zig-tree-sitter#2ffef82a12d0305e646ade0e67c693cabc661e05", + .hash = "tree_sitter-0.26.0-8heIf1HEAQCVBaR2wlDhGTw16Ttn1bzERVIwlzNcyO6b", }, .ts_rust = .{ .url = "git+https://github.com/tree-sitter/tree-sitter-rust#77a3747266f4d621d0757825e6b11edcbf991ca5", .hash = "N-V-__8AAHJYbgAipcanYj2A6rnj9wtGzjGzANkrOsqiFAxm", }, .ts_c = .{ - .url = "git+https://github.com/tree-sitter/tree-sitter-c#b780e47fc780ddc8da13afa35a3f4ed5c157823d", - .hash = "tree_sitter_c-0.24.2-y5boS-ptQADHoCoVfjGT_nFtFQ5LbomIkW0fxG3_cmdB", + .url = "git+https://github.com/pierrelgol/tree-sitter-c#67949e4c8dc630c18a522919f2bde2e025a0d025", + .hash = "tree_sitter_c-0.24.2-y5boS-htQACe3pz6jHq4tpS8dB6WIo4-zk_ct7lAjvQT", }, .ts_typst = .{ .url = "git+https://github.com/uben0/tree-sitter-typst#13863ddcbaa7b68ee6221cea2e3143415e64aea4", @@ -185,7 +192,7 @@ .lazy = true, }, }, - .minimum_zig_version = "0.16.0", + .minimum_zig_version = "0.17.0", .paths = .{""}, .fingerprint = 0x41dc46101fc764d8, } diff --git a/build/snap.zig b/build/snap.zig index 7aa06884..35a246ad 100644 --- a/build/snap.zig +++ b/build/snap.zig @@ -44,7 +44,7 @@ pub fn addTty(b: *std.Build, opts: struct { const run_snap = b.addRunArtifact(snap); run_snap.addArtifactArg(opts.exe); run_snap.addPrefixedFileArg("--lspmock=", lspmock.getEmittedBin()); - if (b.args) |args| run_snap.addArgs(args); + run_snap.addPassthruArgs(); // scripts fork children and drive real ptys: never cached, never "up to date" run_snap.has_side_effects = true; const snap_step = b.step("snap", "run the snapshot parity suite (-- --update regenerates goldens)"); @@ -63,7 +63,7 @@ pub fn addTty(b: *std.Build, opts: struct { const run_monkey = b.addRunArtifact(monkey); run_monkey.addArtifactArg(snap); run_monkey.addArtifactArg(opts.exe); - if (b.args) |args| run_monkey.addArgs(args); + run_monkey.addPassthruArgs(); run_monkey.has_side_effects = true; b.step("monkey", "random snapshot scripts hunting panics (-- [steps] [--from=N] [--out=DIR])").dependOn(&run_monkey.step); const monkey_tests = b.addTest(.{ .root_module = monkey.root_module }); @@ -101,7 +101,7 @@ pub fn addWeb(b: *std.Build) void { run_web_snap.step.dependOn(&build_web_snap.step); run_web_snap.step.dependOn(&driver_test.step); run_web_snap.has_side_effects = true; - if (b.args) |args| run_web_snap.addArgs(args); + run_web_snap.addPassthruArgs(); b.step("web-snap", "run browser highlighting, touch, Look, resize and scroll checks").dependOn(&run_web_snap.step); b.step("web-e2e", "run the Chrome-driven DOM end-to-end suite").dependOn(&run_web_snap.step); } diff --git a/docs/typ/building.typ b/docs/typ/building.typ index ae2f3a60..8d7524f9 100644 --- a/docs/typ/building.typ +++ b/docs/typ/building.typ @@ -9,7 +9,7 @@ pieces fit. To install it, see #doc("setup", section: "install"). = Platforms -Zig *0.16.0* (`build.zig.zon` pins `minimum_zig_version`). Dependencies are +Zig *0.17.0* (`build.zig.zon` pins `minimum_zig_version`). Dependencies are fetched and pinned by the manifest; the terminal build needs no system package. The SDL shell builds SDL3 and FreeType from source. PDF support builds MuPDF and is on by default (`-Dmupdf=false` drops it). 9P over QUIC @@ -24,10 +24,11 @@ builds MuPDF and is on by default (`-Dmupdf=false` drops it). 9P over QUIC [`zig build -Dplatform=esp32p4`], [a freestanding riscv32 editor object for the ESP32-P4, with a 384 KiB heap], ) -A bare `zig build` installs into `~/.local`; `--prefix ` installs -elsewhere, except `--prefix zig-out`, which counts as no prefix. With -`-Dplatform` the default prefix is `zig-out`, so always pass `-Dplatform` -while developing. Test, benchmark and run steps build what they need +A bare `zig build` installs into the prefix (`zig-out`, or `--prefix +`) and copies the binaries into `~/.local/bin`; `-Dhome-install=false` +leaves that copy out. Zig 0.17 keeps the prefix from the build script, so +`--prefix` alone no longer stops it. With `-Dplatform` nothing is copied, +so always pass `-Dplatform` while developing. Test, benchmark and run steps build what they need without installing. `pardes --version` prints `pardes ` (from `build.zig.zon`), plus the commit for release builds and the `~/.local` install; `pardes --help` lists every flag. @@ -50,6 +51,7 @@ there; its namespace is in `src/esp32p4_gpio.zig`. [`-Djpx`], [bool, `true`: JPEG 2000, and with it scanned PDFs], [`-Dtree-sitter`], [`disabled`, `zig`, `minimal`, `full`; `full` natively, `zig` for web, `disabled` for esp32p4], [`-Dembed-sources`], [bool, `false`: serve the sources under #file("/src")], + [`-Dhome-install`], [bool: a bare `zig build` also copies the binaries into `~/.local/bin`; default true], [`-Dstamp-commit`], [bool: the git commit in `--version` and crash records; on for release builds and the `~/.local` install], [`-Dtheme-animation`], [bool; on except for esp32p4], [`-Dworkspace-tag`], [bool: draw the workspace tag row; on except for macOS, whose menu bar carries it], diff --git a/docs/typ/install.typ b/docs/typ/install.typ index 9758777d..d97d30f7 100644 --- a/docs/typ/install.typ +++ b/docs/typ/install.typ @@ -18,7 +18,7 @@ #let first = project.mirrors.first() #cmd("git clone " + first.clone + " && cd pardes") -With Zig 0.16 (the build fetches everything else; once that is +With Zig 0.17 (the build fetches everything else; once that is downloaded, a first build from a clean cache took 62 seconds on 16 cores): #cmd("# pardes, in a terminal\nzig build -Dplatform=tty --prefix ~/.local") #cmd("# pardes-gui, its own window\nzig build -Dplatform=gui --prefix ~/.local") diff --git a/mupdf.zig b/mupdf.zig index eeadaa5d..761fef92 100644 --- a/mupdf.zig +++ b/mupdf.zig @@ -293,7 +293,7 @@ fn configureModule(module: *std.Build.Module, dependency: *std.Build.Dependency, } fn flatCSources(b: *std.Build, io: std.Io, dependency: *std.Build.Dependency, relative_dir: []const u8) []const []const u8 { - var dir = std.Io.Dir.cwd().openDir(io, dependency.path(relative_dir).getPath(b), .{ .iterate = true }) catch + var dir = dependency.builder.root.root_dir.handle.openDir(io, relative_dir, .{ .iterate = true }) catch std.debug.panic("MuPDF source directory is missing: {s}", .{relative_dir}); defer dir.close(io); @@ -317,9 +317,9 @@ fn flatCSources(b: *std.Build, io: std.Io, dependency: *std.Build.Dependency, re } fn readMakelists(b: *std.Build, io: std.Io, dependency: *std.Build.Dependency) []const u8 { - return std.Io.Dir.cwd().readFileAlloc( + return dependency.builder.root.root_dir.handle.readFileAlloc( io, - dependency.path("Makelists").getPath(b), + "Makelists", b.allocator, .limited(1024 * 1024), ) catch @panic("read MuPDF Makelists"); diff --git a/src/9p_io.zig b/src/9p_io.zig index 18c51ece..27d79b93 100644 --- a/src/9p_io.zig +++ b/src/9p_io.zig @@ -1553,7 +1553,7 @@ test "a socket path too long for sun_path is no listener, and leaves the editor' } else { _ = unsetenv("XDG_RUNTIME_DIR"); }; - const long = "n" ** 120; + const long = &@as([120]u8, @splat('n')); try testing.expect(listen(testing.io, testing.allocator, p, long, "", null, null) == null); try testing.expect(pardes.turn.answer_held == null); try testing.expect(pardes.turn.wake_parked == null); @@ -1567,7 +1567,7 @@ test "a listening editor posts itself into the 9P registry and unposts on stop" _ = try std.fmt.bufPrintSentinel(&directory, "/tmp/pardes-post-XXXXXX", .{}, 0); if (mkdtemp(&directory) == null) return error.TempDirectoryFailed; const dir = std.mem.span(@as([*:0]const u8, &directory)); - const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeZ(u8, std.mem.span(v)) else null; + const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeSentinel(u8, std.mem.span(v), 0) else null; defer { if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); @@ -1628,7 +1628,7 @@ test "Unix TCP and QUIC share one listener through reads writes reconnects and r if (mkdtemp(&directory) == null) return error.TempDirectoryFailed; // What the listener posted and left inside goes with it. defer std.Io.Dir.cwd().deleteTree(testing.io, std.mem.sliceTo(&directory, 0)) catch |err| std.debug.print("deleteTree: {s}\n", .{@errorName(err)}); - const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeZ(u8, std.mem.span(v)) else null; + const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeSentinel(u8, std.mem.span(v), 0) else null; defer { if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); @@ -1749,7 +1749,7 @@ test "a change waits while the editor is out mid-step, a read does not, and the if (mkdtemp(&directory) == null) return error.TempDirectoryFailed; // What the listener posted and left inside goes with it. defer std.Io.Dir.cwd().deleteTree(testing.io, std.mem.sliceTo(&directory, 0)) catch |err| std.debug.print("deleteTree: {s}\n", .{@errorName(err)}); - const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeZ(u8, std.mem.span(v)) else null; + const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeSentinel(u8, std.mem.span(v), 0) else null; defer { if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); @@ -1824,7 +1824,7 @@ test "a held read is answered when the log has news, and a flushed one spends no if (mkdtemp(&directory) == null) return error.TempDirectoryFailed; // What the listener posted and left inside goes with it. defer std.Io.Dir.cwd().deleteTree(testing.io, std.mem.sliceTo(&directory, 0)) catch |err| std.debug.print("deleteTree: {s}\n", .{@errorName(err)}); - const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeZ(u8, std.mem.span(v)) else null; + const old_runtime = if (libc.getenv("XDG_RUNTIME_DIR")) |v| try gpa.dupeSentinel(u8, std.mem.span(v), 0) else null; defer { if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); @@ -2051,7 +2051,7 @@ test "a pane's PARDES_MOUNT is the session's directory under 9ns --mntgen, and u const names = [_][*:0]const u8{ "NINE_MNTGEN", "NINE_MOUNT", "PARDES_MOUNT" }; var saved: [names.len]?[:0]u8 = @splat(null); for (names, &saved) |name, *value| { - if (libc.getenv(name)) |old| value.* = try testing.allocator.dupeZ(u8, std.mem.span(old)); + if (libc.getenv(name)) |old| value.* = try testing.allocator.dupeSentinel(u8, std.mem.span(old), 0); } defer for (names, saved) |name, value| { if (value) |old| { @@ -2073,7 +2073,7 @@ test "9P shell environment states being inside pardes apart from how to reach it const names = [_][*:0]const u8{ "XDG_RUNTIME_DIR", "HOME", "PARDES_PID", "PARDES_9P", "PARDES_PANE" }; var saved: [names.len]?[:0]u8 = @splat(null); for (names, &saved) |name, *value| { - if (libc.getenv(name)) |old| value.* = try testing.allocator.dupeZ(u8, std.mem.span(old)); + if (libc.getenv(name)) |old| value.* = try testing.allocator.dupeSentinel(u8, std.mem.span(old), 0); } defer for (names, saved) |name, value| { if (value) |old| { diff --git a/src/CHANGELOG.md b/src/CHANGELOG.md index 4fe0760c..c6c2eadd 100644 --- a/src/CHANGELOG.md +++ b/src/CHANGELOG.md @@ -17,6 +17,14 @@ - A diff paged into a session (`git show | pardes -`) opens its `+Diff` alone; the `+Pager` no longer stays folded under it. +- Built with Zig 0.17. A bare `zig build` still puts the binaries in + `~/.local/bin`, now as a copy beside the prefix, which `--prefix` no + longer prevents; `-Dhome-install=false` does. ghostty, libvaxis, uucode + and SDL (3.4.18) move to their Zig 0.17 versions, and tree-sitter to a + 0.28 development snapshot. +- A terminal's edited copy goes when the program clears the screen, + instead of staying where its rows were. + - n/N follow Helix search registers, counts, wrapping and select mode. `*` sets a search from the selection; `Alt-*` omits word boundaries. The location ring remains as LookNext/LookPrev (Space j n/p). diff --git a/src/Layer.zig b/src/Layer.zig index 7a5fcf7e..ba45bc88 100644 --- a/src/Layer.zig +++ b/src/Layer.zig @@ -51,7 +51,7 @@ cols: u16 = 0, rows: u16 = 0, /// A body's compact rows, from its top. context_rows: u16 = 0, -context_separators: std.bit_set.StaticBitSet(limits.wrap_rows) = .initEmpty(), +context_separators: std.bit_set.StaticBitSet(limits.wrap_rows) = .empty, /// `rows * cols`, row after row. cells: []pardes.Cell = &.{}, cursor: ?struct { x: u16, y: u16, bar: bool = false } = null, @@ -72,7 +72,7 @@ pub fn capacity(layer: *const Layer, metrics: Metrics) u16 { } pub fn setContextLines(layer: *Layer, lines: []const i32) void { - layer.context_separators = .initEmpty(); + layer.context_separators = .empty; const count = @min(lines.len, layer.context_rows); var index: usize = 0; while (index + 1 < count and index < limits.wrap_rows) : (index += 1) { diff --git a/src/Messages.zig b/src/Messages.zig index 2314939f..b67a8402 100644 --- a/src/Messages.zig +++ b/src/Messages.zig @@ -519,20 +519,20 @@ pub fn errorWords(err: anyerror, buf: []u8) []const u8 { /// A ZON parse failure in a person's words: the first error the parser /// found and its line, `line 3: expected ','`. pub fn zonWhy(diag: *const std.zon.parse.Diagnostics, buf: []u8) []const u8 { - var errors = diag.iterateErrors(); - const first = errors.next() orelse return "not valid ZON"; - const loc = first.getLocation(diag); - return std.fmt.bufPrint(buf, "line {d}: {f}", .{ loc.line + 1, first.fmtMessage(diag) }) catch "not valid ZON"; + if (diag.errors.len == 0) return "not valid ZON"; + const first = diag.errors[0]; + return std.fmt.bufPrint(buf, "line {d}: {s}", .{ first.loc.line + 1, first.msg }) catch "not valid ZON"; } test "errors are said in words, and a ZON failure names its line" { var buf: [128]u8 = undefined; try std.testing.expectEqualStrings("file not found", errorWords(error.FileNotFound, &buf)); try std.testing.expectEqualStrings("too large", errorWords(error.FileTooLarge, &buf)); - var diag: std.zon.parse.Diagnostics = .{}; - defer diag.deinit(std.testing.allocator); + var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); + defer arena.deinit(); + var diag: std.zon.parse.Diagnostics = undefined; const T = struct { a: u8 }; - try std.testing.expectError(error.ParseZon, std.zon.parse.fromSliceAlloc(T, std.testing.allocator, ".{\n .a = 1,\n .b = 2 }", &diag, .{})); + try std.testing.expectError(error.ParseZon, std.zon.parse.fromSlice(T, .{ .gpa = std.testing.allocator, .arena = arena.allocator(), .source = ".{\n .a = 1,\n .b = 2 }", .diagnostics = &diag })); try std.testing.expect(std.mem.startsWith(u8, zonWhy(&diag, &buf), "line 3: ")); } diff --git a/src/TerminalImages.zig b/src/TerminalImages.zig index bc1b2ea8..4140813f 100644 --- a/src/TerminalImages.zig +++ b/src/TerminalImages.zig @@ -38,6 +38,72 @@ const GlyphKey = struct { id: u32, generation: u64, src: PixelRect, cols: u16 }; /// left corner (above or left of it is negative). const Rect = struct { x: i64, y: i64, w: i64, h: i64 }; +/// The runs of Kitty placeholder cells between two pins, one row's run at a +/// time, as ghostty's own `kitty.graphics.unicode.placementIterator` gave +/// them before it moved into its RenderState, which pardes does not keep. +const PlaceholderRuns = struct { + rows: ghostty_vt.PageList.RowIterator, + row: ?ghostty_vt.Pin, + /// Where the scan of `row` resumes: a run ended by an incompatible + /// placeholder starts the next one there. + x: usize = 0, + + fn init(top: ghostty_vt.Pin, bottom: ghostty_vt.Pin) PlaceholderRuns { + var rows = top.rowIterator(.right_down, bottom); + const row = rows.next(); + return .{ .rows = rows, .row = row }; + } + + fn next(self: *PlaceholderRuns) ?gfx.unicode.Placement { + while (self.row) |row| { + if (!row.rowAndCell().row.kitty_virtual_placeholder) { + self.advance(); + continue; + } + var run: ?gfx.unicode.IncompletePlacement = null; + var start: usize = 0; + const cells = row.cells(.all); + while (self.x < cells.len) : (self.x += 1) { + const cell = &cells[self.x]; + if (cell.codepoint() != gfx.unicode.placeholder) { + if (run) |done| return complete(row, start, done); + continue; + } + const style = row.style(cell); + const curr: gfx.unicode.IncompletePlacement = .init(&style, row.grapheme(cell) orelse &.{}); + if (run) |*prev| { + if (!prev.append(&curr)) return complete(row, start, prev.*); + } else { + run = curr.start(); + start = self.x; + } + } + self.advance(); + if (run) |done| return complete(row, start, done); + } + return null; + } + + fn advance(self: *PlaceholderRuns) void { + self.row = self.rows.next(); + self.x = 0; + } + + fn complete(row: ghostty_vt.Pin, x: usize, run: gfx.unicode.IncompletePlacement) gfx.unicode.Placement { + var pin = row; + pin.x = @intCast(x); + return .{ + .pin = pin, + .image_id = run.imageId(), + .placement_id = run.placement_id orelse 0, + .col = run.col.?, + .row = run.row.?, + .width = run.width, + .height = 1, + }; + } +}; + pub fn deinit(ti: *TerminalImages, gpa: std.mem.Allocator) void { var conv = ti.converted.valueIterator(); while (conv.next()) |c| gpa.free(c.rgba); @@ -91,7 +157,7 @@ pub fn draw( // Virtual placements: shown through the placeholder cells, one run of // them on a row at a time. The placeholders themselves are not text. - var runs = gfx.unicode.placementIterator(top, bottom); + var runs: PlaceholderRuns = .init(top, bottom); while (runs.next()) |run| { const at = pages.pointFromPin(.viewport, run.pin) orelse continue; const v = at.viewport.y; @@ -104,7 +170,8 @@ pub fn draw( cell.len = 1; } const img = storage.imageById(run.image_id) orelse continue; - const rp = run.renderPlacement(storage, &img, cw, ch) catch continue; + const target = storage.placeholderTarget(run.image_id, run.placement_id) orelse continue; + const rp = run.renderPlacement(&target.placement, &img, cw, ch) catch continue; if (rp.dest_width == 0 or rp.dest_height == 0 or rp.source_width == 0 or rp.source_height == 0) continue; ctx.place(img, .{ .x = rp.source_x, .y = rp.source_y, .w = rp.source_width, .h = rp.source_height }, .{ .x = @as(i64, at.viewport.x) * cw + rp.offset_x, @@ -122,7 +189,8 @@ pub fn draw( const placed = entry.value_ptr.*; const pin = switch (placed.location) { .pin => |pin| pin, - .virtual => continue, + // ponytail: a placement relative to another (P=/Q=, new in ghostty) is not drawn; resolve its chain if a program needs it. + .virtual, .relative => continue, }; const img = storage.images.get(entry.key_ptr.image_id) orelse continue; const at = pages.pointFromPin(.screen, pin.*) orelse continue; @@ -221,7 +289,8 @@ const Ctx = struct { fn pixelsOf(ti: *TerminalImages, gpa: std.mem.Allocator, img: gfx.Image) ?[]const u8 { const n = @as(usize, img.width) * img.height; if (n == 0) return null; - if (img.format == .rgba) return if (img.data.len == n * 4) img.data else null; + const data = img.data.bytes() orelse return null; + if (img.format == .rgba) return if (data.len == n * 4) data else null; const gop = ti.converted.getOrPut(gpa, img.id) catch return null; if (gop.found_existing) { if (gop.value_ptr.generation == img.generation) { @@ -247,10 +316,11 @@ fn convert(gpa: std.mem.Allocator, img: gfx.Image, n: usize) ?[]u8 { .gray => 1, else => return null, }; - if (img.data.len != n * depth) return null; + const data = img.data.bytes() orelse return null; + if (data.len != n * depth) return null; const out = gpa.alloc(u8, n * 4) catch return null; for (0..n) |i| { - const px = img.data[i * depth ..][0..depth]; + const px = data[i * depth ..][0..depth]; out[i * 4 ..][0..4].* = switch (depth) { 3 => .{ px[0], px[1], px[2], 255 }, 2 => .{ px[0], px[0], px[0], px[1] }, diff --git a/src/builtins.zig b/src/builtins.zig index f73c5982..d9cef8c9 100644 --- a/src/builtins.zig +++ b/src/builtins.zig @@ -98,11 +98,11 @@ fn isEnabled(comptime T: type) bool { /// word that is one here, or in no build. pub fn absentWhy(name: []const u8) ?[]const u8 { @setEvalBranchQuota(20000); - inline for (@typeInfo(@This()).@"struct".decls) |d| { - if (comptime @TypeOf(@field(@This(), d.name)) == type) { - const T = @field(@This(), d.name); + inline for (@typeInfo(@This()).@"struct".decl_names) |d| { + if (comptime @TypeOf(@field(@This(), d)) == type) { + const T = @field(@This(), d); if (comptime @typeInfo(T) == .@"struct" and @hasDecl(T, "run") and !isEnabled(T)) { - if (std.mem.eql(u8, d.name, name)) return if (@hasDecl(T, "gui_only")) "GUI-only" else "not in this build"; + if (std.mem.eql(u8, d, name)) return if (@hasDecl(T, "gui_only")) "GUI-only" else "not in this build"; } } } @@ -112,12 +112,12 @@ pub fn absentWhy(name: []const u8) ?[]const u8 { fn manualBuiltinCount() comptime_int { comptime { var count = 0; - for (@typeInfo(@This()).@"struct".decls) |d| { - if (@TypeOf(@field(@This(), d.name)) != type) continue; - const T = @field(@This(), d.name); + for (@typeInfo(@This()).@"struct".decl_names) |d| { + if (@TypeOf(@field(@This(), d)) != type) continue; + const T = @field(@This(), d); if (@typeInfo(T) != .@"struct" or !@hasDecl(T, "run") or !isEnabled(T)) continue; if (@TypeOf(T.run) != fn (Ctx) void) - @compileError(d.name ++ ".run must have signature fn (Ctx) void"); + @compileError(d ++ ".run must have signature fn (Ctx) void"); count += 1; } return count; @@ -129,9 +129,9 @@ fn manualBuiltinList() [manualBuiltinCount()]type { @setEvalBranchQuota(4000); var list: [manualBuiltinCount()]type = undefined; var count = 0; - for (@typeInfo(@This()).@"struct".decls) |d| { - if (@TypeOf(@field(@This(), d.name)) != type) continue; - const T = @field(@This(), d.name); + for (@typeInfo(@This()).@"struct".decl_names) |d| { + if (@TypeOf(@field(@This(), d)) != type) continue; + const T = @field(@This(), d); if (@typeInfo(T) != .@"struct" or !@hasDecl(T, "run") or !isEnabled(T)) continue; list[count] = T; count += 1; @@ -317,7 +317,7 @@ test "Exit over a ctl write or a click names every pane it asks about, a second first.file.?.saved_hash = 0; // edited: what it had is not what it has p.newScratchBelow(p.active); const second = p.panes[p.active].?; - panes.File.setContent(p, &second.file.?, try std.testing.allocator.dupe(u8, "x" ** 120)); + panes.File.setContent(p, &second.file.?, try std.testing.allocator.dupe(u8, &@as([120]u8, @splat('x')))); try std.testing.expect(second.file.?.revision != second.file.?.saved_revision); // A ctl write, or a click through exec (ctl.captured). if (clicked) p.fs.capturing = true else p.fs.no_prompt = true; @@ -341,7 +341,7 @@ test "Exit over a ctl write or a click names every pane it asks about, a second try std.testing.expect(noticed); // Edited since: only that one is named again (acme's rule). p.fs.failure_len = 0; - panes.File.setContent(p, &second.file.?, try std.testing.allocator.dupe(u8, "y" ** 120)); + panes.File.setContent(p, &second.file.?, try std.testing.allocator.dupe(u8, &@as([120]u8, @splat('y')))); try std.testing.expect(p.executeBuiltinLine(p.active, "Exit")); try std.testing.expect(!p.quit); const again = p.fs.failure[0..p.fs.failure_len]; diff --git a/src/colors.zig b/src/colors.zig index 379d1301..dd0d8938 100644 --- a/src/colors.zig +++ b/src/colors.zig @@ -205,7 +205,7 @@ const name_tint_chroma: f32 = 0.06; /// ink where that hue holds, the name reads on its ground (nameTintFloor) /// and it stands off the ink. Null when no colour of the theme will do. fn derivedNameTint(th: *const Theme, ink: [3]u8, ground: [3]u8) ?[3]u8 { - const pal = th.palette orelse [_][3]u8{th.kw} ** 16; + const pal = th.palette orelse @as([16][3]u8, @splat(th.kw)); const sources = [_][3]u8{ th.kw, pal[4], pal[5], pal[6], pal[3], pal[2], pal[1], th.num, th.str, th.box } ++ [_][3]u8{ th.column_box orelse th.num, th.box_dirty orelse th.num, th.sel_bg, th.search_bg orelse th.num } ++ [_][3]u8{ th.diagnostic_info orelse th.num, th.diagnostic_hint orelse th.num, th.diagnostic_warning orelse th.num } ++ @@ -468,28 +468,29 @@ const curated = struct { pub const dark = @import("themes/dark.zig"); }; -fn fold(comptime C: type) [@typeInfo(C).@"struct".decls.len]Theme { +fn fold(comptime C: type) [@typeInfo(C).@"struct".decl_names.len]Theme { comptime { @setEvalBranchQuota(400000); - var out: [@typeInfo(C).@"struct".decls.len]Theme = undefined; - for (@typeInfo(C).@"struct".decls, 0..) |d, i| { + var out: [@typeInfo(C).@"struct".decl_names.len]Theme = undefined; + const theme_info = @typeInfo(Theme).@"struct"; + for (@typeInfo(C).@"struct".decl_names, 0..) |d, i| { var t: Theme = undefined; - const source = @field(C, d.name).theme; - for (@typeInfo(Theme).@"struct".fields) |f| { - @field(t, f.name) = if (f.type == ?Decor and @hasField(@TypeOf(source), f.name)) decor: { + const source = @field(C, d).theme; + for (theme_info.field_names, theme_info.field_types, theme_info.field_attrs) |name, T, attrs| { + @field(t, name) = if (T == ?Decor and @hasField(@TypeOf(source), name)) decor: { // A nested literal is folded field by field too. var decor: Decor = .{}; - const given = @field(source, f.name); - for (@typeInfo(Decor).@"struct".fields) |g| if (@hasField(@TypeOf(given), g.name)) { - @field(decor, g.name) = @field(given, g.name); + const given = @field(source, name); + for (@typeInfo(Decor).@"struct".field_names) |g| if (@hasField(@TypeOf(given), g)) { + @field(decor, g) = @field(given, g); }; break :decor decor; - } else if (@hasField(@TypeOf(source), f.name)) - @field(source, f.name) - else if (f.default_value_ptr) |ptr| - @as(*const f.type, @ptrCast(@alignCast(ptr))).* + } else if (@hasField(@TypeOf(source), name)) + @field(source, name) + else if (attrs.defaultValue(T)) |value| + value else - @compileError("theme " ++ d.name ++ " is missing " ++ f.name); + @compileError("theme " ++ d ++ " is missing " ++ name); } out[i] = t; } @@ -501,7 +502,7 @@ fn fold(comptime C: type) [@typeInfo(C).@"struct".decls.len]Theme { /// takes: a faithful port (`dracula`) keeps the plain name, and helix's /// looser copy of it is still there as `dracula_helix` (the zed ones already /// carry `_zed`). -fn foldImports(comptime ours: anytype, comptime C: type) [@typeInfo(C).@"struct".decls.len]Theme { +fn foldImports(comptime ours: anytype, comptime C: type) [@typeInfo(C).@"struct".decl_names.len]Theme { comptime { @setEvalBranchQuota(400000); var out = fold(C); @@ -514,7 +515,7 @@ fn foldImports(comptime ours: anytype, comptime C: type) [@typeInfo(C).@"struct" pub const themes = fold(curated) ++ foldImports(fold(curated), @import("generated_themes")); /// The first `native_count` of `themes` are pardes's own; the rest imported. -pub const native_count = @typeInfo(curated).@"struct".decls.len; +pub const native_count = @typeInfo(curated).@"struct".decl_names.len; /// The picker's sections (`Themes`), in ring order, each running from /// the theme it names to the next section's; `NextColor` and `Theme` still @@ -677,8 +678,8 @@ pub const ChromeTheme = struct { pub fn interpolate(from: ChromeTheme, to: ChromeTheme, step: u16, steps: u16) ChromeTheme { var out: ChromeTheme = undefined; const t = @as(f32, @floatFromInt(@min(step, steps))) / @as(f32, @floatFromInt(@max(1, steps))); - inline for (@typeInfo(ChromeTheme).@"struct".fields) |field| - @field(out, field.name) = if (step == 0) @field(from, field.name) else if (step >= steps) @field(to, field.name) else mixOklab(@field(from, field.name), @field(to, field.name), t); + inline for (@typeInfo(ChromeTheme).@"struct".field_names) |name| + @field(out, name) = if (step == 0) @field(from, name) else if (step >= steps) @field(to, name) else mixOklab(@field(from, name), @field(to, name), t); return out; } }; @@ -801,14 +802,16 @@ pub fn loadThemeFile(p: *Pardes, generation: u32, bytes: []const u8, animate: bo themeFileError(p, request, error.ThemeFileTooLarge); return false; } - const source = p.gpa.dupeZ(u8, bytes) catch |err| { + const source = p.gpa.dupeSentinel(u8, bytes, 0) catch |err| { themeFileError(p, request, err); return false; }; defer p.gpa.free(source); - var diagnostics: std.zon.parse.Diagnostics = .{}; - defer diagnostics.deinit(p.gpa); - const parsed = std.zon.parse.fromSliceAlloc(Theme, p.gpa, source, &diagnostics, .{}) catch |err| { + // The parse's results live in `arena`; the theme kept owns only its name. + var arena: std.heap.ArenaAllocator = .init(p.gpa); + defer arena.deinit(); + var diagnostics: std.zon.parse.Diagnostics = undefined; + var parsed = std.zon.parse.fromSlice(Theme, .{ .gpa = p.gpa, .arena = arena.allocator(), .source = source, .diagnostics = &diagnostics }) catch |err| { if (err != error.ParseZon) themeFileError(p, request, err) else { // Which line, and what the parser wanted there. var why: [200]u8 = undefined; @@ -817,10 +820,13 @@ pub fn loadThemeFile(p: *Pardes, generation: u32, bytes: []const u8, animate: bo return false; }; if (parsed.name.len == 0 or parsed.name.len > 255) { - std.zon.parse.free(p.gpa, parsed); themeFileError(p, request, error.InvalidThemeName); return false; } + parsed.name = p.gpa.dupe(u8, parsed.name) catch |err| { + themeFileError(p, request, err); + return false; + }; const target_chrome = ChromeTheme.fromTheme(&parsed); if (animate and p.animate_theme_changes) @@ -829,7 +835,7 @@ pub fn loadThemeFile(p: *Pardes, generation: u32, bytes: []const u8, animate: bo p.chrome_animation.snap(target_chrome); const old = p.custom_theme; p.custom_theme = parsed; - if (old) |theme_value| std.zon.parse.free(p.gpa, theme_value); + if (old) |theme_value| p.gpa.free(theme_value.name); var settled: [limits.host_path_cap]u8 = undefined; @memcpy(settled[0..request.path.len], request.path); _ = p.theme_file_loaded.set(settled[0..request.path.len]); @@ -847,7 +853,7 @@ pub fn carryTheme(p: *Pardes, old: *const Pardes, index: bool) !void { const t = old.custom_theme orelse return; var copy = t; copy.name = try p.gpa.dupe(u8, t.name); - if (p.custom_theme) |mine| std.zon.parse.free(p.gpa, mine); + if (p.custom_theme) |mine| p.gpa.free(mine.name); p.custom_theme = copy; _ = p.theme_file_loaded.set(old.theme_file_loaded.get()); } @@ -931,7 +937,7 @@ fn likeness(theme_name: []const u8, name: []const u8) usize { in_order += 1; pos = at + 1; } - const inside: usize = if (name.len > 0 and std.ascii.indexOfIgnoreCase(theme_name, name) != null) name.len * 4 else 0; + const inside: usize = if (name.len > 0 and std.ascii.findIgnoreCase(theme_name, name) != null) name.len * 4 else 0; return prefix * 3 + in_order + inside; } @@ -952,7 +958,7 @@ pub fn setThemeIndex(p: *Pardes, index: usize) void { p.chrome_animation.snap(target_chrome); if (p.custom_theme) |theme_value| { p.custom_theme = null; - std.zon.parse.free(p.gpa, theme_value); + p.gpa.free(theme_value.name); } p.theme_file_loaded.clear(); if (p.theme_file_path.get().len > 0) { @@ -1264,10 +1270,12 @@ test "legacy ThemeFile documents inherit Pardes UI roles without new fields" { rest = rest[at + std.mem.indexOfScalar(u8, rest[at..], '}').? + 1 ..]; rest = std.mem.trimStart(u8, rest, " ,"); } - const source = try std.testing.allocator.dupeZ(u8, legacy.items); + const source = try std.testing.allocator.dupeSentinel(u8, legacy.items, 0); defer std.testing.allocator.free(source); - const parsed = try std.zon.parse.fromSliceAlloc(Theme, std.testing.allocator, source, null, .{}); - defer std.zon.parse.free(std.testing.allocator, parsed); + var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); + defer arena.deinit(); + var diagnostics: std.zon.parse.Diagnostics = undefined; + const parsed = try std.zon.parse.fromSlice(Theme, .{ .gpa = std.testing.allocator, .arena = arena.allocator(), .source = source, .diagnostics = &diagnostics }); try std.testing.expectEqual(@as(?[3]u8, null), parsed.tag_active_bg); try std.testing.expectEqual(@as(?[3]u8, null), parsed.tag_name_fg); try std.testing.expectEqual(@as(?[3]u8, null), parsed.tag_active_name_fg); diff --git a/src/config.zig b/src/config.zig index 50009660..8166954d 100644 --- a/src/config.zig +++ b/src/config.zig @@ -2170,10 +2170,12 @@ pub const User = struct { defer gpa.free(one_path); const bytes = try std.Io.Dir.cwd().readFileAlloc(io, one_path, gpa, .limited(4096)); defer gpa.free(bytes); - const source = try gpa.dupeZ(u8, bytes); + const source = try gpa.dupeSentinel(u8, bytes, 0); defer gpa.free(source); - const parsed = try std.zon.parse.fromSliceAlloc(Sample, gpa, source, null, .{}); - defer std.zon.parse.free(gpa, parsed); + var arena: std.heap.ArenaAllocator = .init(gpa); + defer arena.deinit(); + var diagnostics: std.zon.parse.Diagnostics = undefined; + const parsed = try std.zon.parse.fromSlice(Sample, .{ .gpa = gpa, .arena = arena.allocator(), .source = source, .diagnostics = &diagnostics }); try std.testing.expectEqualStrings("one", parsed.name); try std.testing.expectEqual([3]u8{ 1, 2, 3 }, parsed.rgb); } @@ -2239,25 +2241,26 @@ const doc_typ = struct { fn hasChordField(T: type) bool { if (@typeInfo(T) != .array or @typeInfo(@typeInfo(T).array.child) != .@"struct") return false; - for (@typeInfo(@typeInfo(T).array.child).@"struct".fields) |f| if (f.type == Chord) return true; + for (@typeInfo(@typeInfo(T).array.child).@"struct".field_types) |F| if (F == Chord) return true; return false; } /// Bound somewhere in this file: a binding, a prefix or a key list. fn bound(c: Chord) bool { const config = @import("config.zig"); - inline for (@typeInfo(config).@"struct".decls) |d| { - const v = @field(config, d.name); + inline for (@typeInfo(config).@"struct".decl_names) |d| { + const v = @field(config, d); const T = @TypeOf(v); if (T == []const Chord) { for (v) |b| if (same(b, c)) return true; } else if (T == comptime_int or T == u21 or T == u8) { // a prefix or a bare key; only the tty toggle carries Ctrl - const plain = !c.alt and !c.shift and (!c.ctrl or comptime std.mem.indexOf(u8, d.name, "toggle") != null); + const plain = !c.alt and !c.shift and (!c.ctrl or comptime std.mem.indexOf(u8, d, "toggle") != null); if (plain and v >= 0x20 and v < 0x7f and v == c.cp) return true; } else if (comptime hasChordField(T)) { - for (v) |e| inline for (@typeInfo(@TypeOf(e)).@"struct".fields) |f| - if (f.type == Chord and same(@field(e, f.name), c)) return true; + const info = @typeInfo(@TypeOf(v[0])).@"struct"; + for (v) |e| inline for (info.field_names, info.field_types) |name, F| + if (F == Chord and same(@field(e, name), c)) return true; } } return false; diff --git a/src/detached/client.zig b/src/detached/client.zig index 69ab2fdd..cf1842d9 100644 --- a/src/detached/client.zig +++ b/src/detached/client.zig @@ -1182,12 +1182,12 @@ test "detached session: a frontend is never asked to fork, write, or watch" { // One: the union a frontend decodes into has no other variant. Named // rather than counted, so re-adding `spawn` fails with the name in it. - const fields = @typeInfo(wire.ServerMsg).@"union".fields; - inline for (fields) |f| { + const fields = @typeInfo(wire.ServerMsg).@"union".field_names; + inline for (fields) |name| { for (allowed) |ok| { - if (std.mem.eql(u8, f.name, ok)) break; + if (std.mem.eql(u8, name, ok)) break; } else { - std.debug.print("ServerMsg.{s} is not an effect a frontend may perform\n", .{f.name}); + std.debug.print("ServerMsg.{s} is not an effect a frontend may perform\n", .{name}); return error.MachineLocalEffectOnTheWire; } } diff --git a/src/detached/wire.zig b/src/detached/wire.zig index c9d9b578..f1b7a1b3 100644 --- a/src/detached/wire.zig +++ b/src/detached/wire.zig @@ -20,9 +20,10 @@ const chrome_layout: u64 = 0x5e758b47fb0cf8da; comptime { @setEvalBranchQuota(10_000); var hash = std.hash.Fnv1a_64.init(); - for (@typeInfo(pardes.Chrome).@"struct".fields) |field| { - hash.update(field.name); - hash.update(@typeName(field.type)); + const info = @typeInfo(pardes.Chrome).@"struct"; + for (info.field_names, info.field_types) |name, T| { + hash.update(name); + hash.update(@typeName(T)); } if (hash.final() != chrome_layout) @compileError(std.fmt.comptimePrint("pardes.Chrome changed (layout 0x{x}): bump wire.version, then chrome_layout", .{hash.final()})); } @@ -285,21 +286,22 @@ pub const Frame = struct { region.clip > region.rect.w or region.thumb_y > region.rect.h or region.thumb_h > region.rect.h - region.thumb_y) return error.BadValue; regions.appendAssumeCapacity(region); } - inline for (@typeInfo(pardes.Chrome).@"struct".fields) |field| { - const rgb = &@field(chrome, field.name); - if (field.type == ?[3]u8) { + const info = @typeInfo(pardes.Chrome).@"struct"; + inline for (info.field_names, info.field_types) |name, T| { + const rgb = &@field(chrome, name); + if (T == ?[3]u8) { if (try r.getBool()) { rgb.* = .{ 0, 0, 0 }; for (&rgb.*.?) |*channel| channel.* = try r.getByte(); } - } else if (field.type == u8) { + } else if (T == u8) { rgb.* = try r.getByte(); - } else if (field.type == u32) { + } else if (T == u32) { rgb.* = try r.getU32(); - } else if (field.type == f32) { + } else if (T == f32) { rgb.* = @bitCast(try r.getU32()); if (!std.math.isFinite(rgb.*)) return error.BadValue; - } else if (field.type == [8]f32) { + } else if (T == [8]f32) { for (rgb) |*value| { value.* = @bitCast(try r.getU32()); if (!std.math.isFinite(value.*)) return error.BadValue; @@ -781,7 +783,7 @@ pub fn layersBound(layers: []const pardes.Layer, tags: []const pardes.Layer, reg const region_len = 1 + 1 + 4 + 8 + 1 + 2 + 2 + 1 + 2 + 4 + 4; /// The palette: each colour three bytes, a theme's optional ones a flag /// before them. -const chrome_len = @typeInfo(pardes.Chrome).@"struct".fields.len * 4 + 8 * 4; +const chrome_len = @typeInfo(pardes.Chrome).@"struct".field_names.len * 4 + 8 * 4; pub fn encodeFrameLayers(out: []u8, cols: u16, rows: u16, cursor: ?Cursor, pointer_shape: pardes.Surface.PointerShape, cells: []const pardes.Cell, prev: []const pardes.Cell, layers: []const pardes.Layer, tags: []const pardes.Layer, regions: []const pardes.Region, chrome: pardes.Chrome) Error![]const u8 { // The protocol's ceiling, enforced by the ENCODER too, and BEFORE the @@ -865,18 +867,19 @@ pub fn encodeFrameLayers(out: []u8, cols: u16, rows: u16, cursor: ?Cursor, point try w.putU32(@bitCast(region.lift)); try w.putU32(@bitCast(region.line)); } - inline for (@typeInfo(pardes.Chrome).@"struct".fields) |field| { - const rgb = @field(chrome, field.name); - if (field.type == ?[3]u8) { + const info = @typeInfo(pardes.Chrome).@"struct"; + inline for (info.field_names, info.field_types) |name, T| { + const rgb = @field(chrome, name); + if (T == ?[3]u8) { try w.putBool(rgb != null); if (rgb) |value| for (value) |channel| try w.putByte(channel); - } else if (field.type == u8) { + } else if (T == u8) { try w.putByte(rgb); - } else if (field.type == u32) { + } else if (T == u32) { try w.putU32(rgb); - } else if (field.type == f32) { + } else if (T == f32) { try w.putU32(@bitCast(rgb)); - } else if (field.type == [8]f32) { + } else if (T == [8]f32) { for (rgb) |value| try w.putU32(@bitCast(value)); } else for (rgb) |channel| try w.putByte(channel); } diff --git a/src/draw.zig b/src/draw.zig index b35d55e8..e6420ed4 100644 --- a/src/draw.zig +++ b/src/draw.zig @@ -1598,7 +1598,7 @@ test "a tag of three rows has its grip on the first, and band under it that stil test "place puts every part where the painters and layers put it, in every layout" { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 100, .rows = 30 }); defer p.deinit(); - _ = try p.setTestFile("placed\n" ** 80); + _ = try p.setTestFile(std.mem.asBytes(&@as([80][7]u8, @splat("placed\n".*)))); p.newColumn(0); p.newScratchIn(0, p.activeColumn().?); p.newScratchBelow(0); @@ -1713,7 +1713,7 @@ test "a notice wider than its band, as a shader's compile error is, draws its en _ = try p.setTestFile("text\n"); var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - const long = "Shader /home/someone/shaders/test_shadertoy_invalid.glsl: error: Linking fragment stage: No function definition (body) found: " ** 3; + const long = std.mem.asBytes(&@as([3][126]u8, @splat("Shader /home/someone/shaders/test_shadertoy_invalid.glsl: error: Linking fragment stage: No function definition (body) found: ".*))); for ([_][]const u8{ "Msg " ++ long, "" }) |line| { // The Msg builtin, and the path a failed compile takes (Post.zig). if (line.len > 0) try std.testing.expect(p.executeBuiltinLine(p.active, line)) else p.setMessage(p.active, long); @@ -1952,7 +1952,7 @@ test "the focused cursor glides a jump, lands a step at once, and is exact at re if (!builtins.capabilities.lift) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .cols = 80, .rows = 30 }); defer p.deinit(); - _ = try p.setTestFile("some text on a line\n" ** 20); + _ = try p.setTestFile(std.mem.asBytes(&@as([20][20]u8, @splat("some text on a line\n".*)))); p.settings.cursor_blink = false; var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); @@ -2078,7 +2078,7 @@ test "a terminal's cursor lands with scrolling output, never gliding against it" defer arena.deinit(); const ms = std.time.ns_per_ms; var now: u64 = std.time.ns_per_s; - p.update(.{ .output = .{ .pane = @intCast(term), .bytes = "line\r\n" ** 20 ++ "$ some command typed" } }); + p.update(.{ .output = .{ .pane = @intCast(term), .bytes = std.mem.asBytes(&@as([20][6]u8, @splat("line\r\n".*))) ++ "$ some command typed" } }); p.advance(now); _ = try p.render(arena.allocator()); // Enter: three lines of output scroll up, the cursor to a new prompt. @@ -2101,10 +2101,10 @@ test "a monkey over notices, prompts, tags and tiny panes at a gui's metrics nev const random = prng.random(); const p = try Pardes.init(std.testing.allocator, .{ .cols = 80, .rows = 24 }); defer p.deinit(); - _ = try p.setTestFile("fuzz\n\u{754c}\u{754c} wide\n" ** 20); + _ = try p.setTestFile(std.mem.asBytes(&@as([20][17]u8, @splat("fuzz\n\u{754c}\u{754c} wide\n".*)))); var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - const words = [_][]const u8{ "error:", "\u{754c}", "x", "Shader /tmp/a.glsl:", "e\u{301}", "\t", "long-" ** 20, " " }; + const words = [_][]const u8{ "error:", "\u{754c}", "x", "Shader /tmp/a.glsl:", "e\u{301}", "\t", std.mem.asBytes(&@as([20][5]u8, @splat("long-".*))), " " }; const commands = [_][]const u8{ "New", "Newcol", "Collapse", "Tagbottom", "TreeContext", "Wrap", "Find", "Del", "FocusTint", "Debug" }; var text: std.ArrayListUnmanaged(u8) = .empty; defer text.deinit(std.testing.allocator); @@ -2143,7 +2143,7 @@ test "a monkey over notices, prompts, tags and tiny panes at a gui's metrics nev // A tag of several rows. 5 => if (p.panes[p.active]) |pane| { text.clearRetainingCapacity(); - for (0..random.uintAtMost(usize, 5)) |_| try text.appendSlice(std.testing.allocator, if (random.boolean()) " Look\nmore \u{754c}" else "\n" ++ "x" ** 150); + for (0..random.uintAtMost(usize, 5)) |_| try text.appendSlice(std.testing.allocator, if (random.boolean()) " Look\nmore \u{754c}" else "\n" ++ &@as([150]u8, @splat('x'))); if (pane.tag.own) |own| std.testing.allocator.free(own); pane.tag.own = try std.testing.allocator.dupe(u8, text.items); }, @@ -2173,7 +2173,7 @@ test "a wide glyph cut by a tag's or a notice's edge is joined into the grid as defer arena.deinit(); const pane = p.panes[p.active].?; // Both alignments, so one of them puts a wide glyph across each edge. - for ([_][]const u8{ "\u{754c}" ** 30, "x" ++ "\u{754c}" ** 30 }) |wide| { + for ([_][]const u8{ std.mem.asBytes(&@as([30][3]u8, @splat("\u{754c}".*))), "x" ++ std.mem.asBytes(&@as([30][3]u8, @splat("\u{754c}".*))) }) |wide| { if (pane.tag.own) |own| std.testing.allocator.free(own); pane.tag.own = try std.testing.allocator.dupe(u8, wide); var message: [200]u8 = undefined; @@ -2201,7 +2201,7 @@ test "a body with context rows is painted once, and the grid's rows are what a p if (comptime !@import("syntax.zig").enabled) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 60, .rows = 16 }); defer p.deinit(); - const pane = try p.setTestFile("pub const Box = struct {\n pub fn run() void {\n" ++ (" // body line\n" ** 30) ++ " }\n};\n"); + const pane = try p.setTestFile("pub const Box = struct {\n pub fn run() void {\n" ++ (std.mem.asBytes(&@as([30][21]u8, @splat(" // body line\n".*)))) ++ " }\n};\n"); p.gpa.free(pane.file.?.path); pane.file.?.path = try p.gpa.dupe(u8, "/test.zig"); if (!panes.File.supportsContext(pane)) return error.SkipZigTest; @@ -2250,7 +2250,7 @@ test "a body with context rows is painted once, and the grid's rows are what a p /// Two columns, the file in the left one and a shell in the right, the /// right one focused with its cursor settled. fn twoColumns(p: *Pardes, arena: *std.heap.ArenaAllocator) !void { - _ = try p.setTestFile("some text on a line\n" ** 6); + _ = try p.setTestFile(std.mem.asBytes(&@as([6][20]u8, @splat("some text on a line\n".*)))); _ = try p.newShell(1, ""); try std.testing.expect(layout.splitColumn(p, 0, 1, false)); p.sync(); @@ -2386,7 +2386,7 @@ test "a pane focus reaches brightens on the same spring the one it leaves dims o if (!builtins.capabilities.lift) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .cols = 80, .rows = 24 }); defer p.deinit(); - _ = try p.setTestFile("some text on a line\n" ** 4); + _ = try p.setTestFile(std.mem.asBytes(&@as([4][20]u8, @splat("some text on a line\n".*)))); p.newScratchBelow(0); p.settings.inactive_dim = 40; p.settings.motion = .smooth; diff --git a/src/dump.zig b/src/dump.zig index 402918c4..545d09ec 100644 --- a/src/dump.zig +++ b/src/dump.zig @@ -421,9 +421,9 @@ pub const Parsed = struct { pub fn check(gpa: std.mem.Allocator, bytes: []const u8, why: []u8) !void { var arena: std.heap.ArenaAllocator = .init(gpa); defer arena.deinit(); - const source = try arena.allocator().dupeZ(u8, bytes); - var diag: std.zon.parse.Diagnostics = .{}; - const state = std.zon.parse.fromSliceAlloc(State, arena.allocator(), source, &diag, .{ .free_on_error = false }) catch |err| { + const source = try arena.allocator().dupeSentinel(u8, bytes, 0); + var diag: std.zon.parse.Diagnostics = undefined; + const state = std.zon.parse.fromSlice(State, .{ .gpa = gpa, .arena = arena.allocator(), .source = source, .diagnostics = &diag }) catch |err| { const said = pardes.Messages.zonWhy(&diag, why[0 .. why.len - 1]); if (said.ptr != why.ptr) @memcpy(why[0..said.len], said); why[said.len] = 0; @@ -436,21 +436,13 @@ pub fn readZon(gpa: std.mem.Allocator, bytes: []const u8, label: []const u8) !Pa var parsed: Parsed = .{ .value = undefined, .arena = .init(gpa) }; errdefer parsed.deinit(); const arena = parsed.arena.allocator(); - const source = try gpa.dupeZ(u8, bytes); + const source = try gpa.dupeSentinel(u8, bytes, 0); defer gpa.free(source); - parsed.value = if (builtin.os.tag == .emscripten or builtin.os.tag == .freestanding) - std.zon.parse.fromSliceAlloc(State, arena, source, null, .{ .free_on_error = false }) catch |err| { - if (err == error.ParseZon) log.err("parse dump {s}: {s}", .{ label, @errorName(err) }); - return err; - } - else blk: { - var diag: std.zon.parse.Diagnostics = .{}; - defer diag.deinit(arena); - break :blk std.zon.parse.fromSliceAlloc(State, arena, source, &diag, .{ .free_on_error = false }) catch |err| { - if (err == error.ParseZon) log.err("parse dump {s}: {f}", .{ label, diag }); - return err; - }; + var diag: std.zon.parse.Diagnostics = undefined; + parsed.value = std.zon.parse.fromSlice(State, .{ .gpa = gpa, .arena = arena, .source = source, .diagnostics = &diag }) catch |err| { + if (err == error.ParseZon) log.err("{f}", .{diag.fmt(label)}); + return err; }; try validate(parsed.value); return parsed; diff --git a/src/esp32p4.zig b/src/esp32p4.zig index 6772e75d..ed18efd9 100644 --- a/src/esp32p4.zig +++ b/src/esp32p4.zig @@ -506,7 +506,7 @@ fn apply(c: *pardes.Pardes, ev: vaxis.Event) void { c.update(.pointer_leave); dirty = true; }, - .focus_in, .mouse_leave => {}, + .focus_in, .mouse_leave, .cursor_position => {}, // Capability replies. vaxis's own Loop sets these fields directly (`Loop.zig:377-403`); // with no Loop, this is where they land. DA1 is the terminator: every terminal answers it // last, so it is the signal that the whole handshake is in and the detected features can be diff --git a/src/esp32p4/input_rescue.zig b/src/esp32p4/input_rescue.zig index ba6974d2..d7556d46 100644 --- a/src/esp32p4/input_rescue.zig +++ b/src/esp32p4/input_rescue.zig @@ -245,7 +245,7 @@ test "a long transmit does not lose the input that arrives during it" { defer port.sent.deinit(std.testing.allocator); var ring: Ring = .{}; - const frame = "\x1b[1;1H" ++ "x" ** 400; + const frame = "\x1b[1;1H" ++ &@as([400]u8, @splat('x')); try std.testing.expectEqual(@as(u32, 0), pump(&port, &ring, frame, 1_000_000)); // Every output byte went out, in order. diff --git a/src/esp32p4/selftest.zig b/src/esp32p4/selftest.zig index 97b43059..19b71e90 100644 --- a/src/esp32p4/selftest.zig +++ b/src/esp32p4/selftest.zig @@ -270,7 +270,7 @@ export fn zig_main() noreturn { const typed = "the quick brown fox jumps over the lazy dog, and then some more besides"; var port: FakePort = .{ .tx_cap = 2, .incoming = typed }; var ring: input_rescue.Ring = .{}; - const frame = "\x1b[1;1H" ++ "x" ** 300; + const frame = "\x1b[1;1H" ++ &@as([300]u8, @splat('x')); const abandoned = input_rescue.pump(&port, &ring, frame, 1_000_000); check("the frame went out whole", abandoned == 0 and port.sent == frame.len); diff --git a/src/fs.zig b/src/fs.zig index fb76cee3..061d4225 100644 --- a/src/fs.zig +++ b/src/fs.zig @@ -677,7 +677,7 @@ pub fn statPath(io: std.Io, path: []const u8, options: std.Io.Dir.StatFileOption } test "a path with a part over 255 bytes is NameTooLong, never a panic" { - const long = "/tmp/" ++ "x" ** 300 ++ "/f"; + const long = "/tmp/" ++ &@as([300]u8, @splat('x')) ++ "/f"; try std.testing.expectError(error.NameTooLong, statPath(std.testing.io, long, .{})); } @@ -1152,7 +1152,7 @@ pub fn find(arena: std.mem.Allocator, dir: []const u8, pat: []const u8, out: []u notes.skipped += 1; continue; }) orelse break; - if (std.ascii.indexOfIgnoreCase(e.basename, pat) != null) { + if (std.ascii.findIgnoreCase(e.basename, pat) != null) { hits[hits_len] = try arena.dupe(u8, e.path); hits_len += 1; } @@ -1167,7 +1167,7 @@ pub fn find(arena: std.mem.Allocator, dir: []const u8, pat: []const u8, out: []u } else { for (sources) |s| { if (hits_len >= hits.len) break; - if (std.ascii.indexOfIgnoreCase(std.fs.path.basename(s.path), pat) != null) { + if (std.ascii.findIgnoreCase(std.fs.path.basename(s.path), pat) != null) { hits[hits_len] = s.path; hits_len += 1; } @@ -1201,7 +1201,7 @@ fn grepText(path: []const u8, text: []const u8, pat: []const u8, out: []u8, budg while (it.next()) |raw| { line += 1; if (result.hits >= budget) break; - const at = std.ascii.indexOfIgnoreCase(raw, pat) orelse continue; + const at = std.ascii.findIgnoreCase(raw, pat) orelse continue; const ln = std.mem.trimEnd(u8, raw, " \t\r"); var cut = @min(ln.len, 200); while (cut > 0 and cut < ln.len and ln[cut] & 0xc0 == 0x80) cut -= 1; diff --git a/src/gui/Post.zig b/src/gui/Post.zig index fc256135..8d0e14d9 100644 --- a/src/gui/Post.zig +++ b/src/gui/Post.zig @@ -610,13 +610,13 @@ const GhosttyUniforms = extern struct { }; test "the uniform block is ghostty's, offset for offset, and the prefix names it in that order" { - const ours = @typeInfo(Uniforms).@"struct".fields; - const theirs = @typeInfo(GhosttyUniforms).@"struct".fields; - try std.testing.expectEqual(theirs.len, ours.len); + const ours = @typeInfo(Uniforms).@"struct"; + const theirs = @typeInfo(GhosttyUniforms).@"struct"; + try std.testing.expectEqual(theirs.field_names.len, ours.field_names.len); try std.testing.expectEqual(@sizeOf(GhosttyUniforms), @sizeOf(Uniforms)); - inline for (ours, theirs) |a, b| { - try std.testing.expectEqual(@offsetOf(GhosttyUniforms, b.name), @offsetOf(Uniforms, a.name)); - try std.testing.expectEqual(b.type, a.type); + inline for (ours.field_names, ours.field_types, theirs.field_names, theirs.field_types) |a, a_type, b, b_type| { + try std.testing.expectEqual(@offsetOf(GhosttyUniforms, b), @offsetOf(Uniforms, a)); + try std.testing.expectEqual(b_type, a_type); } // By NAME: iSelectionForegroundColor sits where ghostty's Zig says // selection_background, which is where its GLSL puts the foreground. @@ -633,7 +633,7 @@ test "the uniform block is ghostty's, offset for offset, and the prefix names it .{ "vec3", "iBackgroundColor" }, .{ "vec3", "iForegroundColor" }, .{ "vec3", "iCursorColor" }, .{ "vec3", "iCursorText" }, .{ "vec3", "iSelectionForegroundColor" }, .{ "vec3", "iSelectionBackgroundColor" }, }; - try std.testing.expectEqual(ours.len, glsl.len); + try std.testing.expectEqual(ours.field_names.len, glsl.len); // Its three blocks (Globals, Palette, Colors) read as one, in order. const start = std.mem.indexOf(u8, prefix, "uniform Globals {").?; const end = std.mem.indexOf(u8, prefix, "uniform PardesPost {").?; diff --git a/src/gui/gui.zig b/src/gui/gui.zig index 5a026664..b8687fd3 100644 --- a/src/gui/gui.zig +++ b/src/gui/gui.zig @@ -25,11 +25,7 @@ const ninep_io = @import("../9p_io.zig"); const detached_client = @import("../detached/client.zig"); const tracy = @import("../tracy.zig"); -pub const c = @cImport({ - @cDefine("SDL_DISABLE_OLD_NAMES", "1"); - @cInclude("SDL3/SDL.h"); - @cInclude("font.h"); -}); +pub const c = @import("gui_c"); // SDL3 and font.h, translated by build.zig const host_io = @import("../host_io.zig"); extern "c" fn setenv(name: [*:0]const u8, value: [*:0]const u8, overwrite: c_int) c_int; @@ -4806,7 +4802,7 @@ fn dumpThemes(ctx: ?*anyopaque, pane: u8) void { } fn putClipboard(gpa: std.mem.Allocator, text: []const u8) void { - const z = gpa.dupeZ(u8, text) catch return; + const z = gpa.dupeSentinel(u8, text, 0) catch return; defer gpa.free(z); _ = c.SDL_SetClipboardText(z.ptr); } @@ -5131,7 +5127,7 @@ test "a wheel over codex in a terminal reaches its pty as the mouse reports it a wheelAt(&g, &in, .{ .y = 1, .integer_y = 1 }, mc); try std.testing.expectEqualStrings("\x1b[<64;6;4M", try Sent.take(core, &sent)); wheelAt(&g, &in, .{ .y = -3, .integer_y = -3 }, mc); - try std.testing.expectEqualStrings("\x1b[<65;6;4M" ** 3, try Sent.take(core, &sent)); + try std.testing.expectEqualStrings(std.mem.asBytes(&@as([3][10]u8, @splat("\x1b[<65;6;4M".*))), try Sent.take(core, &sent)); try std.testing.expectEqual(@as(i32, 0), g.scroll_notch); try std.testing.expect(g.scroll_pane == null); // A touchpad's fractions add up to a turn. @@ -7157,7 +7153,7 @@ test "every pixel of a grip column is band, button or ring, and the button sits const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .cols = 60, .rows = 20 }); defer core.deinit(); - const pane = try core.setTestFile("text\n" ** 60); + const pane = try core.setTestFile(std.mem.asBytes(&@as([60][5]u8, @splat("text\n".*)))); pane.tag.own = try gpa.dupe(u8, "second\nthird"); core.newScratchBelow(0); for ([_]struct { w: u32, h: u32, tag: u32 }{ .{ .w = 10, .h = 20, .tag = 12 }, .{ .w = 17, .h = 35, .tag = 30 }, .{ .w = 9, .h = 19, .tag = 15 }, .{ .w = 12, .h = 24, .tag = 24 } }) |m| for ([_]bool{ false, true }) |bottom| for ([_]usize{ 0, 1 }) |focus| for ([_]bool{ false, true }) |dirty| { @@ -10156,7 +10152,7 @@ test "the selection glows round its block, on the body only, fading in over 100 const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .cols = 80, .rows = 24 }); defer core.deinit(); - _ = try core.setTestFile("some text on a line\n" ** 30); + _ = try core.setTestFile(std.mem.asBytes(&@as([30][20]u8, @splat("some text on a line\n".*)))); core.settings.selection_glow = true; pardes.test_api.sync(core); const s = try core.render(core.scratch.allocator()); @@ -10206,7 +10202,7 @@ test "a hovered word glows along its foot, never over the cursor, fading in over const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .cols = 80, .rows = 24 }); defer core.deinit(); - _ = try core.setTestFile("some text on a line\n" ** 30); + _ = try core.setTestFile(std.mem.asBytes(&@as([30][20]u8, @splat("some text on a line\n".*)))); core.settings.hover_glow = true; pardes.test_api.sync(core); const s = try core.render(core.scratch.allocator()); @@ -10256,7 +10252,7 @@ test "occlusion shades each body's edges, inside it, never a tag, a grip or the const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .cols = 80, .rows = 24 }); defer core.deinit(); - _ = try core.setTestFile("some text on a line\n" ** 30); + _ = try core.setTestFile(std.mem.asBytes(&@as([30][20]u8, @splat("some text on a line\n".*)))); core.settings.occlusion = true; pardes.test_api.sync(core); const s = try core.render(core.scratch.allocator()); @@ -10295,7 +10291,7 @@ test "a body tells its first line, and lapis's dots under it move a quarter as f const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .cols = 80, .rows = 24 }); defer core.deinit(); - const pane = try core.setTestFile("line\n" ** 80); + const pane = try core.setTestFile(std.mem.asBytes(&@as([80][5]u8, @splat("line\n".*)))); pane.scrollBy(8); try std.testing.expect(core.executeBuiltinLine(0, "Theme lapis")); core.settings.parallax = true; diff --git a/src/host_io.zig b/src/host_io.zig index a90c903d..d789a77d 100644 --- a/src/host_io.zig +++ b/src/host_io.zig @@ -172,7 +172,7 @@ pub const Lsp = struct { const declared_path = if (file) |f| f.path else ""; const path = try gpa.dupe(u8, filesystem.localPath(declared_path) orelse declared_path); errdefer gpa.free(path); - const source = try gpa.dupeZ(u8, if (file) |f| f.content else ""); + const source = try gpa.dupeSentinel(u8, if (file) |f| f.content else "", 0); errdefer gpa.free(source); const arg = try gpa.dupe(u8, req.arg); errdefer gpa.free(arg); @@ -210,7 +210,7 @@ pub const Lsp = struct { if (f.output != null) continue; const local = filesystem.localPath(f.path) orelse continue; const other_path = try gpa.dupe(u8, local); - const other_source = gpa.dupeZ(u8, f.content) catch |err| { + const other_source = gpa.dupeSentinel(u8, f.content, 0) catch |err| { gpa.free(other_path); return err; }; @@ -785,7 +785,7 @@ pub const Shell = struct { var dir_buf: [4096]u8 = undefined; const dir = dir_buf[0..try tmp.dir.realPath(std.testing.io, &dir_buf)]; var path: [4200:0]u8 = undefined; - const p = try std.fmt.bufPrintZ(&path, "{s}/mysh", .{dir}); + const p = try std.fmt.bufPrintSentinel(&path, "{s}/mysh", .{dir}, 0); try std.testing.expectEqual(Family.bash, familyOf(p.ptr)); try std.testing.expectEqual(Family.none, familyOf("/bin/sh")); } @@ -1662,7 +1662,7 @@ pub fn forkShell( test "shell spawn rejects invalid directories before creating a child" { const rcs: Shell.PromptFiles = .{}; try std.testing.expectError(error.InvalidPath, forkShell(null, 0, &rcs, "/bin/sh", "/tmp\x00/ignored", 24, 80, null)); - const too_long = [_]u8{'x'} ** 4096; + const too_long = @as([4096]u8, @splat('x')); try std.testing.expectError(error.NameTooLong, forkShell(null, 0, &rcs, "/bin/sh", &too_long, 24, 80, null)); var tmp = std.testing.tmpDir(.{}); @@ -1675,7 +1675,7 @@ test "shell spawn rejects invalid directories before creating a child" { try std.testing.expectError(error.NotDir, forkShell(null, 0, &rcs, "/bin/sh", file, 24, 80, null)); const explicit_file = try std.fmt.bufPrint(&path_buf, "/n/os{s}/file", .{directory}); try std.testing.expectError(error.NotDir, forkShell(null, 0, &rcs, "/bin/sh", explicit_file, 24, 80, null)); - const missing = try std.fmt.bufPrint(&path_buf, "{s}/missing/" ++ ("child/" ** 50), .{directory}); + const missing = try std.fmt.bufPrint(&path_buf, "{s}/missing/" ++ (std.mem.asBytes(&@as([50][6]u8, @splat("child/".*)))), .{directory}); try std.testing.expect(missing.len > 256); try std.testing.expectError(error.FileNotFound, forkShell(null, 0, &rcs, "/bin/sh", missing, 24, 80, null)); } @@ -1720,7 +1720,7 @@ test "shell spawn uses an explicit OS directory longer than 256 bytes" { if (!haveFile("/bin/sh")) return error.SkipZigTest; var tmp = std.testing.tmpDir(.{}); defer tmp.cleanup(); - const sub_path = "nested-directory-with-more-than-forty-characters/" ** 7; + const sub_path = std.mem.asBytes(&@as([7][49]u8, @splat("nested-directory-with-more-than-forty-characters/".*))); try tmp.dir.createDirPath(std.testing.io, sub_path); var directory_buf: [4096]u8 = undefined; const directory = directory_buf[0..try tmp.dir.realPathFile(std.testing.io, sub_path, &directory_buf)]; @@ -1801,7 +1801,7 @@ test "shell cwd rejects truncation and preserves an owned child path longer than const io = std.testing.io; var tmp = std.testing.tmpDir(.{}); defer tmp.cleanup(); - const nested = "nested-directory-with-more-than-forty-characters/" ** 24; + const nested = std.mem.asBytes(&@as([24][49]u8, @splat("nested-directory-with-more-than-forty-characters/".*))); try tmp.dir.createDirPath(io, nested); var path_buf: [pardes.memory.limits.host_path_cap + 1]u8 = undefined; const directory = path_buf[0..try tmp.dir.realPathFile(io, nested, &path_buf)]; diff --git a/src/image.zig b/src/image.zig index 72520ad6..8bf9de08 100644 --- a/src/image.zig +++ b/src/image.zig @@ -242,11 +242,11 @@ pub const GlyphArt = struct { const glyphs_ascii = glyphs ++ ascii_glyphs; fn matchCell(cell: *const [64][3]u8, pal: Palette, gset: []const Glyph) Cell { - var counts = [_]u16{0} ** 16; + var counts = @as([16]u16, @splat(0)); for (cell) |px| counts[nearest(px, pal)] += 1; var cand: [4]u4 = undefined; var ncand: usize = 0; - var used = [_]bool{false} ** 16; + var used = @as([16]bool, @splat(false)); while (ncand < 4) : (ncand += 1) { var best: ?usize = null; for (counts, 0..) |c, i| { @@ -432,7 +432,7 @@ pub const GlyphArt = struct { test "render: tiny image produces a grid within bounds" { const a = std.testing.allocator; // 2x2 checker, RGBA - var img = [_]u8{0} ** (2 * 2 * 4); + var img = @as([2 * 2 * 4]u8, @splat(0)); img[0] = 255; img[1] = 255; img[2] = 255; diff --git a/src/linux/v9fs.zig b/src/linux/v9fs.zig index 5e031bfe..15945e7b 100644 --- a/src/linux/v9fs.zig +++ b/src/linux/v9fs.zig @@ -64,11 +64,11 @@ pub fn helperPath(buf: []u8) ![:0]u8 { const path = if (std.c.getenv("PARDES_V9FS_HELPER")) |env| blk: { const override = std.mem.span(env); if (!std.fs.path.isAbsolute(override)) return error.AbsoluteHelperPathRequired; - break :blk try std.fmt.bufPrintZ(buf, "{s}", .{override}); + break :blk try std.fmt.bufPrintSentinel(buf, "{s}", .{override}, 0); } else blk: { var own: [4096]u8 = undefined; const parent = std.fs.path.dirname(try selfPath(&own)) orelse return error.ExecutablePathUnavailable; - break :blk try std.fmt.bufPrintZ(buf, "{s}/{s}", .{ parent, executable }); + break :blk try std.fmt.bufPrintSentinel(buf, "{s}/{s}", .{ parent, executable }, 0); }; if (std.c.access(path.ptr, 1) != 0) return error.V9fsHelperNotFound; return path; diff --git a/src/locations.zig b/src/locations.zig index 745393e5..2828566d 100644 --- a/src/locations.zig +++ b/src/locations.zig @@ -587,12 +587,13 @@ pub const Config = struct { const key = token[0..colon]; const value = token[colon + 1 ..]; var found = false; - inline for (@typeInfo(Config).@"struct".fields) |field| { - if (std.mem.eql(u8, key, field.name)) { - @field(result, field.name) = switch (@typeInfo(field.type)) { + const info = @typeInfo(Config).@"struct"; + inline for (info.field_names, info.field_types) |name, T| { + if (std.mem.eql(u8, key, name)) { + @field(result, name) = switch (@typeInfo(T)) { .bool => if (std.mem.eql(u8, value, "on")) true else if (std.mem.eql(u8, value, "off")) false else return error.ExpectedOnOrOff, - .int => std.fmt.parseInt(field.type, value, 10) catch return error.InvalidContext, - .@"enum" => std.meta.stringToEnum(field.type, value) orelse return error.InvalidLayout, + .int => std.fmt.parseInt(T, value, 10) catch return error.InvalidContext, + .@"enum" => std.meta.stringToEnum(T, value) orelse return error.InvalidLayout, else => @compileError("unsupported location setting type"), }; found = true; @@ -604,11 +605,12 @@ pub const Config = struct { } pub fn write(config: Config, writer: *std.Io.Writer) !void { - inline for (@typeInfo(Config).@"struct".fields, 0..) |field, index| { + const info = @typeInfo(Config).@"struct"; + inline for (info.field_names, info.field_types, 0..) |name, T, index| { if (index > 0) try writer.writeByte(' '); - try writer.writeAll(field.name ++ ":"); - const value = @field(config, field.name); - switch (@typeInfo(field.type)) { + try writer.writeAll(name ++ ":"); + const value = @field(config, name); + switch (@typeInfo(T)) { .bool => try writer.writeAll(if (value) "on" else "off"), .int => try writer.print("{d}", .{value}), .@"enum" => try writer.writeAll(@tagName(value)), @@ -809,11 +811,11 @@ test "locations cache bounds memory and entries and evicts least recently used" try std.testing.expectEqual(@as(usize, 2), cache.len); try std.testing.expect(cache.bytes <= 40); try std.testing.expectEqualStrings("b.unknown", cache.entries[0].path); - try std.testing.expectError(error.SourceTooLarge, cache.getBounded(gpa, "large.unknown", "x" ** 41, false, false, 40)); + try std.testing.expectError(error.SourceTooLarge, cache.getBounded(gpa, "large.unknown", &@as([41]u8, @splat('x')), false, false, 40)); try std.testing.expectEqual(@as(usize, 2), cache.len); if (syntax.supportsPath("large.zig")) { const before = cache.analyses; - try std.testing.expectError(error.SourceTooLarge, cache.getBounded(gpa, "large.zig", "x" ** 25, true, true, 40)); + try std.testing.expectError(error.SourceTooLarge, cache.getBounded(gpa, "large.zig", &@as([25]u8, @splat('x')), true, true, 40)); try std.testing.expectEqual(before, cache.analyses); } cache.deinit(gpa); diff --git a/src/look.zig b/src/look.zig index 1065cb02..f9c470cf 100644 --- a/src/look.zig +++ b/src/look.zig @@ -671,7 +671,7 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start const cl: usize = @intCast(@max(0, pane.body.cur_row)); const cc: usize = @intCast(@max(0, pane.body.cur_col)); for (lines, 0..) |ln, i| { - const at = std.ascii.indexOfIgnoreCase(ln, pat) orelse continue; + const at = std.ascii.findIgnoreCase(ln, pat) orelse continue; const row = try std.fmt.allocPrint(arena, "{s}:{d}:{d}{c}{d} {s}\n", .{ loc, i + 1, at + 1, config.range_sep, at + pat.len, std.mem.trimEnd(u8, ln, " \t"), }); @@ -1273,7 +1273,7 @@ fn missText(text: []const u8, buf: *[miss_path_cap + "…".len]u8) []const u8 { test "a miss names a long path by its start and its end, its middle given up to …" { var cut: [miss_path_cap + "…".len]u8 = undefined; - const long = "/home/someone/" ++ "deep/" ** 80 ++ "the-file.zig"; + const long = "/home/someone/" ++ std.mem.asBytes(&@as([80][5]u8, @splat("deep/".*))) ++ "the-file.zig"; const said = missText(long, &cut); try std.testing.expect(said.len <= miss_path_cap); try std.testing.expect(std.mem.startsWith(u8, said, "/home/someone/")); @@ -2185,7 +2185,7 @@ test "Look ignores missing and out of bounds pane addresses" { test "a pointer resting where a look-hover found nothing does not wake the core every frame" { const p = try Pardes.init(std.testing.allocator, .{ .cols = 80, .rows = 24 }); defer p.deinit(); - _ = try p.setTestFile(" \n" ** 30); + _ = try p.setTestFile(std.mem.asBytes(&@as([30][6]u8, @splat(" \n".*)))); p.sync(); const r = p.rects[p.active]; // Resting on blanks in the body: a wait, then nothing to open. diff --git a/src/lsp/lsp.zig b/src/lsp/lsp.zig index 96d42bbd..4a70efe5 100644 --- a/src/lsp/lsp.zig +++ b/src/lsp/lsp.zig @@ -92,7 +92,7 @@ else pub const backend_name = "lsp-client"; pub const supports: std.EnumSet(Kind) = blk: { - var s: std.EnumSet(Kind) = .initEmpty(); + var s: std.EnumSet(Kind) = .empty; for (0..backends.len) |i| s.setUnion(backends[i].supports); break :blk s; }; diff --git a/src/lsp/lsp_client.zig b/src/lsp/lsp_client.zig index 54bbaa79..b396879b 100644 --- a/src/lsp/lsp_client.zig +++ b/src/lsp/lsp_client.zig @@ -159,7 +159,7 @@ pub const specs = [_]Spec{ /// answered per connection; a kind its server never advertised simply returns /// no rows, which the harness reports as CLAIMED-EMPTY per language — honest, /// since the claim here is about the protocol, not about every server. -pub const supports: std.EnumSet(lsp.Kind) = .initFull(); +pub const supports: std.EnumSet(lsp.Kind) = .full; /// Routing: an extension in the table whose spec is not disabled by its env /// var. Deliberately does NOT probe for the binary — this runs on the Tab diff --git a/src/macos.zig b/src/macos.zig index 20463481..37db0abd 100644 --- a/src/macos.zig +++ b/src/macos.zig @@ -724,7 +724,7 @@ fn initCore(runtime: ?*const Runtime, cols_arg: u16, rows_arg: u16) !void { // in Debug, libc's malloc otherwise. libc's measured at least as fast as // std.heap.smp_allocator everywhere and faster on MuPDF's page-sized // rasters, so this shell no longer differs. - const gpa = if (builtin.mode == .Debug) debug_gpa.allocator() else std.heap.c_allocator; + const gpa = if (builtin.mode == .debug) debug_gpa.allocator() else std.heap.c_allocator; const allocs = pardes.memory.init(gpa); errdefer pardes.memory.deinit(); @@ -858,7 +858,7 @@ export fn pardes_deinit() void { st.gpa.destroy(st.threaded); pardes.memory.deinit(); // Leaks are logged and do not fail, as std.start treats the others'. - if (builtin.mode == .Debug) { + if (builtin.mode == .debug) { _ = debug_gpa.deinit(); debug_gpa = .init; } @@ -2871,22 +2871,22 @@ test "pardes.h matches the Zig boundary" { try expectEqual(@offsetOf(c.pardes_cell_s, "len"), @offsetOf(Cell, "len")); try expectEqual(@offsetOf(c.pardes_cell_s, "flags"), @offsetOf(Cell, "flags")); try expectEqual(@sizeOf(c.pardes_scene_s), @sizeOf(Scene)); - inline for (@typeInfo(Scene).@"struct".fields) |field| - try expectEqual(@offsetOf(c.pardes_scene_s, field.name), @offsetOf(Scene, field.name)); + inline for (@typeInfo(Scene).@"struct".field_names) |name| + try expectEqual(@offsetOf(c.pardes_scene_s, name), @offsetOf(Scene, name)); try expectEqual(@sizeOf(c.pardes_panel_box_s), @sizeOf(PanelBox)); - inline for (@typeInfo(PanelBox).@"struct".fields) |field| - try expectEqual(@offsetOf(c.pardes_panel_box_s, field.name), @offsetOf(PanelBox, field.name)); + inline for (@typeInfo(PanelBox).@"struct".field_names) |name| + try expectEqual(@offsetOf(c.pardes_panel_box_s, name), @offsetOf(PanelBox, name)); try expectEqual(@sizeOf(c.pardes_panel_track_s), @sizeOf(PanelTrack)); - inline for (@typeInfo(PanelTrack).@"struct".fields) |field| { - const c_name = comptime if (std.mem.eql(u8, field.name, "motion")) + inline for (@typeInfo(PanelTrack).@"struct".field_names) |name| { + const c_name = comptime if (std.mem.eql(u8, name, "motion")) "reserved0" else - field.name; - try expectEqual(@offsetOf(c.pardes_panel_track_s, c_name), @offsetOf(PanelTrack, field.name)); + name; + try expectEqual(@offsetOf(c.pardes_panel_track_s, c_name), @offsetOf(PanelTrack, name)); } try expectEqual(@sizeOf(c.pardes_image_s), @sizeOf(Image)); - inline for (@typeInfo(Image).@"struct".fields) |field| - try expectEqual(@offsetOf(c.pardes_image_s, field.name), @offsetOf(Image, field.name)); + inline for (@typeInfo(Image).@"struct".field_names) |name| + try expectEqual(@offsetOf(c.pardes_image_s, name), @offsetOf(Image, name)); try expectEqual(@sizeOf(c.pardes_runtime_s), @sizeOf(Runtime)); try expectEqual(@as(u32, c.PARDES_COLOR_DEFAULT), color_default); @@ -3177,7 +3177,7 @@ test "mac tag layer ABI preserves logical capacity and physical grip" { }); try std.testing.expectEqual(@as(u32, 11), pardes_tag_layer_value(0, 8)); try std.testing.expectEqual(@as(u8, 'Z'), pardes_tag_layer_cells(0).?[11].text[0]); - var canonical: [24]Cell = @splat(encodeCell(.{ .text = .{'G'} ++ .{0} ** 6, .len = 1, .default = false })); + var canonical: [24]Cell = @splat(encodeCell(.{ .text = .{ 'G', 0, 0, 0, 0, 0, 0 }, .len = 1, .default = false })); var source: [24]pardes.Cell = @splat(.{}); blankTagCells(&canonical, 8, 3, &source, &core.surface.tag_layers); try std.testing.expectEqual(@as(u8, 'G'), canonical[9].text[0]); diff --git a/src/main.zig b/src/main.zig index 26259b43..e3bae75f 100644 --- a/src/main.zig +++ b/src/main.zig @@ -464,6 +464,12 @@ fn nativeMain(init: std.process.Init) !void { } opts.tty_only = explicit_tty or session_only; opts.mounts = mounts[0..mounts_len]; + // `zig build run-isolated` with nothing named: something to look at on + // arrival, read out of the binary itself. + if (pardes.isolated and positional == null) { + session_only = false; + positional = "src/pardes.zig"; + } if (detach != null and attach != null) badArgs(init.io, "--detach and --attach are opposites: one runs the session, the other joins one", .{}); // --wait waits on the pane a FILE opens: with none there is nothing to diff --git a/src/memory.zig b/src/memory.zig index abfd40f6..48403c55 100644 --- a/src/memory.zig +++ b/src/memory.zig @@ -38,7 +38,7 @@ pub const limits = struct { }; const Allocator = std.mem.Allocator; -const debug_enabled = builtin.mode == .Debug; +const debug_enabled = builtin.mode == .debug; pub const Allocators = struct { pardes: Allocator, diff --git a/src/modal.zig b/src/modal.zig index 50e0ddf4..82f1a3a0 100644 --- a/src/modal.zig +++ b/src/modal.zig @@ -163,7 +163,7 @@ pub const Normal = struct { }; pub const Input = struct { - roles: std.EnumSet(Role) = .initEmpty(), + roles: std.EnumSet(Role) = .empty, cp: u21, ctrl: bool = false, alt: bool = false, diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index b78c46e7..3feda5a2 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -646,7 +646,7 @@ test "the address language, form by form" { try testing.expectEqualStrings(c[1], th.wr(p, addr, c[0]).reply.ename); } - const nested = "," ** 4096; + const nested = &@as([4096]u8, @splat(',')); _ = th.wr(p, addr, "#0"); try testing.expectEqual(E.INVAL, th.wr(p, addr, nested).errno()); } diff --git a/src/ninep/cols.zig b/src/ninep/cols.zig index e3b456ba..59f6ae42 100644 --- a/src/ninep/cols.zig +++ b/src/ninep/cols.zig @@ -288,18 +288,18 @@ test "exactly 4096 bytes written with echo's newline after a truncation fit ever const p = try th.withFile(testing.allocator, "x\n"); defer p.deinit(); const serial = th.serialOf(p); - const text = "w" ** 4096 ++ "\n"; + const text = &@as([4096]u8, @splat('w')) ++ "\n"; for ([_]u64{ tree.Node.of(serial, .tag), tree.Node.ofCol(layout.columnSerial(p, 0), .tag), @intFromEnum(tree.TopFile.tag) }) |node| { _ = th.call(p, .{ .tag = 1, .op = .setattr, .node = node, .truncate = true }); const h = th.call(p, .{ .tag = 2, .op = .open, .node = node, .omode = 1 }).reply.handle; const r = th.call(p, .{ .tag = 3, .op = .write, .node = node, .handle = h, .data = text }); try testing.expectEqual(tree.Status.ok, r.reply.status); _ = th.call(p, .{ .tag = 4, .op = .release, .node = node, .handle = h, .opened = true }); - try testing.expect(std.mem.endsWith(u8, th.rd(p, node, 0, 1 << 16).bytes, "w" ** 4096)); + try testing.expect(std.mem.endsWith(u8, th.rd(p, node, 0, 1 << 16).bytes, &@as([4096]u8, @splat('w')))); // One more byte is over. _ = th.call(p, .{ .tag = 5, .op = .setattr, .node = node, .truncate = true }); const h2 = th.call(p, .{ .tag = 6, .op = .open, .node = node, .omode = 1 }).reply.handle; - try testing.expectEqual(E.NOSPC, th.call(p, .{ .tag = 7, .op = .write, .node = node, .handle = h2, .data = "w" ** 4097 ++ "\n" }).errno()); + try testing.expectEqual(E.NOSPC, th.call(p, .{ .tag = 7, .op = .write, .node = node, .handle = h2, .data = &@as([4097]u8, @splat('w')) ++ "\n" }).errno()); _ = th.call(p, .{ .tag = 8, .op = .release, .node = node, .handle = h2, .opened = true }); } } @@ -316,7 +316,7 @@ test "a Newcol refused because the tags would not fit logs its err alone and spe for (p.col_panes[col][0..p.col_n[col]]) |k| { const each = p.panes[k].?; if (each.tag.own) |old| p.gpa.free(old); - each.tag.own = try p.gpa.dupe(u8, "word " ** 60); + each.tag.own = try p.gpa.dupe(u8, std.mem.asBytes(&@as([60][5]u8, @splat("word ".*)))); } try testing.expect(p.col_n[col] >= 2); p.sync(); @@ -588,7 +588,7 @@ test "a refused write after a truncation leaves every kind of tag as it was, and defer p.deinit(); const serial = th.serialOf(p); const col = layout.columnSerial(p, 0); - const big = "w" ** (limits.max_tag_tail + 1); + const big = &@as([limits.max_tag_tail + 1]u8, @splat('w')); // Each tag: a truncating open, then a write too long: nothing changes. for ([_]u64{ Node.of(serial, .tag), Node.ofCol(col, .tag), @intFromEnum(tree.TopFile.tag) }) |node| { _ = th.wr(p, node, " Mine"); diff --git a/src/ninep/ctl.zig b/src/ninep/ctl.zig index f6457b83..6b4bbafb 100644 --- a/src/ninep/ctl.zig +++ b/src/ninep/ctl.zig @@ -1682,9 +1682,9 @@ test "a look's miss repeats the path whole up to 256 bytes, then gives up its mi var line: [512]u8 = undefined; _ = wr(p, root, try std.fmt.bufPrint(&line, "{s}\n", .{long})); try testing.expect(th.logHas(p, try std.fmt.bufPrint(&line, "look: no match for \"{s}\"\n", .{long}))); - const huge = "é" ** 150; // 300 bytes: its middle cut to fit 256, between characters + const huge = std.mem.asBytes(&@as([150][2]u8, @splat("é".*))); // 300 bytes: its middle cut to fit 256, between characters _ = wr(p, root, huge ++ "\n"); - try testing.expect(th.logHas(p, "look: no match for \"" ++ "é" ** 63 ++ "…" ++ "é" ** 63 ++ "\"\n")); + try testing.expect(th.logHas(p, "look: no match for \"" ++ std.mem.asBytes(&@as([63][2]u8, @splat("é".*))) ++ "…" ++ std.mem.asBytes(&@as([63][2]u8, @splat("é".*))) ++ "\"\n")); } test "exec runs a builtin at the pane and records the pane it acted on" { @@ -2041,7 +2041,7 @@ test "Zerox twins dump their text once and come back on one buffer from a Restor defer p.deinit(); const first = serialOf(p); _ = wr(p, Node.of(first, .name), "/tmp/pardes-zerox-dump.txt\n"); - _ = wr(p, Node.of(first, .body), "unsaved " ** 64 ++ "\n"); + _ = wr(p, Node.of(first, .body), std.mem.asBytes(&@as([64][8]u8, @splat("unsaved ".*))) ++ "\n"); try testing.expect(p.executeBuiltinLine(0, "Zerox")); try testing.expect(p.executeBuiltinLine(0, "Zerox")); try pardes.dump.dumpState(p); @@ -2888,7 +2888,7 @@ test "a builtin that fails a ctl write logs only its err, so the same failure ag test "an exec line over the command limit fails the write and says the limit; a builtin's may be longer" { const p = try withFile(testing.allocator, "x\n"); defer p.deinit(); - const long = "echo " ++ "y" ** pardes.exec.command_max ++ "\n"; + const long = "echo " ++ &@as([pardes.exec.command_max]u8, @splat('y')) ++ "\n"; const refused = wr(p, root_exec, long); try testing.expectEqual(E.INVAL, refused.errno()); try testing.expectEqualStrings(e_too_long, refused.reply.ename); @@ -2896,7 +2896,7 @@ test "an exec line over the command limit fails the write and says the limit; a try testing.expect(th.logHas(p, try std.fmt.bufPrint(&rec, "err {d} exec: invalid command line: a command line is at most 1024 bytes", .{serialOf(p)}))); try testing.expectEqualStrings(e_control, wr(p, root_exec, "echo a\x01b\n").reply.ename); try testing.expect(th.logHas(p, try std.fmt.bufPrint(&rec, "err {d} exec: invalid command line: it holds", .{serialOf(p)}))); - try testing.expectEqual(Status.ok, wr(p, root_exec, "Msg " ++ "z" ** 1100 ++ "\n").reply.status); + try testing.expectEqual(Status.ok, wr(p, root_exec, "Msg " ++ &@as([1100]u8, @splat('z')) ++ "\n").reply.status); } test "Edit's p and = never fail for want of room for +Errors: the text is logged instead" { @@ -4012,8 +4012,8 @@ test "Back and Forward written to exec keep to the focused file, and JumpScope a defer p.deinit(); var tmp = testing.tmpDir(.{}); defer tmp.cleanup(); - try tmp.dir.writeFile(testing.io, .{ .sub_path = "a.txt", .data = "row\n" ** 100 }); - try tmp.dir.writeFile(testing.io, .{ .sub_path = "b.txt", .data = "row\n" ** 100 }); + try tmp.dir.writeFile(testing.io, .{ .sub_path = "a.txt", .data = std.mem.asBytes(&@as([100][4]u8, @splat("row\n".*))) }); + try tmp.dir.writeFile(testing.io, .{ .sub_path = "b.txt", .data = std.mem.asBytes(&@as([100][4]u8, @splat("row\n".*))) }); var dir_buf: [4096]u8 = undefined; const dir = dir_buf[0..try tmp.dir.realPath(testing.io, &dir_buf)]; var line: [4200]u8 = undefined; diff --git a/src/ninep/events.zig b/src/ninep/events.zig index 3910ac05..7b73dc0e 100644 --- a/src/ninep/events.zig +++ b/src/ninep/events.zig @@ -938,7 +938,7 @@ test "event records are acme's bytes, one per read, and .again when empty" { try testing.expectEqualStrings("ML0 ", rd(p, event, 0, 4).bytes); try testing.expectEqualStrings("3 4 3 one\n", rd(p, event, 0, 4096).bytes); - const big = "z" ** max_record_text; + const big = &@as([max_record_text]u8, @splat('z')); _ = noteAction(p, 0, .body_exec, 0, max_record_text, 0, big); try testing.expectEqualStrings("MX0 256 0 0 \n", rd(p, event, 0, 4096).bytes); @@ -1388,7 +1388,7 @@ test "opens of the log share the open records, and a closed one frees its record test "a click in the body reports the offsets of the word it takes, so the record can be written back" { const config = @import("../config.zig"); const gpa = testing.allocator; - const text = "alpha qty beta\n qty\n\tqty\ndef f(qty): return qty.x\n" ++ "w" ** 100 ++ " qty\n"; + const text = "alpha qty beta\n qty\n\tqty\ndef f(qty): return qty.x\n" ++ &@as([100]u8, @splat('w')) ++ " qty\n"; const p = try withFile(gpa, text); defer p.deinit(); // A scratch made over 9P, clicked while another pane has the keyboard. @@ -1459,7 +1459,7 @@ test "follow new skips what the log held and waits for what comes after" { test "a long msg record is cut between words at its cap, with an ellipsis" { const p = try withFile(testing.allocator, "x\n"); defer p.deinit(); - const long = "word " ** 60; // 300 bytes + const long = std.mem.asBytes(&@as([60][5]u8, @splat("word ".*))); // 300 bytes _ = wr(p, Node.of(serialOf(p), .ctl), "Msg " ++ long ++ "\n"); const log = @intFromEnum(tree.TopFile.log); const h = call(p, .{ .tag = 1, .op = .open, .node = log }).reply.handle; @@ -1485,7 +1485,7 @@ test "a record is one line of UTF-8: DEL and C1 are spaces, bytes not UTF-8 are test "a long err record keeps its end, its middle given up to an ellipsis" { const p = try withFile(testing.allocator, "x\n"); defer p.deinit(); - _ = wr(p, @intFromEnum(tree.TopFile.ctl), "Bogus" ++ "abcdefgh" ** 40 ++ "\n"); + _ = wr(p, @intFromEnum(tree.TopFile.ctl), "Bogus" ++ std.mem.asBytes(&@as([40][8]u8, @splat("abcdefgh".*))) ++ "\n"); const log = @intFromEnum(tree.TopFile.log); const h = call(p, .{ .tag = 1, .op = .open, .node = log }).reply.handle; const text = call(p, .{ .tag = 2, .op = .read, .node = log, .handle = h, .size = 1 << 16 }).bytes; @@ -1500,7 +1500,7 @@ test "a long err record keeps its end, its middle given up to an ellipsis" { test "two failures that agree only in what their records cut are two records, not one counted" { const p = try th.withFile(testing.allocator, "x\n"); defer p.deinit(); - const long = "why " ** 80; + const long = std.mem.asBytes(&@as([80][4]u8, @splat("why ".*))); p.setMessage(0, long ++ "first"); p.setMessage(0, long ++ "second"); try testing.expect(!th.logHas(p, "(x2)")); diff --git a/src/ninep/pane.zig b/src/ninep/pane.zig index 3b70d62f..8605cab8 100644 --- a/src/ninep/pane.zig +++ b/src/ninep/pane.zig @@ -1427,7 +1427,7 @@ test "a tag write past the limit is refused whole, naming the limit" { const tag = Node.of(serialOf(p), .tag); const before = try testing.allocator.dupe(u8, rd(p, tag, 0, 1 << 16).bytes); defer testing.allocator.free(before); - const big = "w" ** (limits.max_tag_tail + 10); + const big = &@as([limits.max_tag_tail + 10]u8, @splat('w')); const r = wr(p, tag, big); try testing.expectEqual(E.NOSPC, r.errno()); // `no space` is what 9ns maps to ENOSPC, so a shell through a mount @@ -1549,20 +1549,21 @@ test "a name with a component over 255 bytes is refused, and a long path's faile const p = try withFile(testing.allocator, "x\n"); defer p.deinit(); const serial = serialOf(p); - const refused = wr(p, Node.of(serial, .name), "/tmp/" ++ "c" ** 256 ++ "/f.txt\n"); + const refused = wr(p, Node.of(serial, .name), "/tmp/" ++ &@as([256]u8, @splat('c')) ++ "/f.txt\n"); try testing.expectEqual(E.INVAL, refused.errno()); try testing.expectEqualStrings("invalid file name: a component over 255 bytes", refused.reply.ename); - try testing.expectEqual(Status.ok, wr(p, Node.of(serial, .name), "/tmp/" ++ "c" ** 255 ++ "\n").reply.status); + try testing.expectEqual(Status.ok, wr(p, Node.of(serial, .name), "/tmp/" ++ &@as([255]u8, @splat('c')) ++ "\n").reply.status); // Held with no newline, it is refused by that write, not at the close. const held = call(p, .{ .tag = 1, .op = .open, .node = Node.of(serial, .name), .omode = 1 }).reply.handle; - try testing.expectEqual(E.INVAL, call(p, .{ .tag = 2, .op = .write, .node = Node.of(serial, .name), .handle = held, .data = "/tmp/" ++ "e" ** 256 }).errno()); + try testing.expectEqual(E.INVAL, call(p, .{ .tag = 2, .op = .write, .node = Node.of(serial, .name), .handle = held, .data = "/tmp/" ++ &@as([256]u8, @splat('e')) }).errno()); _ = call(p, .{ .tag = 3, .op = .release, .node = Node.of(serial, .name), .handle = held, .opened = true }); try testing.expectEqual(@as(usize, 0), p.fs.closed_lines.items.len); // A Save of a 3000-byte path the host refuses: the waiting write's // reason is at the end, the path giving up its middle. // (Under /tmp: under / a user's Save is refused permission, not // missing.) - const long = "/tmp/nonexistent-pardes-root/" ++ ("d" ** 200 ++ "/") ** 15 ++ "f.txt"; + const segment = &@as([200]u8, @splat('d')) ++ "/"; + const long = "/tmp/nonexistent-pardes-root/" ++ std.mem.asBytes(&@as([15][201]u8, @splat(segment.*))) ++ "f.txt"; const id = p.paneBySerial(serial).?; try nameBuffer(p, id, long, false); p.fs.late_failure_len = 0; diff --git a/src/ninep/pty.zig b/src/ninep/pty.zig index c58cf676..aeb57a00 100644 --- a/src/ninep/pty.zig +++ b/src/ninep/pty.zig @@ -1044,12 +1044,12 @@ test "a run's answer says cut when its output's start is gone, and reads a bound const q = try withTerm(gpa); defer q.deinit(); const kept = q.panes[0].?; - kept.terminal.?.vt.screens.get(.primary).?.pages.explicit_max_size = 1; + kept.terminal.?.vt.screens.get(.primary).?.pages.limits.set(.bytes, 1); const run = Node.of(serialOf(q), .pty_run); sh.put(q, prompt); h = sh.send(q, run); sh.put(q, c); - const lines = "pruned\r\n" ** 64; + const lines = std.mem.asBytes(&@as([64][8]u8, @splat("pruned\r\n".*))); var fed: usize = 0; while (!kept.terminal.?.stream.handler.out_start.?.garbage) : (fed += 1) { if (fed == 1000) return error.TestUnexpectedResult; @@ -1146,7 +1146,7 @@ test "a run's answer longer than its room is cut, its newline kept" { var slot: Run = .{}; const p = try th.withTerm(testing.allocator); defer p.deinit(); - answer(p, &slot, "busy: {s} is running", .{"x" ** 200}); + answer(p, &slot, "busy: {s} is running", .{&@as([200]u8, @splat('x'))}); try testing.expectEqual(slot.answer.len, slot.len); try testing.expectEqual(@as(u8, '\n'), slot.answer[slot.len - 1]); } diff --git a/src/ninep/screen.zig b/src/ninep/screen.zig index b991c3e4..8d4707e0 100644 --- a/src/ninep/screen.zig +++ b/src/ninep/screen.zig @@ -333,7 +333,7 @@ test "a terminal's body is its logical lines: a row the terminal wrapped is join const p = try th.withTerm(testing.allocator); defer p.deinit(); const pane = p.panes[0].?; - const long = "w" ** 50 ++ "x" ** 150; // wider than the pane: it wraps + const long = &@as([50]u8, @splat('w')) ++ &@as([150]u8, @splat('x')); // wider than the pane: it wraps try testing.expect(long.len > pane.cols); p.update(.{ .output = .{ .pane = 0, .bytes = long ++ "\r\nnext" } }); while (p.nextEffect()) |_| {} diff --git a/src/ninep/tree.zig b/src/ninep/tree.zig index eedab43b..5678832b 100644 --- a/src/ninep/tree.zig +++ b/src/ninep/tree.zig @@ -509,7 +509,7 @@ pub fn fitErr(text: []const u8, out: []u8) []const u8 { test "a reason past 128 bytes keeps its end: the path in it gives up its middle" { var out: [128]u8 = undefined; - const long = "Save /home/someone/projects/" ++ "deep/" ** 30 ++ "file.txt: no such directory"; + const long = "Save /home/someone/projects/" ++ std.mem.asBytes(&@as([30][5]u8, @splat("deep/".*))) ++ "file.txt: no such directory"; const fit = fitErr(long, &out); try testing.expect(fit.len <= 128); try testing.expect(std.mem.startsWith(u8, fit, "Save /home/someone/")); @@ -517,7 +517,7 @@ test "a reason past 128 bytes keeps its end: the path in it gives up its middle" try testing.expect(std.mem.indexOf(u8, fit, "…") != null); try testing.expect(std.unicode.utf8ValidateSlice(fit)); // No path: the whole text's middle goes, the end stays. - const words = "why " ** 40 ++ "the reason"; + const words = std.mem.asBytes(&@as([40][4]u8, @splat("why ".*))) ++ "the reason"; try testing.expect(std.mem.endsWith(u8, fitErr(words, &out), "the reason")); try testing.expectEqualStrings("short", fitErr("short", &out)); // Through the tree: a look at a long ./ name not there, a miss, says @@ -525,7 +525,7 @@ test "a reason past 128 bytes keeps its end: the path in it gives up its middle" const p = try th.withFile(testing.allocator, "x\n"); defer p.deinit(); var line: [512]u8 = undefined; - const r = th.wr(p, Node.of(serialOf(p), .look), try std.fmt.bufPrint(&line, "./{s}x.txt\n", .{"no-such-dir/" ** 20})); + const r = th.wr(p, Node.of(serialOf(p), .look), try std.fmt.bufPrint(&line, "./{s}x.txt\n", .{std.mem.asBytes(&@as([20][12]u8, @splat("no-such-dir/".*)))})); try testing.expectEqual(Status.ok, r.reply.status); try testing.expect(th.logHas(p, "x.txt: no such file")); } @@ -1492,17 +1492,17 @@ test "README names every file the tree serves" { return false; } }; - inline for (@typeInfo(TopFile).@"enum".fields) |f| { - const t: TopFile = @enumFromInt(f.value); + inline for (@typeInfo(TopFile).@"enum".field_values) |value| { + const t: TopFile = @enumFromInt(value); if (t != .root and !words.has(t.fileName())) { std.debug.print("README does not mention /{s}\n", .{t.fileName()}); return error.Undocumented; } } - inline for (@typeInfo(PaneFile).@"enum".fields) |f| { - const t: PaneFile = @enumFromInt(f.value); + inline for (@typeInfo(PaneFile).@"enum".field_values) |value| { + const t: PaneFile = @enumFromInt(value); const ok = if (t == .dir) true else if (t.inPty() and t != .pty) - std.mem.indexOf(u8, fs.help, "pty/" ++ comptime @as(PaneFile, @enumFromInt(f.value)).fileName()) != null + std.mem.indexOf(u8, fs.help, "pty/" ++ comptime @as(PaneFile, @enumFromInt(value)).fileName()) != null else words.has(t.fileName()); if (!ok) { diff --git a/src/normal.zig b/src/normal.zig index 77feb7cf..8de23d1c 100644 --- a/src/normal.zig +++ b/src/normal.zig @@ -1015,7 +1015,7 @@ test "gj/gk step the wrapped rows a body draws while j/k keep the file's lines" const p = try Pardes.init(gpa, .{ .tty_only = true }); defer p.deinit(); while (p.nextEffect()) |_| {} - const long = "a" ** 400; + const long = &@as([400]u8, @splat('a')); const pane = try p.setTestFile(long ++ "\nsecond\n"); p.settings.wrap = true; const width = panes.File.wrapWidth(pane, true); diff --git a/src/pager.zig b/src/pager.zig index 30418623..feabd194 100644 --- a/src/pager.zig +++ b/src/pager.zig @@ -38,7 +38,7 @@ pub fn parse(gpa: std.mem.Allocator, raw: []const u8, colour: bool) !Page { errdefer h.text.deinit(gpa); errdefer h.spans.deinit(gpa); try h.text.ensureTotalCapacity(gpa, raw.len); - var stream: ghostty_vt.Stream(*Handler) = .init(&h); + var stream: ghostty_vt.Stream(*Handler) = .init(.{ .handler = &h }); defer stream.deinit(); stream.nextSlice(raw); if (h.failed) return error.OutOfMemory; diff --git a/src/panes.zig b/src/panes.zig index ba1dedb5..07d5ccb1 100644 --- a/src/panes.zig +++ b/src/panes.zig @@ -318,7 +318,7 @@ pub const Pane = struct { /// `older` is an earlier message line still on screen; `index` says /// which (`Pane.msg_older`). pub const Kind = enum { older, message, leader, prompt }; - pub const max = @typeInfo(Kind).@"enum".fields.len - 1 + message_stack; + pub const max = @typeInfo(Kind).@"enum".field_names.len - 1 + message_stack; kinds: [max]Kind = undefined, index: [max]u8 = @splat(0), diff --git a/src/pardes.zig b/src/pardes.zig index 5f330712..8da0922d 100644 --- a/src/pardes.zig +++ b/src/pardes.zig @@ -807,7 +807,7 @@ test "TreeContext tag style preserves message rows and its opt-out survives rest if (comptime !syntax.enabled) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 60, .rows = 16 }); defer p.deinit(); - const pane = try p.setTestFile("pub const Box = struct {\n pub fn run() void {\n" ++ (" // body\n" ** 30) ++ " }\n};\n"); + const pane = try p.setTestFile("pub const Box = struct {\n pub fn run() void {\n" ++ (std.mem.asBytes(&@as([30][16]u8, @splat(" // body\n".*)))) ++ " }\n};\n"); p.gpa.free(pane.file.?.path); pane.file.?.path = try p.gpa.dupe(u8, "/test.zig"); if (!panes.File.supportsContext(pane)) return error.SkipZigTest; @@ -1247,7 +1247,9 @@ test "Def and Refs are a speaking file's tag words; gd, gr and a 2-1 chord all r const r = p.rects[0]; const x0 = r.x + config.GUTTER + panes.File.gutterWidth(pane); const use_y = p.bodyTop(pane, r) + 1; - const tag = try tagline.tagText(p, p.scratch.allocator(), pane); + // Copied out of the scratch arena, which each p.update resets. + const tag = try std.testing.allocator.dupe(u8, try tagline.tagText(p, p.scratch.allocator(), pane)); + defer std.testing.allocator.free(tag); for ([_]struct { word: []const u8, kind: lsp.Kind }{ .{ .word = " Def ", .kind = .definition }, .{ .word = " Refs", .kind = .references } }) |w| { // B1 sweeps the use of `alpha` on the second line (bytes 30..35), // leaving the cursor at its far end. @@ -1348,7 +1350,7 @@ test "acme placement: under a pane's text when it has many blank rows, else halv try std.testing.expectEqual(text_end, p.rects[second].y); // Fill both with text: no blank space is big, so the biggest pane is // halved. - const full = "x\n" ** 80; + const full = std.mem.asBytes(&@as([80][2]u8, @splat("x\n".*))); for ([_]usize{ top, second }) |id| panes.File.setContent(p, &p.panes[id].?.file.?, try p.gpa.dupe(u8, full)); p.sync(); const big = if (p.rects[second].h >= p.rects[top].h) second else top; @@ -2006,7 +2008,7 @@ test "a command pane shows how its command ended, and the next command there run try std.testing.expectEqual(@as(usize, 0), Count.signals); p.host = host_before; // A command line longer than a command pane takes is refused. - const long = "x" ** (exec.command_max + 1); + const long = &@as([exec.command_max + 1]u8, @splat('x')); try std.testing.expectEqual(@as(?usize, null), exec.execute(p, 1, long)); } @@ -2970,9 +2972,9 @@ test "restored terminals keep monotonic nonzero pane identities" { fn chordName(comptime c: config.Chord) []const u8 { comptime { const mods = (if (c.ctrl) "C-" else "") ++ (if (c.alt) "A-" else "") ++ (if (c.shift) "S-" else ""); - for (@typeInfo(Key).@"struct".decls) |d| { - if (@TypeOf(@field(Key, d.name)) != u21) continue; - if (@field(Key, d.name) == c.cp) return mods ++ d.name; + for (@typeInfo(Key).@"struct".decl_names) |d| { + if (@TypeOf(@field(Key, d)) != u21) continue; + if (@field(Key, d) == c.cp) return mods ++ d; } if (c.cp < 0x20 or c.cp > 0x7e) @compileError("chord has no name for the index"); return mods ++ &[_]u8{@as(u8, @intCast(c.cp))}; @@ -3088,11 +3090,11 @@ pub const builtin_rows: [std.enums.values(Builtin).len]Row = blk: { for (std.enums.values(Builtin), 0..) |b, i| { const keys = if (config.leader_path.get(b)) |path| leaderKeys(path) else ""; const rest = shortcuts(b); - const named = @tagName(b) ++ (if (rest.len == 0) "" else (" " ** (namew - @tagName(b).len)) ++ " " ++ rest); + const named = @tagName(b) ++ (if (rest.len == 0) "" else (&@as([namew - @tagName(b).len]u8, @splat(' '))) ++ " " ++ rest); rows[i] = .{ .path = config.leader_path.get(b), .cmd = b, - .line = keys ++ (" " ** (4 + keyw - keys.len)) ++ " " ++ named, + .line = keys ++ (&@as([4 + keyw - keys.len]u8, @splat(' '))) ++ " " ++ named, }; } const last = "\x7f"; @@ -3111,8 +3113,8 @@ comptime { @setEvalBranchQuota(20000); // one branch per string byte, behind the decl walk that folds Builtin var it = std.mem.tokenizeScalar(u8, config.topbar_str, ' '); words: while (it.next()) |w| { - for (@typeInfo(Builtin).@"enum".fields) |f| - if (std.mem.eql(u8, f.name, w)) continue :words; + for (@typeInfo(Builtin).@"enum".field_names) |name| + if (std.mem.eql(u8, name, w)) continue :words; @compileError("topbar word is not a builtin: " ++ w); } } @@ -4499,7 +4501,7 @@ pub const Pardes = struct { /// Panes made this step that found no room in their column /// (exec.placeNew): closed at the step's end, once whatever made them /// is done with them. - unplaced: std.StaticBitSet(MAX_PANES) = .initEmpty(), + unplaced: std.StaticBitSet(MAX_PANES) = .empty, /// A look's word search is running (look.lookUnresolved): from a /// terminal its rows refill the directory's one +Search (Output). word_look: bool = false, @@ -5018,7 +5020,7 @@ pub const Pardes = struct { if (p.dump_out) |d| gpa.free(d); p.fallback.deinit(); if (p.last_dump) |d| gpa.free(d); - if (p.custom_theme) |theme_value| std.zon.parse.free(gpa, theme_value); + if (p.custom_theme) |theme_value| gpa.free(theme_value.name); if (p.chord_arg) |a| gpa.free(a); for (p.spawn_dirs) |d| if (d) |dir| gpa.free(dir); p.pipe.deinit(gpa); @@ -8508,7 +8510,7 @@ test "raw tty Ctrl-B switches to editor mode without sending child input" { test "the jump list never holds two entries in a row the same, when the one between them goes" { const p = try Pardes.init(std.testing.allocator, .{ .cols = 60, .rows = 40 }); defer p.deinit(); - const pane = try p.setTestFile("a\n" ** 200); + const pane = try p.setTestFile(std.mem.asBytes(&@as([200][2]u8, @splat("a\n".*)))); const at: Loc = .{ .pane = 0, .serial = pane.serial, .line = 1, .col = 1 }; // A, a pane long closed with no file to keep it, A again, then far on. p.jumps[0] = at; @@ -9054,7 +9056,7 @@ const JumpsTest = struct { fn init(order: []const u8) !JumpsTest { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 100, .rows = 80 }); errdefer p.deinit(); - const text = "row\n" ** 100; + const text = std.mem.asBytes(&@as([100][4]u8, @splat("row\n".*))); _ = try p.setTestFile(text); var t: JumpsTest = .{ .p = p }; for ([_][]const u8{ "/b.txt", "/test.txt", "" }, 1..) |path, n| { diff --git a/src/pdf.zig b/src/pdf.zig index d98d5ff2..5adc62d1 100644 --- a/src/pdf.zig +++ b/src/pdf.zig @@ -4,9 +4,7 @@ const std = @import("std"); const c_heap = @import("c_heap"); -const c = @cImport({ - @cInclude("pdf_bridge.h"); -}); +const c = @import("pdf_c"); // src/pdf_bridge.h, translated by build.zig /// What MuPDF allocates from, itself and the libraries it bundles (they all /// allocate through its context): `start`'s allocator, through the shared C @@ -1130,7 +1128,7 @@ pub const Document = struct { const page_number = try document.checkedPage(page); if (std.mem.indexOfScalar(u8, needle, 0) != null) return error.InvalidNeedle; - const needle_z = try gpa.dupeZ(u8, needle); + const needle_z = try gpa.dupeSentinel(u8, needle, 0); defer gpa.free(needle_z); var found: c.pardes_pdf_search_result = @@ -2067,7 +2065,7 @@ test "MuPDF search returns normalized oriented quads and word selection text" { const projected = try document.pageText(std.testing.allocator, 0); defer std.testing.allocator.free(projected); - try std.testing.expect(std.ascii.indexOfIgnoreCase(projected, "Pardes") != null); + try std.testing.expect(std.ascii.findIgnoreCase(projected, "Pardes") != null); // Exercise transactional cache replacement before searching page zero // again; a failed ownership handoff here tends to surface as a double drop. if (document.pages > 1) { @@ -2093,7 +2091,7 @@ test "MuPDF search returns normalized oriented quads and word selection text" { const text = try document.copySelection(std.testing.allocator, 0, selected.start, selected.end); defer std.testing.allocator.free(text); - try std.testing.expect(std.ascii.indexOfIgnoreCase(text, "Pardes") != null); + try std.testing.expect(std.ascii.findIgnoreCase(text, "Pardes") != null); } test "highlighted render changes pixels while plain render stays stable" { diff --git a/src/regexp.zig b/src/regexp.zig index 86132f60..4cbca57e 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -58,7 +58,7 @@ pub const Regex = struct { /// Timed on this machine's ReleaseSafe build: 10M steps of the worst /// patterns take 45-80 ms, and a Debug build is ten times slower. - pub const budget: u64 = if (builtin.mode == .Debug) 4_000_000 else 32_000_000; + pub const budget: u64 = if (builtin.mode == .debug) 4_000_000 else 32_000_000; pub const e_long = std.fmt.comptimePrint("bad regular expression: longer than mvzr's {d} operations (about {d} pattern characters)", .{ max_ops, max_ops }); pub const e_anchor = "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n"; @@ -503,10 +503,10 @@ test "a class with non-ASCII runes matches those runes: [éa-z], a range of them } test "a pattern past 64 characters compiles, and one past the limit says so" { - var long = try Regex.compile("a" ** 200); - const text = "x" ++ "a" ** 200 ++ "\n"; + var long = try Regex.compile(&@as([200]u8, @splat('a'))); + const text = "x" ++ &@as([200]u8, @splat('a')) ++ "\n"; try std.testing.expectEqual(@as(usize, 1), (try long.find(text, 0, text.len, text.len)).?.start); - try std.testing.expectError(error.TooLong, Regex.compile("a" ** (max_ops + 8))); + try std.testing.expectError(error.TooLong, Regex.compile(&@as([max_ops + 8]u8, @splat('a')))); try std.testing.expectError(error.Bad, Regex.compile("a[b")); } @@ -546,13 +546,13 @@ test "a search that would backtrack without end runs out of budget, promptly" { // As long a chain as mvzr compiles, over runs one `a` short of it: 2^20 // steps from every start. for (0..line.len / 20) |i| line[i * 20 + 19] = 'b'; - var chain = try Regex.compile("a?" ** 20 ++ "a" ** 20); + var chain = try Regex.compile(std.mem.asBytes(&@as([20][2]u8, @splat("a?".*))) ++ &@as([20]u8, @splat('a'))); try std.testing.expectError(error.TooSlow, chain.find(&line, 0, line.len, line.len)); } test "the match is the leftmost, however long the line" { // A window cut into the identifier would find `x... =` from its middle. - const text = "let " ++ "x" ** 100 ++ " = 1"; + const text = "let " ++ &@as([100]u8, @splat('x')) ++ " = 1"; var assign = try Regex.compile("\\s*(\\w+)\\s*="); const m = (try assign.find(text, 0, text.len, text.len)).?; try std.testing.expectEqual(@as(usize, 3), m.start); diff --git a/src/sam_edit.zig b/src/sam_edit.zig index d0c51bdb..f6253c89 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -1102,7 +1102,7 @@ fn expectEdit(text: []const u8, command: []const u8, want: []const u8) !void { test "an x over 100k lines is 100k searches, each with its own step budget" { const th = @import("ninep/testing.zig"); - const line = "x" ** 80 ++ "\n"; + const line = &@as([80]u8, @splat('x')) ++ "\n"; const text = try std.testing.allocator.alloc(u8, line.len * 100_000); defer std.testing.allocator.free(text); for (0..100_000) |k| @memcpy(text[line.len * k ..][0..line.len], line); diff --git a/src/syntax.zig b/src/syntax.zig index b2f5ea99..fa212ca7 100644 --- a/src/syntax.zig +++ b/src/syntax.zig @@ -1859,7 +1859,7 @@ test "syntax declaration locations clear snapshots without muting ordinary conte defer stop(); const header = "pub fn run() void {"; const content = "| \t" ++ header ++ "\n| \t const value = 42;\nfile.zig:3\t}\n"; - const snapshots = [_]u8{0xff} ** header.len; + const snapshots = @as([header.len]u8, @splat(0xff)); const Row = struct { path: []const u8 = "file.zig", code_start: usize, diff --git a/src/terminal.zig b/src/terminal.zig index 3190c6e3..a8e7931c 100644 --- a/src/terminal.zig +++ b/src/terminal.zig @@ -207,7 +207,7 @@ const LivePalette = struct { var seed = GColor.default; for (base, 0..) |rgb, i| seed[i] = rgb; - self.colors = GColor.generate256Color(seed, .initEmpty(), bg, fg, false); + self.colors = GColor.generate256Color(seed, .empty, bg, fg, false); self.key = key; return &self.colors; } @@ -609,10 +609,10 @@ pub fn create(gpa: std.mem.Allocator, cols: u16, rows: u16) !*Pane { state.vt = try ghostty_vt.Terminal.init(terminalIo(), gpa, .{ .cols = cols, .rows = rows, - .max_scrollback = 16 * 1024 * 1024, + .max_scrollback_bytes = 16 * 1024 * 1024, .kitty_image_storage_limit = if (images_enabled) TerminalImages.storage_limit else 0, }); - state.stream = .initAlloc(gpa, .{ .inner = state.vt.vtHandler() }); + state.stream = .init(.{ .handler = .{ .inner = state.vt.vtHandler() }, .allocator = gpa }); state.replay.head = 0; state.replay.len = 0; state.reply_len = 0; @@ -1055,7 +1055,11 @@ fn ingestWithPositions(pane: *Pane, bytes: []const u8) void { } }; } else failed = true; + // Whether the program cleared the screen (ED 2, a switch of screens); + // pardes reads the flag nowhere else. + pane.terminal.?.vt.flags.dirty.clear = false; ingest(pane, bytes); + const cleared = pane.terminal.?.vt.flags.dirty.clear; // RIS can destroy the alternate screen and every pin it owned. const alive = if (screens.generation(key) == generation) pages else null; var moved = false; @@ -1065,7 +1069,15 @@ fn ingestWithPositions(pane: *Pane, bytes: []const u8) void { if (pin.garbage) pointer_rows_lost = true; }; }; + // The program cleared the screen, or the rows the live copy stood on are + // gone (pruned, a reset): the copy goes with them, as what the program + // drew next. Kept, ghostty would move its rows to the top of what is + // left, over the program's new text. + var copy_lost = cleared or alive == null; for (spans[0..nspans]) |span| { + if (span.overlay == overlay) if (span.start.pin) |pin| if (alive != null and pin.garbage) { + copy_lost = true; + }; const start = pinnedRow(alive, span.start); const rows = @max(1, pinnedRow(alive, span.end) -| start +| 1); moved = moved or start != span.overlay.row or rows != span.overlay.rows; @@ -1096,6 +1108,11 @@ fn ingestWithPositions(pane: *Pane, bytes: []const u8) void { pane.clearPointerSelection(button); pane.sel[button].state = .none; }; + if (copy_lost) if (pane.ovl) |copy| { + pane.gpa.free(copy.text); + pane.ovl = null; + moved = true; + }; if (moved) pane.nsel_snap = 0; // Saved regex-preview offsets name the old flat text. if (failed) { const message = "terminal edit position reset: out of memory"; @@ -1211,7 +1228,7 @@ pub fn commandOutput(pane: *Pane, gpa: std.mem.Allocator, max_rows: usize) !Comm .whole = whole, }; } else true; - return .{ .text = try gpa.dupeZ(u8, ""), .whole = whole }; + return .{ .text = try gpa.dupeSentinel(u8, "", 0), .whole = whole }; } /// At a prompt with nothing typed on its line: no input left of the @@ -2813,7 +2830,7 @@ pub fn redo(p: *Pardes, pane: *Pane) void { pane.ensureCursorVisible(); } -pub fn ptyReport(handler: *ghostty_vt.TerminalStream.Handler, data: [:0]const u8) void { +pub fn ptyReport(handler: *ghostty_vt.TerminalStream.Handler, data: []const u8) void { const state: *State = @alignCast(@fieldParentPtr("vt", handler.terminal)); const room = state.reply.len - state.reply_len; const n = @min(room, data.len); @@ -2968,7 +2985,8 @@ test "a program that asked for focus reports hears CSI I and CSI O as its pane t /// over two cells). inline fn kittyTransmit(comptime extra: []const u8) []const u8 { return comptime blk: { - const pixels = [_]u8{ 255, 0, 0, 255, 0, 0, 0, 0, 255, 0, 0, 255 } ** 2; + const row = [_]u8{ 255, 0, 0, 255, 0, 0, 0, 0, 255, 0, 0, 255 }; + const pixels = row ++ row; var buf: [std.base64.standard.Encoder.calcSize(pixels.len)]u8 = undefined; const b64 = std.base64.standard.Encoder.encode(&buf, &pixels)[0..buf.len].*; break :blk "\x1b_Ga=T,f=24,s=4,v=2,i=7," ++ extra ++ ";" ++ b64 ++ "\x1b\\"; diff --git a/src/tty/tty.zig b/src/tty/tty.zig index ca2cb939..82cb40bd 100644 --- a/src/tty/tty.zig +++ b/src/tty/tty.zig @@ -129,7 +129,7 @@ fn readInput(loop: *Loop, tty: anytype, cache: *vaxis.GraphemeCache) !void { var parser: vaxis.Parser = .{}; var buf: [1024]u8 = undefined; var carried: usize = 0; - while (!loop.should_quit) { + while (true) { // stopInput cancels the task; the read then fails with Canceled if (carried == buf.len) return error.InputSequenceTooLong; const received = try tty.read(buf[carried..]); if (received == 0) return; @@ -295,7 +295,7 @@ test "terminal input preserves fragmented keys queries paste and text after EOF" fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), _: u32) bool { + pub fn inputFollows(self: *@This(), _: u32) bool { return self.next < self.parts.len; } fn read(self: *@This(), buf: []u8) !usize { @@ -380,7 +380,7 @@ test "terminal replies and malformed sequences never become typed text" { fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), _: u32) bool { + pub fn inputFollows(self: *@This(), _: u32) bool { return self.next < self.parts.len; } fn read(self: *@This(), buf: []u8) !usize { @@ -465,7 +465,7 @@ test "terminal input split at every byte with a gap reads as it does whole" { fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), ms: u32) bool { + pub fn inputFollows(self: *@This(), ms: u32) bool { return self.next < self.parts.len and self.gap_ms <= ms; } fn read(self: *@This(), buf: []u8) !usize { @@ -561,7 +561,7 @@ test "an ESC read by itself is the Escape key at once" { fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), _: u32) bool { + pub fn inputFollows(self: *@This(), _: u32) bool { self.asked = true; return self.next < self.parts.len; } @@ -609,7 +609,7 @@ test "a frame's answer is taken wherever the reads cut it, and never typed" { fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), _: u32) bool { + pub fn inputFollows(self: *@This(), _: u32) bool { return self.next < self.parts.len; } fn read(self: *@This(), buf: []u8) !usize { @@ -701,7 +701,7 @@ test "terminal input cancellation joins blocked reads and queued EOF" { fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(_: *@This(), _: u32) bool { + pub fn inputFollows(_: *@This(), _: u32) bool { return false; } fn read(self: *@This(), buf: []u8) !usize { @@ -2332,7 +2332,7 @@ test "a terminal pane gets function keys as their sequences and lone modifiers n fn getWinsize(_: *@This()) !vaxis.Winsize { return .{ .rows = 24, .cols = 80, .x_pixel = 0, .y_pixel = 0 }; } - fn inputFollows(self: *@This(), _: u32) bool { + pub fn inputFollows(self: *@This(), _: u32) bool { return self.next < self.parts.len; } fn read(self: *@This(), buf: []u8) !usize { @@ -3033,7 +3033,7 @@ test "a jump leaves a trail of a few cells behind the cursor that fades in 120 m const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 24 }); defer core.deinit(); - _ = try core.setTestFile("some text on a line\n" ** 20); + _ = try core.setTestFile(std.mem.asBytes(&@as([20][20]u8, @splat("some text on a line\n".*)))); try std.testing.expect(core.executeBuiltinLine(0, "Theme forge")); core.settings.jump_trail = true; for (0..40) |_| core.update(.tick); @@ -3078,7 +3078,7 @@ test "a notice chip casts a cell's shadow right and down, inside its body, off t const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 24 }); defer core.deinit(); - _ = try core.setTestFile("some text on a line\n" ** 20); + _ = try core.setTestFile(std.mem.asBytes(&@as([20][20]u8, @splat("some text on a line\n".*)))); try std.testing.expect(core.executeBuiltinLine(0, "Theme forge")); core.settings.chip_shadow = true; try std.testing.expect(core.executeBuiltinLine(0, "Msg a notice")); @@ -3114,7 +3114,7 @@ test "a scrolled pane's thumb brightens and fades back in 250 ms" { const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 24 }); defer core.deinit(); - const pane = try core.setTestFile("some text on a line\n" ** 200); + const pane = try core.setTestFile(std.mem.asBytes(&@as([200][20]u8, @splat("some text on a line\n".*)))); try std.testing.expect(core.executeBuiltinLine(0, "Theme forge")); core.settings.thumb_flash = true; for (0..40) |_| core.update(.tick); diff --git a/test/deck/replay.zig b/test/deck/replay.zig index 475f7476..49614b46 100644 --- a/test/deck/replay.zig +++ b/test/deck/replay.zig @@ -20,8 +20,9 @@ const Tally = struct { fn replay(src: [:0]const u8) !Tally { const gpa = std.testing.allocator; - const rec = try schema.parse(gpa, src); - defer schema.free(gpa, rec); + var arena: std.heap.ArenaAllocator = .init(gpa); + defer arena.deinit(); + const rec = try schema.parse(gpa, arena.allocator(), src); var d: deck.Deck = .{}; var tally: Tally = .{}; for (rec.events) |ev| { diff --git a/test/deck/schema.zig b/test/deck/schema.zig index 782c829a..401e15e8 100644 --- a/test/deck/schema.zig +++ b/test/deck/schema.zig @@ -61,10 +61,8 @@ pub const Event = struct { }; }; -pub fn parse(gpa: std.mem.Allocator, src: [:0]const u8) error{ OutOfMemory, ParseZon }!Recording { - return std.zon.parse.fromSliceAlloc(Recording, gpa, src, null, .{}); -} - -pub fn free(gpa: std.mem.Allocator, rec: Recording) void { - std.zon.parse.free(gpa, rec); +/// The recording's slices live in `arena`. +pub fn parse(gpa: std.mem.Allocator, arena: std.mem.Allocator, src: [:0]const u8) error{ OutOfMemory, ParseZon }!Recording { + var diagnostics: std.zon.parse.Diagnostics = undefined; + return std.zon.parse.fromSlice(Recording, .{ .gpa = gpa, .arena = arena, .source = src, .diagnostics = &diagnostics }); } diff --git a/test/e2e_harness.zig b/test/e2e_harness.zig index 9e67c49f..a2010c19 100644 --- a/test/e2e_harness.zig +++ b/test/e2e_harness.zig @@ -104,7 +104,7 @@ pub const Harness = struct { term.* = try ghostty_vt.Terminal.init(std.Io.Threaded.global_single_threaded.io(), gpa, .{ .cols = cols, .rows = rows, - .max_scrollback = 1024 * 1024, + .max_scrollback_bytes = 1024 * 1024, }); errdefer term.deinit(gpa); // vtStream() captures a pointer to `term`; term is heap-allocated so the diff --git a/test/fs_bench.zig b/test/fs_bench.zig index 746fe737..4764049f 100644 --- a/test/fs_bench.zig +++ b/test/fs_bench.zig @@ -272,7 +272,7 @@ pub fn main(init: std.process.Init) !void { .tag = 9, .op = .write, .node = s.node(.body), - .data = "x" ** 1024, + .data = &@as([1024]u8, @splat('x')), .size = 1024, }, "append: whole-body swap + undo snapshot"); @@ -327,7 +327,7 @@ pub fn main(init: std.process.Init) !void { try w.writeAll("]}\n"); } else { try w.print("{s:<26} {s:>9} {s:>8} {s}\n", .{ "request", "ns/op", "allocs", "note" }); - try w.print("{s:<26} {s:>9} {s:>8} {s}\n", .{ "-" ** 26, "-" ** 9, "-" ** 8, "-" ** 20 }); + try w.print("{s:<26} {s:>9} {s:>8} {s}\n", .{ &@as([26]u8, @splat('-')), &@as([9]u8, @splat('-')), &@as([8]u8, @splat('-')), &@as([20]u8, @splat('-')) }); for (rows.items) |r| try w.print("{s:<26} {d:>9} {d:>8} {s}\n", .{ r.name, r.ns, r.allocs, r.note }); } diff --git a/test/history.zig b/test/history.zig index 5ce45239..d399f7aa 100644 --- a/test/history.zig +++ b/test/history.zig @@ -432,24 +432,24 @@ test "history benchmark cases are nonempty unique and matched by name" { try std.testing.expectError(error.MissingField, readBenchmarks(arena, "{\"case\":\"first\"}")); const before = try readBenchmarks(arena, benchmark_fixture); const missing = try readBenchmarks(arena, "{\"case\":\"first\",\"bytes\":10,\"median_ns\":100,\"cold_ns\":100,\"allocations\":10,\"allocated_bytes\":1000}"); - try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(.{before} ** 4, .{missing} ** 4, null, &report.writer)); + try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(@splat(before), @splat(missing), null, &report.writer)); const renamed = \\{"case":"first","bytes":10,"median_ns":100,"cold_ns":100,"allocations":10,"allocated_bytes":1000} \\{"case":"renamed","bytes":20,"median_ns":100,"cold_ns":100,"allocations":10,"allocated_bytes":1000} ; const different = try readBenchmarks(arena, renamed); - try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(.{before} ** 4, .{different} ** 4, null, &report.writer)); + try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(@splat(before), @splat(different), null, &report.writer)); const reordered = \\{"case":"last","bytes":20,"median_ns":100,"cold_ns":100,"allocations":10,"allocated_bytes":1000} \\{"case":"first","bytes":10,"median_ns":100,"cold_ns":100,"allocations":10,"allocated_bytes":1000} ; const same = try readBenchmarks(arena, reordered); - try std.testing.expect(!try compareBenchmarks(.{before} ** 4, .{same} ** 4, 1, &report.writer)); + try std.testing.expect(!try compareBenchmarks(@splat(before), @splat(same), 1, &report.writer)); var samples = [4][]Benchmark{ before, before, before, different }; - try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(.{before} ** 4, samples, null, &report.writer)); + try std.testing.expectError(error.DifferentBenchmarks, compareBenchmarks(@splat(before), samples, null, &report.writer)); samples[3] = try arena.dupe(Benchmark, before); samples[3][0].bytes += 1; - try std.testing.expectError(error.DifferentBenchmarkInputs, compareBenchmarks(.{before} ** 4, samples, null, &report.writer)); + try std.testing.expectError(error.DifferentBenchmarkInputs, compareBenchmarks(@splat(before), samples, null, &report.writer)); } test "history benchmark gate uses repeated process captures rather than the first process alone" { diff --git a/test/hxdiff.zig b/test/hxdiff.zig index 48c7ab93..b12bcd8e 100644 --- a/test/hxdiff.zig +++ b/test/hxdiff.zig @@ -39,7 +39,7 @@ pub fn main(init: std.process.Init) !void { var pane_override: ?[]const u8 = null; var goldens_path: ?[]const u8 = null; var waivers_path: ?[]const u8 = null; - const paths = try arena.alloc([]const u8, args.len - 1); + const paths = try arena.alloc([]const u8, @max(args.len - 1, 3)); // room for the defaults below var n_paths: usize = 0; var want_waivers = false; for (args[1..]) |a| { @@ -67,6 +67,18 @@ pub fn main(init: std.process.Init) !void { n_paths += 1; } } + // Named nothing (`zig build hxdiff` or `hxparity` bare): the checked-in + // suite, run from the repository root. + if (n_paths == 0 and !want_waivers) { + strict = true; + const defaults: []const []const u8 = if (parity) + &.{ "test/hxcases/cases.jsonl", "test/hxcases/parity.jsonl" } + else + &.{ "test/hxcases/cases.jsonl", "test/hxcases/goldens.jsonl", "test/hxcases/waivers.jsonl" }; + if (parity) waivers_path = waivers_path orelse "test/hxcases/parity-waivers.jsonl"; + @memcpy(paths[0..defaults.len], defaults); + n_paths = defaults.len; + } if (n_paths == 0 or want_waivers) fatal("usage: pardes-hxdiff [--parity] ... [goldens.jsonl [waivers.jsonl]]", .{}); const case_paths = paths[0..if (parity) n_paths else 1]; diff --git a/test/image_harness.zig b/test/image_harness.zig index 474ec48b..b45851cc 100644 --- a/test/image_harness.zig +++ b/test/image_harness.zig @@ -38,7 +38,7 @@ pub fn main(init: std.process.Init) !void { const args = try init.minimal.args.toSlice(arena); if (args.len != 3 and args.len != 4) return error.BadArgs; - const exe = try arena.dupeZ(u8, args[1]); + const exe = try arena.dupeSentinel(u8, args[1], 0); const backend = args[2]; const pdf_mode = args.len == 4 and std.mem.eql(u8, args[3], "pdf"); if (args.len == 4 and !pdf_mode) return error.BadArgs; diff --git a/test/lspbench.zig b/test/lspbench.zig index 02138d24..d0e25aee 100644 --- a/test/lspbench.zig +++ b/test/lspbench.zig @@ -216,7 +216,7 @@ fn reportText(rs: []const Result, rss: u64) void { const o = std.debug.print; o("backend: {s}\n\n", .{lsp.backend_name}); o("{s:<24} {s:>9} {s:>7} {s:>10} {s:>10} {s}\n", .{ "kind", "supports", "rows", "cold(us)", "warm(us)", "verdict" }); - o("{s}\n", .{"-" ** 78}); + o("{s}\n", .{&@as([78]u8, @splat('-'))}); var claimed: usize = 0; var working: usize = 0; var lies: usize = 0; diff --git a/test/monkey.zig b/test/monkey.zig index 83869ce4..dcea0b43 100644 --- a/test/monkey.zig +++ b/test/monkey.zig @@ -44,7 +44,7 @@ fn generate(gpa: std.mem.Allocator, seed: u64, steps: usize) ![]u8 { var s: Script = .{ .gpa = gpa }; try s.line("# pardes-monkey seed {d}, {d} steps", .{ seed, steps }); try s.line("file bad.txt hello \\xff\\xfe world\\nlatin-1 caf\\xe9 tail\\ntruncated \\xe6\\x97 here\\nend\\n", .{}); - try s.line("file wide.txt {s}tail\\nsecond line\\n", .{"\\xe6\\x97\\xa5\\xe6\\x9c\\xac\\xe8\\xaa\\x9e " ** 8}); + try s.line("file wide.txt {s}tail\\nsecond line\\n", .{std.mem.asBytes(&@as([8][37]u8, @splat("\\xe6\\x97\\xa5\\xe6\\x9c\\xac\\xe8\\xaa\\x9e ".*)))}); try s.line("lines long.txt 400 some words after the number, long enough to wrap in a narrow pane", .{}); try s.line("start 24 80 bad.txt", .{}); try s.line("wait 8000 Newcol Joincol", .{}); diff --git a/test/panes.zig b/test/panes.zig index 86e52838..54100f91 100644 --- a/test/panes.zig +++ b/test/panes.zig @@ -42,7 +42,7 @@ const TagLayerTests = struct { p.row_metrics = .{ .body_w = 10, .body_h = 20, .tagline_w = 5, .tagline_h = 10 }; p.gpa.free(pane.file.?.path); pane.file.?.path = try p.gpa.dupe(u8, "/x"); - try setTag(pane, " " ++ "a" ** 38 ++ "\t界 END"); + try setTag(pane, " " ++ &@as([38]u8, @splat('a')) ++ "\t界 END"); var frame = std.heap.ArenaAllocator.init(p.gpa); defer frame.deinit(); _ = try p.render(frame.allocator()); @@ -135,7 +135,7 @@ const TagLayerTests = struct { try std.testing.expectEqual(@as(i32, 0), p.panes[second].?.input.cur_col); p.update(.{ .key = .{ .cp = Key.escape } }); const left = p.panes[0].?; - try setTag(left, " " ++ "a" ** 110); + try setTag(left, " " ++ &@as([110]u8, @splat('a'))); _ = try p.render(frame.allocator()); const left_layer = try findLayer(p, .pane, 0); const left_text = try pardes.tagline.tagText(p, frame.allocator(), left); @@ -149,7 +149,7 @@ const TagLayerTests = struct { try std.testing.expectEqual(panes.File.rawAtDisplay(left_text, end_col), @as(usize, @intCast(left.tag.cur_col))); try std.testing.expect(left.sel[0].state == .done and left.sel[0].r0 == 0); p.update(.{ .key = .{ .cp = Key.escape } }); - const long = "a" ** 60 ++ "界END"; + const long = &@as([60]u8, @splat('a')) ++ "界END"; p.global_tag.own = try p.gpa.dupe(u8, long); for (p.col_tags[0..p.ncol]) |*tag| tag.own = try p.gpa.dupe(u8, long); _ = try p.render(frame.allocator()); @@ -581,7 +581,7 @@ const BodyHitTests = struct { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 64, .rows = 18 }); defer p.deinit(); p.row_metrics = .{ .body_w = 10, .body_h = 20, .tagline_w = 8, .tagline_h = 10 }; - const pane = try TreeContextTests.setZigFile(p, "const Box = struct {\n pub fn work(\n value: usize,\n ) void {\n" ++ " // body\n" ** 60 ++ " }\n};\n"); + const pane = try TreeContextTests.setZigFile(p, "const Box = struct {\n pub fn work(\n value: usize,\n ) void {\n" ++ std.mem.asBytes(&@as([60][16]u8, @splat(" // body\n".*))) ++ " }\n};\n"); try std.testing.expect(p.executeBuiltinLine(0, "TreeContext")); pane.scrollBy(10); var frame = std.heap.ArenaAllocator.init(p.gpa); @@ -629,7 +629,7 @@ const LineNumberLookTests = struct { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 64, .rows = 18 }); defer p.deinit(); p.row_metrics = .{ .body_w = 10, .body_h = 20, .tagline_w = 8, .tagline_h = 10 }; - const pane = try TreeContextTests.setZigFile(p, "const Box = struct {\n pub fn work(\n value: usize,\n ) void {\n" ++ " // body\n" ** 60 ++ " }\n};\n"); + const pane = try TreeContextTests.setZigFile(p, "const Box = struct {\n pub fn work(\n value: usize,\n ) void {\n" ++ std.mem.asBytes(&@as([60][16]u8, @splat(" // body\n".*))) ++ " }\n};\n"); try std.testing.expect(p.executeBuiltinLine(0, "TreeContext")); pane.scrollBy(30); var frame = std.heap.ArenaAllocator.init(p.gpa); @@ -686,7 +686,7 @@ const TreeContextTests = struct { if (!syntax.enabled) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 64, .rows = 18 }); defer p.deinit(); - const pane = try setZigFile(p, "const Box = struct {\n\n // omitted member comment\n\n pub fn work(\n value: usize,\n ) void {\n" ++ " // body\n" ** 30 ++ " }\n};\n"); + const pane = try setZigFile(p, "const Box = struct {\n\n // omitted member comment\n\n pub fn work(\n value: usize,\n ) void {\n" ++ std.mem.asBytes(&@as([30][16]u8, @splat(" // body\n".*))) ++ " }\n};\n"); p.row_metrics = .{ .body_w = 10, .body_h = 20, .tagline_w = 8, .tagline_h = 10 }; try std.testing.expect(p.executeBuiltinLine(0, "TreeContext")); pane.scrollBy(10); @@ -717,7 +717,7 @@ const TreeContextTests = struct { if (context and !syntax.enabled) continue; const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 64, .rows = 18 }); defer p.deinit(); - const pane = try setZigFile(p, "pub fn work() void {\n" ++ " // body\n" ** 100 ++ "}\n"); + const pane = try setZigFile(p, "pub fn work() void {\n" ++ std.mem.asBytes(&@as([100][12]u8, @splat(" // body\n".*))) ++ "}\n"); if (context) try std.testing.expect(p.executeBuiltinLine(0, "TreeContext")); var frame = std.heap.ArenaAllocator.init(p.gpa); defer frame.deinit(); @@ -792,7 +792,7 @@ const TreeContextTests = struct { if (!syntax.enabled) return error.SkipZigTest; const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 48, .rows = 14 }); defer p.deinit(); - const pane = try setZigFile(p, "pub fn @\"界name\"(\n\tvalue: usize,\n) void {\n" ++ " _ = value;\n" ** 12 ++ "}\n"); + const pane = try setZigFile(p, "pub fn @\"界name\"(\n\tvalue: usize,\n) void {\n" ++ std.mem.asBytes(&@as([12][15]u8, @splat(" _ = value;\n".*))) ++ "}\n"); pane.tag_expanded = false; // a one-row tag keeps this small screen's rows p.sync(); p.settings.wrap = false; @@ -820,7 +820,7 @@ const TreeContextTests = struct { // Read the body top again: the click cleared the announcement band the // builtin above put up, so the body moved with it. try std.testing.expectEqual(p.bodyTop(pane, p.rects[0]), cursor.y); - panes.File.setContent(p, &pane.file.?, try p.gpa.dupe(u8, "// no enclosing declaration\n" ** 18)); + panes.File.setContent(p, &pane.file.?, try p.gpa.dupe(u8, std.mem.asBytes(&@as([18][28]u8, @splat("// no enclosing declaration\n".*))))); _ = try p.render(frame.allocator()); try std.testing.expectEqual(@as(u16, 0), pane.context_rows); try std.testing.expectEqual(pane.scroll(), pane.wrapAt(0).line); @@ -942,7 +942,7 @@ const TtySelectionTests = struct { try noChildInput(p); _ = try p.render(frame.allocator()); try std.testing.expectEqualStrings(expected, pardes.test_api.heldSelection(p, 0).?); - p.update(.{ .output = .{ .pane = 0, .bytes = "\x1b]133;C\x07" ++ "\r\nnew output" ** 30 } }); + p.update(.{ .output = .{ .pane = 0, .bytes = "\x1b]133;C\x07" ++ std.mem.asBytes(&@as([30][12]u8, @splat("\r\nnew output".*))) } }); _ = try p.render(frame.allocator()); try std.testing.expectEqual(panes.Text.Mode.tty, pane.body.mode); try std.testing.expectEqualStrings(expected, pardes.test_api.heldSelection(p, 0).?); @@ -1266,7 +1266,7 @@ const ClickCursorTests = struct { test "click operands and hover follow the scrolled source selection" { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 40, .rows = 14 }); defer p.deinit(); - const pane = try p.setTestFile("zero\none\ntwo\nthree\nalpha beta\nfive\ngamma delta\nseven\n" ++ "tail\n" ** 20); + const pane = try p.setTestFile("zero\none\ntwo\nthree\nalpha beta\nfive\ngamma delta\nseven\n" ++ std.mem.asBytes(&@as([20][5]u8, @splat("tail\n".*)))); p.presentation.enabled = false; p.settings.wrap = false; var frame = std.heap.ArenaAllocator.init(p.gpa); @@ -1323,7 +1323,7 @@ const JumpSelectionTests = struct { const p = try Pardes.init(std.testing.allocator, .{ .tty_only = true, .cols = 34, .rows = 10 }); defer p.deinit(); p.settings.wrap = wrap; - const text = try std.fmt.allocPrint(p.gpa, "{s}{s}", .{ prefix, "unrelated destination text\n" ** 80 }); + const text = try std.fmt.allocPrint(p.gpa, "{s}{s}", .{ prefix, std.mem.asBytes(&@as([80][27]u8, @splat("unrelated destination text\n".*))) }); defer p.gpa.free(text); const pane = try p.setTestFile(text); var frame = std.heap.ArenaAllocator.init(p.gpa); @@ -1813,7 +1813,7 @@ test "a word the tag's wrapping breaks across rows is still one word to a click" var pad: usize = 1; const shown = while (pad < 80) : (pad += 1) { p.gpa.free(pane.file.?.path); - const path = try std.fmt.allocPrint(arena.allocator(), "/{s}/f.zig", .{("a" ** 80)[0..pad]}); + const path = try std.fmt.allocPrint(arena.allocator(), "/{s}/f.zig", .{(&@as([80]u8, @splat('a')))[0..pad]}); pane.file.?.path = try p.gpa.dupe(u8, path); p.sync(); const s = try pardes.tagline.shownTag(p, arena.allocator(), pane); @@ -1866,7 +1866,7 @@ test "terminal output keeps following new rows after scrollback eviction" { pane.body.mode = .tty; pane.tty_filter = false; const pages = &pane.terminal.?.vt.screens.active.pages; - pages.explicit_max_size = 64 * 1024; + pages.limits.set(.bytes, 64 * 1024); var frame = std.heap.ArenaAllocator.init(gpa); defer frame.deinit(); var evictions: usize = 0; @@ -1885,7 +1885,7 @@ test "terminal output keeps following new rows after scrollback eviction" { if (pages.scrollbar().total < previous) evictions += 1; } try std.testing.expect(evictions > phase); - try std.testing.expect(pages.page_size <= pages.maxSize()); + try std.testing.expect(pages.page_size <= pages.limits.max(.bytes)); const sb = pages.scrollbar(); try std.testing.expectEqual(sb.total - pane.rows, sb.offset); const full = try panes.terminal.screenTextAlloc(pane, gpa); @@ -1926,7 +1926,7 @@ test "terminal edit stays on its surviving row when old scrollback is evicted" { const pane = p.panes[0].?; pane.body.mode = .tty; const pages = &pane.terminal.?.vt.screens.active.pages; - pages.explicit_max_size = 64 * 1024; + pages.limits.set(.bytes, 64 * 1024); var next: usize = 0; while (next < 3000) { var buf: [32]u8 = undefined; @@ -1977,9 +1977,9 @@ test "terminal edits and undo survive partial history clearing and whole-screen pane.body.mode = .tty; if (action == .reset_alternate) p.update(.{ .output = .{ .pane = 0, .bytes = "\x1b[?1049h" } }); - p.update(.{ .output = .{ .pane = 0, .bytes = "old row\r\n" ** 40 } }); + p.update(.{ .output = .{ .pane = 0, .bytes = std.mem.asBytes(&@as([40][9]u8, @splat("old row\r\n".*))) } }); const pages = &pane.terminal.?.vt.screens.active.pages; - pages.explicit_max_size = 64 * 1024; + pages.limits.set(.bytes, 64 * 1024); const row: i32 = @as(i32, @intCast(pages.total_rows - pane.rows)) + @as(i32, if (action == .partial_clear_history) -1 else 2); pane.body.mode = .normal; @@ -2005,7 +2005,7 @@ test "terminal edits and undo survive partial history clearing and whole-screen const bytes = switch (action) { .clear_history, .partial_clear_history => "\x1b[3J", .reset_primary, .reset_alternate => "\x1bc", - .evict_all, .hidden_undo => "new row\r\n" ** 4000, + .evict_all, .hidden_undo => std.mem.asBytes(&@as([4000][9]u8, @splat("new row\r\n".*))), }; p.update(.{ .output = .{ .pane = 0, .bytes = bytes } }); const expected: i32 = if (action == .clear_history) 2 else 0; @@ -2045,7 +2045,7 @@ test "terminal lower scroll regions do not move edits above them" { var setup: [64]u8 = undefined; const bytes = try std.fmt.bufPrint(&setup, "\x1b[4;{d}r\x1b[{d};1H", .{ pane.rows, pane.rows }); p.update(.{ .output = .{ .pane = 0, .bytes = bytes } }); - p.update(.{ .output = .{ .pane = 0, .bytes = "tail\r\n" ** 50 } }); + p.update(.{ .output = .{ .pane = 0, .bytes = std.mem.asBytes(&@as([50][6]u8, @splat("tail\r\n".*))) } }); try std.testing.expectEqual(total, panes.terminal.scrollbar(pane).total); try std.testing.expectEqual(@as(i32, 1), pane.ovl.?.row); try std.testing.expectEqual(@as(i32, 1), pane.body.cur_row); @@ -2064,7 +2064,7 @@ test "terminal eviction and lower-region scrolling in one read preserve a surviv const pane = p.panes[0].?; pane.body.mode = .tty; const pages = &pane.terminal.?.vt.screens.active.pages; - pages.explicit_max_size = 64 * 1024; + pages.limits.set(.bytes, 64 * 1024); var next: usize = 0; while (next < 3000) { var buf: [32]u8 = undefined; @@ -2229,7 +2229,7 @@ test "prompt cancellation restores the selection before regex previews" { } test "a prompt that cannot be opened leaves the open one as it was" { - const Request = @typeInfo(@TypeOf(Pardes.startPrompt)).@"fn".params[3].type.?; + const Request = @typeInfo(@TypeOf(Pardes.startPrompt)).@"fn".param_types[3].?; for ([_]Request{ .{ .search = config.search_marker }, .save, .{ .pipe = .replace }, .{ .pipe = .ignore }, .{ .pipe = .insert }, .{ .pipe = .append } }) |request| { var failing = std.testing.FailingAllocator.init(std.testing.allocator, .{}); const p = try Pardes.init(failing.allocator(), .{ .tty_only = true }); @@ -2268,7 +2268,7 @@ test "Save's prompt starts from the pane's directory" { } test "a prompt opens with one allocation" { - const Request = @typeInfo(@TypeOf(Pardes.startPrompt)).@"fn".params[3].type.?; + const Request = @typeInfo(@TypeOf(Pardes.startPrompt)).@"fn".param_types[3].?; for ([_]Request{ .{ .search = config.search_marker }, .save, .{ .pipe = .replace }, .{ .pipe = .ignore }, .{ .pipe = .insert }, .{ .pipe = .append } }) |request| { var failing = std.testing.FailingAllocator.init(std.testing.allocator, .{}); const p = try Pardes.init(failing.allocator(), .{ .tty_only = true }); @@ -3441,7 +3441,7 @@ const PdfTests = struct { defer probe.deinit(std.testing.allocator); try std.testing.expectEqual(@as(usize, 0), probe.page); try std.testing.expect(probe.quads.len > 0); - try std.testing.expect(std.ascii.indexOfIgnoreCase(probe.text, "Pardes") != null); + try std.testing.expect(std.ascii.findIgnoreCase(probe.text, "Pardes") != null); } test "PDF point mapping restores band origin and fractional placement" { @@ -4812,7 +4812,7 @@ test "growing a file pane colors every newly visible Zig keyword" { const gpa = std.testing.allocator; const p = try Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 8 }); defer p.deinit(); - const pane = try p.setTestFile("pub fn f() void { if (true) {} }\n" ** 100); + const pane = try p.setTestFile(std.mem.asBytes(&@as([100][33]u8, @splat("pub fn f() void { if (true) {} }\n".*)))); const file = &pane.file.?; gpa.free(file.path); file.path = try gpa.dupe(u8, "/resize.zig"); diff --git a/test/pdf.zig b/test/pdf.zig index a0eb8154..a51fd1a9 100644 --- a/test/pdf.zig +++ b/test/pdf.zig @@ -2080,14 +2080,14 @@ test "PDF native mouse selection, Look, and highlights share page geometry" { @as(i32, r.y + r.h - 1), )); if (!panes.Pdf.beginSelection(p, pane, col, row)) continue; - if (std.ascii.indexOfIgnoreCase(pv.selection_text, "Pardes") != null) { + if (std.ascii.findIgnoreCase(pv.selection_text, "Pardes") != null) { selected_col = col; selected_row = row; break :find_word; } }; const word_col = selected_col orelse return error.PdfMouseMappingMissedWord; - try std.testing.expect(std.ascii.indexOfIgnoreCase( + try std.testing.expect(std.ascii.findIgnoreCase( pardes.test_api.heldSelection(p, 0).?, "Pardes", ) != null); @@ -2142,7 +2142,7 @@ test "PDF native mouse selection, Look, and highlights share page geometry" { const hover = p.pdf_hover_preview orelse return error.MissingPdfHoverPreview; try std.testing.expectEqual(@as(usize, 0), hover.probe.page); try std.testing.expect(hover.probe.quads.len > 0); - try std.testing.expect(std.ascii.indexOfIgnoreCase(hover.probe.text, "Pardes") != null); + try std.testing.expect(std.ascii.findIgnoreCase(hover.probe.text, "Pardes") != null); try std.testing.expect(p.look_hover_preview == null); try std.testing.expect(p.look_hover_wait == null); @@ -2188,7 +2188,7 @@ test "PDF native mouse selection, Look, and highlights share page geometry" { // search highlights without carrying an old selection into the result. p.update(.{ .mouse = .{ .button = config.look_button, .kind = .press, .col = word_col, .row = selected_row } }); p.update(.{ .mouse = .{ .button = config.look_button, .kind = .release, .col = word_col, .row = selected_row } }); - try std.testing.expect(std.ascii.indexOfIgnoreCase(pv.search_query, "Pardes") != null); + try std.testing.expect(std.ascii.findIgnoreCase(pv.search_query, "Pardes") != null); pv.resolveSearch(p.pdf_gpa); const highlights = (try panes.Pdf.buildHighlights( pv, diff --git a/test/pdf_scroll_bench.zig b/test/pdf_scroll_bench.zig index 166fa809..fdc42660 100644 --- a/test/pdf_scroll_bench.zig +++ b/test/pdf_scroll_bench.zig @@ -597,7 +597,7 @@ fn reportText(config: Config, travel: Travel, results: []const Result) void { "total_us", "rasters", "kept", "peak_kib", "visual", "transport", }); - std.debug.print("{s}\n", .{"-" ** 160}); + std.debug.print("{s}\n", .{&@as([160]u8, @splat('-'))}); for (results) |result| std.debug.print( "{s:<16} {d:>6} {d:>7} {d:>10} {d:>10} {d:>10} {d:>10} {d:>10} {d:>11} {d:>7} {d:>5} {d:>9} {x:0>16} {x:0>16}\n", .{ diff --git a/test/pdf_sections_bench.zig b/test/pdf_sections_bench.zig index 92c9ef7c..ee26e83d 100644 --- a/test/pdf_sections_bench.zig +++ b/test/pdf_sections_bench.zig @@ -454,7 +454,7 @@ fn reportText(config: Config, results: []const Result) void { std.debug.print("{s:<42} {s:>10} {s:>10} {s:>10} {s:>10} {s:>8} {s:>8} {s}\n", .{ "scenario", "min", "median", "p90", "max", "items", "bytes", "checksum", }); - std.debug.print("{s}\n", .{"-" ** 124}); + std.debug.print("{s}\n", .{&@as([124]u8, @splat('-'))}); for (results) |result| std.debug.print( "{s:<42} {d:>10} {d:>10} {d:>10} {d:>10} {d:>8} {d:>8} {x:0>16}\n", .{ diff --git a/test/perf-baseline-gui-Debug.json b/test/perf-baseline-gui-Debug.json deleted file mode 100644 index 39f5913c..00000000 --- a/test/perf-baseline-gui-Debug.json +++ /dev/null @@ -1 +0,0 @@ -{"harness":"303736f83a625e87","build":{"compiler":"0.16.0","target":"x86_64-linux.5.10...6.19-gnu.2.38","cpu":"x86_64","driver_optimize":"ReleaseFast","core_optimize":"Debug","c_optimize":"ReleaseFast","terminal_optimize":"ReleaseSafe","platform":"gui","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1818,"med_us":2104,"p90_us":2287,"max_us":2293},{"op":"render","fixture":"medium","min_us":78,"med_us":82,"p90_us":87,"max_us":87},{"op":"key-down","fixture":"medium","min_us":79,"med_us":81,"p90_us":89,"max_us":109},{"op":"key-right","fixture":"medium","min_us":79,"med_us":84,"p90_us":85,"max_us":89},{"op":"page-down","fixture":"medium","min_us":79,"med_us":81,"p90_us":84,"max_us":87},{"op":"wheel","fixture":"medium","min_us":77,"med_us":78,"p90_us":82,"max_us":82},{"op":"edit-char","fixture":"medium","min_us":1019,"med_us":1349,"p90_us":1597,"max_us":1632},{"op":"edit-twin","fixture":"medium","min_us":2029,"med_us":2353,"p90_us":2455,"max_us":2676},{"op":"body-2m","fixture":"medium","min_us":5978,"med_us":9708,"p90_us":11361,"max_us":11666},{"op":"body-appends","fixture":"medium","min_us":117933,"med_us":211247,"p90_us":268014,"max_us":275006},{"op":"deep-sel","fixture":"medium","min_us":73,"med_us":75,"p90_us":80,"max_us":84}]} diff --git a/test/perf-baseline-gui-debug.json b/test/perf-baseline-gui-debug.json new file mode 100644 index 00000000..4b60520b --- /dev/null +++ b/test/perf-baseline-gui-debug.json @@ -0,0 +1 @@ +{"harness":"6cf9f4b8943e7e29","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"gui","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1732,"med_us":1881,"p90_us":2035,"max_us":2072},{"op":"render","fixture":"medium","min_us":89,"med_us":92,"p90_us":101,"max_us":102},{"op":"key-down","fixture":"medium","min_us":92,"med_us":94,"p90_us":100,"max_us":100},{"op":"key-right","fixture":"medium","min_us":91,"med_us":93,"p90_us":105,"max_us":109},{"op":"page-down","fixture":"medium","min_us":91,"med_us":95,"p90_us":99,"max_us":106},{"op":"wheel","fixture":"medium","min_us":90,"med_us":95,"p90_us":101,"max_us":101},{"op":"edit-char","fixture":"medium","min_us":1226,"med_us":1678,"p90_us":2093,"max_us":2323},{"op":"edit-twin","fixture":"medium","min_us":2275,"med_us":2697,"p90_us":2868,"max_us":2920},{"op":"body-2m","fixture":"medium","min_us":5213,"med_us":11333,"p90_us":18974,"max_us":18999},{"op":"body-appends","fixture":"medium","min_us":175012,"med_us":260087,"p90_us":317184,"max_us":317469},{"op":"deep-sel","fixture":"medium","min_us":83,"med_us":85,"p90_us":86,"max_us":98}]} diff --git a/test/perf-baseline-tty-Debug.json b/test/perf-baseline-tty-Debug.json deleted file mode 100644 index 9110398c..00000000 --- a/test/perf-baseline-tty-Debug.json +++ /dev/null @@ -1 +0,0 @@ -{"harness":"303736f83a625e87","build":{"compiler":"0.16.0","target":"x86_64-linux.5.10...6.19-gnu.2.38","cpu":"x86_64","driver_optimize":"ReleaseFast","core_optimize":"Debug","c_optimize":"ReleaseFast","terminal_optimize":"ReleaseSafe","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1881,"med_us":2148,"p90_us":2324,"max_us":2644},{"op":"render","fixture":"medium","min_us":79,"med_us":87,"p90_us":90,"max_us":93},{"op":"key-down","fixture":"medium","min_us":81,"med_us":84,"p90_us":86,"max_us":90},{"op":"key-right","fixture":"medium","min_us":80,"med_us":85,"p90_us":89,"max_us":102},{"op":"page-down","fixture":"medium","min_us":85,"med_us":86,"p90_us":88,"max_us":88},{"op":"wheel","fixture":"medium","min_us":79,"med_us":85,"p90_us":90,"max_us":91},{"op":"edit-char","fixture":"medium","min_us":1037,"med_us":1300,"p90_us":1639,"max_us":1647},{"op":"edit-twin","fixture":"medium","min_us":2046,"med_us":2378,"p90_us":2529,"max_us":2585},{"op":"body-2m","fixture":"medium","min_us":5790,"med_us":9510,"p90_us":12226,"max_us":12462},{"op":"body-appends","fixture":"medium","min_us":132421,"med_us":224849,"p90_us":257368,"max_us":274890},{"op":"deep-sel","fixture":"medium","min_us":75,"med_us":80,"p90_us":84,"max_us":86}]} diff --git a/test/perf-baseline-tty-ReleaseFast.json b/test/perf-baseline-tty-ReleaseFast.json deleted file mode 100644 index 4cbc8b95..00000000 --- a/test/perf-baseline-tty-ReleaseFast.json +++ /dev/null @@ -1 +0,0 @@ -{"harness":"303736f83a625e87","build":{"compiler":"0.16.0","target":"x86_64-linux.5.10...6.19-gnu.2.38","cpu":"x86_64","driver_optimize":"ReleaseFast","core_optimize":"ReleaseFast","c_optimize":"ReleaseFast","terminal_optimize":"ReleaseFast","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1775,"med_us":1995,"p90_us":2092,"max_us":2094},{"op":"render","fixture":"medium","min_us":80,"med_us":86,"p90_us":89,"max_us":95},{"op":"key-down","fixture":"medium","min_us":81,"med_us":83,"p90_us":95,"max_us":101},{"op":"key-right","fixture":"medium","min_us":79,"med_us":84,"p90_us":85,"max_us":98},{"op":"page-down","fixture":"medium","min_us":80,"med_us":82,"p90_us":96,"max_us":97},{"op":"wheel","fixture":"medium","min_us":78,"med_us":80,"p90_us":87,"max_us":88},{"op":"edit-char","fixture":"medium","min_us":1036,"med_us":1377,"p90_us":1602,"max_us":1621},{"op":"edit-twin","fixture":"medium","min_us":2266,"med_us":2715,"p90_us":3695,"max_us":3927},{"op":"body-2m","fixture":"medium","min_us":4739,"med_us":8675,"p90_us":11518,"max_us":13657},{"op":"body-appends","fixture":"medium","min_us":111613,"med_us":214959,"p90_us":264859,"max_us":283506},{"op":"deep-sel","fixture":"medium","min_us":74,"med_us":76,"p90_us":89,"max_us":96}]} diff --git a/test/perf-baseline-tty-debug.json b/test/perf-baseline-tty-debug.json new file mode 100644 index 00000000..a01335bd --- /dev/null +++ b/test/perf-baseline-tty-debug.json @@ -0,0 +1 @@ +{"harness":"6cf9f4b8943e7e29","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":2169,"med_us":2697,"p90_us":3126,"max_us":3207},{"op":"render","fixture":"medium","min_us":89,"med_us":96,"p90_us":101,"max_us":115},{"op":"key-down","fixture":"medium","min_us":93,"med_us":98,"p90_us":104,"max_us":115},{"op":"key-right","fixture":"medium","min_us":96,"med_us":97,"p90_us":102,"max_us":108},{"op":"page-down","fixture":"medium","min_us":96,"med_us":98,"p90_us":102,"max_us":103},{"op":"wheel","fixture":"medium","min_us":89,"med_us":100,"p90_us":128,"max_us":168},{"op":"edit-char","fixture":"medium","min_us":1821,"med_us":2274,"p90_us":2404,"max_us":2610},{"op":"edit-twin","fixture":"medium","min_us":2710,"med_us":3180,"p90_us":3413,"max_us":3636},{"op":"body-2m","fixture":"medium","min_us":9090,"med_us":14438,"p90_us":18158,"max_us":22063},{"op":"body-appends","fixture":"medium","min_us":241367,"med_us":375874,"p90_us":470623,"max_us":528094},{"op":"deep-sel","fixture":"medium","min_us":88,"med_us":89,"p90_us":90,"max_us":90}]} diff --git a/test/perf-baseline-tty-fast.json b/test/perf-baseline-tty-fast.json new file mode 100644 index 00000000..04fc88d3 --- /dev/null +++ b/test/perf-baseline-tty-fast.json @@ -0,0 +1 @@ +{"harness":"6cf9f4b8943e7e29","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"fast","c_optimize":"fast","terminal_optimize":"fast","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1695,"med_us":2024,"p90_us":2180,"max_us":2292},{"op":"render","fixture":"medium","min_us":79,"med_us":82,"p90_us":92,"max_us":94},{"op":"key-down","fixture":"medium","min_us":81,"med_us":82,"p90_us":83,"max_us":84},{"op":"key-right","fixture":"medium","min_us":80,"med_us":82,"p90_us":91,"max_us":91},{"op":"page-down","fixture":"medium","min_us":81,"med_us":84,"p90_us":90,"max_us":171},{"op":"wheel","fixture":"medium","min_us":79,"med_us":80,"p90_us":90,"max_us":101},{"op":"edit-char","fixture":"medium","min_us":1332,"med_us":1691,"p90_us":1967,"max_us":1987},{"op":"edit-twin","fixture":"medium","min_us":2160,"med_us":2457,"p90_us":2718,"max_us":2757},{"op":"body-2m","fixture":"medium","min_us":5047,"med_us":13525,"p90_us":16257,"max_us":20323},{"op":"body-appends","fixture":"medium","min_us":155834,"med_us":261597,"p90_us":329936,"max_us":333538},{"op":"deep-sel","fixture":"medium","min_us":75,"med_us":76,"p90_us":84,"max_us":96}]} diff --git a/test/perf.zig b/test/perf.zig index c9235954..d01d1bb5 100644 --- a/test/perf.zig +++ b/test/perf.zig @@ -1054,7 +1054,7 @@ fn measureTerminalSession(io: std.Io, mib: usize, reps: usize, json: bool) !void .undo_output = undo_output, .history_rows = state.vt.screens.active.pages.scrollbar().total, .history_bytes = state.vt.screens.active.pages.page_size, - .history_limit_bytes = state.vt.screens.active.pages.maxSize(), + .history_limit_bytes = state.vt.screens.active.pages.limits.max(.bytes), .row_count_decreases = row_count_decreases, .allocator_calls = counter.calls, .allocator_live_after_output = live_after_output, @@ -1385,7 +1385,7 @@ fn reportText( for (fixtures) |fx| { if (base != null) o(" {s:>19}", .{fx.name}) else o(" {s:>12}", .{fx.name}); } - o("\n{s}\n", .{if (base != null) "-" ** (12 + fixtures.len * 20) else "-" ** (12 + fixtures.len * 13)}); + o("\n{s}\n", .{if (base != null) &@as([12 + fixtures.len * 20]u8, @splat('-')) else &@as([12 + fixtures.len * 13]u8, @splat('-'))}); var worst_jitter: f64 = 0; for (std.enums.values(Op), 0..) |op, oi| { o("{s:<12}", .{op.label()}); @@ -1412,7 +1412,7 @@ fn reportText( for (term_fixtures) |fx| { if (base != null) o(" {s:>19}", .{fx.name}) else o(" {s:>12}", .{fx.name}); } - o("\n{s}\n", .{if (base != null) "-" ** (12 + term_fixtures.len * 20) else "-" ** (12 + term_fixtures.len * 13)}); + o("\n{s}\n", .{if (base != null) &@as([12 + term_fixtures.len * 20]u8, @splat('-')) else &@as([12 + term_fixtures.len * 13]u8, @splat('-'))}); for (std.enums.values(TermOp), 0..) |op, oi| { o("{s:<12}", .{op.label()}); for (0..term_fixtures.len) |fi| { @@ -1524,10 +1524,11 @@ fn parseBase(src: []const u8, opts: Options, bytes: *const [fixtures.len]usize) defer parsed.deinit(); const report = parsed.value; if (!std.mem.eql(u8, report.harness, harness_id)) return error.BaselineHarnessMismatch; - inline for (std.meta.fields(BuildIdentity)) |field| { - const before = @field(report.build, field.name); - const after = @field(build_identity, field.name); - const same = if (field.type == []const u8) std.mem.eql(u8, before, after) else before == after; + const info = @typeInfo(BuildIdentity).@"struct"; + inline for (info.field_names, info.field_types) |name, T| { + const before = @field(report.build, name); + const after = @field(build_identity, name); + const same = if (T == []const u8) std.mem.eql(u8, before, after) else before == after; if (!same) return error.BaselineBuildMismatch; } if (report.cols != opts.cols or report.rows != opts.rows or report.reps != opts.reps or diff --git a/test/snapshot.zig b/test/snapshot.zig index 5522fbb9..2d8a8664 100644 --- a/test/snapshot.zig +++ b/test/snapshot.zig @@ -157,8 +157,8 @@ pub fn main(init: std.process.Init) !void { script_i += 1; } if (exe == null) fatal("usage: pardes-snap [scripts...] [--update] [--jobs=N]", .{}); - const exe_z = try arena.dupeZ(u8, exe.?); - lspmock_bin = if (lspmock) |p| try arena.dupeZ(u8, p) else null; + const exe_z = try arena.dupeSentinel(u8, exe.?, 0); + lspmock_bin = if (lspmock) |p| try arena.dupeSentinel(u8, p, 0) else null; if (scripts.len == 0) { scripts = try listSnaps(arena, try absPath(arena, orig_cwd, "test/snapshots")); @@ -443,7 +443,7 @@ fn runBatch( fn runOne(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const u8, update: bool) u8 { if (careful) beCareful(); const stem = std.fs.path.stem(script_path); - const report_z = arena.dupeZ(u8, reportPath(arena, script_path) catch return 1) catch return 1; + const report_z = arena.dupeSentinel(u8, reportPath(arena, script_path) catch return 1, 0) catch return 1; const fd = libc.open(report_z, .{ .ACCMODE = .WRONLY, .CREAT = true, .TRUNC = true }, @as(libc.mode_t, 0o644)); if (fd >= 0) { _ = libc.dup2(fd, 1); @@ -457,7 +457,7 @@ fn runOne(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const u8 // dirty to every manual `cmp` audit. Unlink only this script's exact path // before it starts; a new mismatch below writes it back. const actual_path = std.fmt.allocPrint(arena, "{s}.actual", .{base_path}) catch return 1; - const actual_path_z = arena.dupeZ(u8, actual_path) catch return 1; + const actual_path_z = arena.dupeSentinel(u8, actual_path, 0) catch return 1; if (record_dir == null) _ = libc.unlink(actual_path_z); const golden: ?[]u8 = if (record_dir == null) readFileAlloc(arena, golden_path) catch null else null; const started = eh.nowMs(); @@ -473,7 +473,7 @@ fn runOne(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const u8 return 0; } if (update) { - eh.writeFile(arena.dupeZ(u8, golden_path) catch return 1, out) catch return 1; + eh.writeFile(arena.dupeSentinel(u8, golden_path, 0) catch return 1, out) catch return 1; std.debug.print("UPDATED {s} ({d} bytes)\n", .{ stem, out.len }); return 0; } @@ -485,7 +485,7 @@ fn runOne(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const u8 std.debug.print("PASS {s} ({d}ms)\n", .{ stem, took }); return 0; } - eh.writeFile(arena.dupeZ(u8, actual_path) catch return 1, out) catch |err| { + eh.writeFile(arena.dupeSentinel(u8, actual_path, 0) catch return 1, out) catch |err| { std.debug.print("FAIL {s}: differs from golden; cannot write {s}: {s} ({d}ms)\n", .{ stem, actual_path, @errorName(err), took }); printFirstDiff(golden.?, out); return 1; @@ -557,7 +557,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const try mkdir(arena, work, false); // deterministic shell: fixed prompt, no history, no user rc leakage try eh.writeFile(try std.fmt.allocPrintSentinel(arena, "{s}/.bashrc", .{home}, 0), "PS1='$ '\nHISTFILE=\n"); - _ = setenv("HOME", try arena.dupeZ(u8, home), 1); + _ = setenv("HOME", try arena.dupeSentinel(u8, home, 0), 1); // Keep startup configuration hermetic even when the developer exports an // XDG directory that points outside this per-snapshot fake home. const config_home = try std.fmt.allocPrintSentinel(arena, "{s}/.config", .{home}, 0); @@ -602,8 +602,8 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const // take turns — which cost more wall clock than the other 51 scripts put // together. Scripts name it `$DUMP`. const dump_path = try std.fmt.allocPrint(arena, "{s}/dump.zon", .{base}); - _ = setenv("PARDES_DUMP", try arena.dupeZ(u8, dump_path), 1); - if (chdirRc(try arena.dupeZ(u8, work)) < 0) return error.ChdirFailed; + _ = setenv("PARDES_DUMP", try arena.dupeSentinel(u8, dump_path, 0), 1); + if (chdirRc(try arena.dupeSentinel(u8, work, 0)) < 0) return error.ChdirFailed; const chunks = try arena.alloc([]const u8, countCaptures(src)); var chunk_len: usize = 0; @@ -636,7 +636,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const try eh.appendFile(try std.fmt.allocPrintSentinel(arena, "{s}/pardes/init", .{config_home}, 0), setting); } else if (std.mem.eql(u8, cmd, "file")) { const name = tok.next() orelse return error.BadScript; - try eh.writeFile(try arena.dupeZ(u8, name), try unescape(arena, tok.rest())); + try eh.writeFile(try arena.dupeSentinel(u8, name, 0), try unescape(arena, tok.rest())); } else if (std.mem.eql(u8, cmd, "lines")) { const name = tok.next() orelse return error.BadScript; const n = try std.fmt.parseInt(usize, tok.next() orelse return error.BadScript, 10); @@ -646,7 +646,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const // scroll test needs and `file` can only give as one huge script line const tail = tok.rest(); const buf = try numberedLines(arena, n, tail); - try eh.writeFile(try arena.dupeZ(u8, name), buf); + try eh.writeFile(try arena.dupeSentinel(u8, name, 0), buf); } else if (std.mem.eql(u8, cmd, "ppmdiag")) { // The diagonal-split test PPM from the old e2e (white over blue), // grown to 512x512 with the white half filled by a fixed LCG @@ -658,7 +658,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const // down to flat color and a block always wins). const name = tok.next() orelse return error.BadScript; const ppm = try ppmDiag(arena); - try eh.writeFile(try arena.dupeZ(u8, name), ppm); + try eh.writeFile(try arena.dupeSentinel(u8, name, 0), ppm); } else if (std.mem.eql(u8, cmd, "dirmk")) { try mkdir(arena, tok.next() orelse return error.BadScript, false); } else if (std.mem.eql(u8, cmd, "run")) { @@ -668,7 +668,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const // different inotify event. This runs in the script's cwd, as a // child of the RUNNER, so nothing it does reaches the app's ptys: // whatever the app then shows, it woke up for by itself. - if (system(try arena.dupeZ(u8, tok.rest())) != 0) return error.RunFailed; + if (system(try arena.dupeSentinel(u8, tok.rest(), 0)) != 0) return error.RunFailed; } else if (std.mem.eql(u8, cmd, "start")) { const rows = try std.fmt.parseInt(u16, tok.next() orelse return error.BadScript, 10); const cols = try std.fmt.parseInt(u16, tok.next() orelse return error.BadScript, 10); @@ -676,7 +676,7 @@ fn runScript(arena: std.mem.Allocator, exe_z: [:0]const u8, script_path: []const var nextra: usize = 0; while (tok.next()) |a| : (nextra += 1) { if (nextra >= extra.len) return error.BadScript; - extra[nextra] = (try arena.dupeZ(u8, a)).ptr; + extra[nextra] = (try arena.dupeSentinel(u8, a, 0)).ptr; } if (h) |*hp| hp.deinit(); // scripts may restart (e.g. dump then -l) // a new process is a new screen: capture it whole, or its first @@ -963,13 +963,13 @@ fn treeAsleep(root: libc.pid_t) bool { while (n > 0) { n -= 1; const pid = stack[n]; - const dir_path = std.fmt.bufPrintZ(&path_buf, "/proc/{d}/task", .{pid}) catch continue; + const dir_path = std.fmt.bufPrintSentinel(&path_buf, "/proc/{d}/task", .{pid}, 0) catch continue; const dir = libc.opendir(dir_path) orelse continue; // exited between steps defer _ = libc.closedir(dir); while (libc.readdir(dir)) |d| { const tid = std.mem.sliceTo(@as([*:0]const u8, @ptrCast(&d.name)), 0); if (tid[0] == '.') continue; - const stat = readProc(&buf, std.fmt.bufPrintZ(&path_buf, "/proc/{d}/task/{s}/stat", .{ pid, tid }) catch continue) orelse continue; + const stat = readProc(&buf, std.fmt.bufPrintSentinel(&path_buf, "/proc/{d}/task/{s}/stat", .{ pid, tid }, 0) catch continue) orelse continue; // "pid (comm) S ..." — comm can hold spaces and parens, so the // state is the char two past the LAST ')' const close = std.mem.lastIndexOfScalar(u8, stat, ')') orelse continue; @@ -983,7 +983,7 @@ fn treeAsleep(root: libc.pid_t) bool { 'R', 'D', 'Z' => return false, else => {}, } - const kids = readProc(&buf, std.fmt.bufPrintZ(&path_buf, "/proc/{d}/task/{s}/children", .{ pid, tid }) catch continue) orelse continue; + const kids = readProc(&buf, std.fmt.bufPrintSentinel(&path_buf, "/proc/{d}/task/{s}/children", .{ pid, tid }, 0) catch continue) orelse continue; var it = std.mem.tokenizeAny(u8, kids, " \n"); while (it.next()) |k| { if (n >= stack.len) return false; // deeper than we can walk: assume busy @@ -1554,7 +1554,7 @@ fn absPath(arena: std.mem.Allocator, cwd: []const u8, p: []const u8) ![]const u8 } fn listSnaps(arena: std.mem.Allocator, dir_path: []const u8) ![][]const u8 { - const dir_z = try arena.dupeZ(u8, dir_path); + const dir_z = try arena.dupeSentinel(u8, dir_path, 0); const count = countSnapEntries(dir_z, dir_path); const scripts = try arena.alloc([]const u8, count); const dir = libc.opendir(dir_z) orelse fatal("no snapshots/ dir at {s}", .{dir_path}); @@ -1583,7 +1583,7 @@ fn countSnapEntries(dir_z: [:0]const u8, dir_path: []const u8) usize { } fn readFileAlloc(arena: std.mem.Allocator, path: []const u8) ![]u8 { - const path_z = try arena.dupeZ(u8, path); + const path_z = try arena.dupeSentinel(u8, path, 0); const fd = libc.open(path_z, .{ .ACCMODE = .RDONLY }); if (fd < 0) return error.OpenFailed; defer _ = libc.close(fd); @@ -1611,7 +1611,7 @@ fn readFileAlloc(arena: std.mem.Allocator, path: []const u8) ![]u8 { } fn mkdir(arena: std.mem.Allocator, path: []const u8, exist_ok: bool) !void { - const path_z = try arena.dupeZ(u8, path); + const path_z = try arena.dupeSentinel(u8, path, 0); const rc = libc.mkdir(path_z, 0o755); if (rc != 0) { if (exist_ok and libc.errno(rc) == .EXIST) return; diff --git a/test/snapshots/exec.golden b/test/snapshots/exec.golden index df49336c..30473c9e 100644 --- a/test/snapshots/exec.golden +++ b/test/snapshots/exec.golden @@ -22,15 +22,17 @@ == snap kept grid=140x31 cursor=6,3 |3: ZZARG ls |4: ./c.sh e.txt +|8: |9: == snap chord grid=140x31 cursor=6,3 |3: ZZARG Exec ls -|9: CHORDARG:ZZARG +|5: ./c.sh ZZARG +|6: CHORDARG:ZZARG == snap deleted grid=140x31 cursor=4,5 |3: ls ls |4: e.txt e.txt -|8: -|9: +|5: +|6: |17: |18: |19: diff --git a/test/snapshots/tinywin.golden b/test/snapshots/tinywin.golden deleted file mode 100644 index 787a9d8f..00000000 --- a/test/snapshots/tinywin.golden +++ /dev/null @@ -1,83 +0,0 @@ -== snap stacked grid=100x31 cursor=14,3 -| Newcol Joincol Find Grep Help Changelog Tutor Dump Themes Config Debug Exit -| New Tty Find Grep Joincol Delcol -|$ /tmp/pardes-snap/tinywin/cwd Tty+bash Save Mode Filter Collapse Del -| $ :^[n^[n^[n -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| /tmp/pardes-snap/tinywin/cwd/+New Save Tty Collapse Del -| 1 -== snap shrunk grid=60x4 cursor=59,3 -| Newcol Joincol Find Grep Help Changelog Tutor Dump Themes -| New Tty Find Grep Joincol Delcol -|$ /tmp/pardes-snap/tinywin/cwd Tty+bash Save Mode Filter Col -| /tmp/pardes-snap/tinywin/cwd/+New Save Tty Collapse Del -== snap columns grid=100x31 cursor=23,3 -| Newcol Joincol Find Grep Help Changelog Tutor Dump Themes Config Debug Exit -| New Tty Find Grep Joincol Delcol -|$ /tmp/pardes-snap/tinywin/cwd Tty+bash Save Mode Filter Collapse Del -| $ :^[n^[n^[n cn cn cn -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| -| /tmp/pardes-snap/tinywin/cwd/+New Save Tty Collapse Del -| 1 -== snap narrow grid=4x31 cursor=2,20 -| Ne -| Ne -|$ /t -| mp -| /p -| ar -| de -| s- -| sn -| ap -| $ -| ^[ -| n^ -| [n -| ^[ -| n -| cn -| c -| n -| cn -| -| /t -| mp -| /p -| ar -| de -| s- -| sn -| ap diff --git a/test/snapshots/tinywin.snap b/test/snapshots/tinywin.snap deleted file mode 100644 index 3d22d583..00000000 --- a/test/snapshots/tinywin.snap +++ /dev/null @@ -1,31 +0,0 @@ -# Shrinking the window past "one row per pane" used to panic: computeGeom gave -# every pane in a column at least one row, so `y` walked off the bottom and -# renderPane got a rect outside the surface (assert). Panes that no longer fit -# now get h = 0 and simply do not render until the window grows back. -start 31 100 -wait 8000 $ -key a-n -stable 400 8000 -key a-n -stable 400 8000 -key a-n -stable 400 8000 -key a-n -stable 700 15000 -snap stacked -resize 4 60 -stable 700 10000 -snap shrunk -# ...and the same sideways: more columns than the window is wide -resize 31 100 -stable 700 10000 -key space c n -stable 600 8000 -key space c n -stable 600 8000 -key space c n -stable 700 15000 -snap columns -resize 31 4 -stable 700 10000 -snap narrow diff --git a/test/unit_profile_test.zig b/test/unit_profile_test.zig index 721f1212..a4cdd09e 100644 --- a/test/unit_profile_test.zig +++ b/test/unit_profile_test.zig @@ -149,11 +149,8 @@ fn fixture(init: std.process.Init) !void { var write_buffer: [1024]u8 = undefined; var reader = std.Io.File.stdin().readerStreaming(io, &read_buffer); var writer = std.Io.File.stdout().writerStreaming(io, &write_buffer); - var server = try std.zig.Server.init(.{ - .in = &reader.interface, - .out = &writer.interface, - .zig_version = if (std.mem.eql(u8, mode, "version")) "wrong-version" else builtin.zig_version_string, - }); + var server: std.zig.Server = .{ .in = &reader.interface, .out = &writer.interface }; + try server.serveStringMessage(.zig_version, if (std.mem.eql(u8, mode, "version")) "wrong-version" else builtin.zig_version_string); while (true) { const request = try server.receiveMessage(); switch (request.tag) { diff --git a/tools/gen_themes.zig b/tools/gen_themes.zig index 65e85d1c..be5fde2f 100644 --- a/tools/gen_themes.zig +++ b/tools/gen_themes.zig @@ -152,13 +152,14 @@ fn byName(_: void, a: Theme, b: Theme) bool { fn emit(w: *std.Io.Writer, t: Theme) !void { try w.writeAll("// Generated by tools/gen_themes.zig. Do not edit.\npub const theme = .{\n"); try w.print(" .name = \"{s}\",\n", .{t.name}); - inline for (@typeInfo(Theme).@"struct".fields) |f| { - if (comptime std.mem.eql(u8, f.name, "name")) continue; - if (comptime std.mem.eql(u8, f.name, "palette")) continue; - try w.print(" .{s} = ", .{f.name}); - if (comptime @typeInfo(f.type) == .optional) { - if (@field(t, f.name)) |c| try rgb(w, c) else try w.writeAll("null"); - } else try rgb(w, @field(t, f.name)); + const info = @typeInfo(Theme).@"struct"; + inline for (info.field_names, info.field_types) |name, T| { + if (comptime std.mem.eql(u8, name, "name")) continue; + if (comptime std.mem.eql(u8, name, "palette")) continue; + try w.print(" .{s} = ", .{name}); + if (comptime @typeInfo(T) == .optional) { + if (@field(t, name)) |c| try rgb(w, c) else try w.writeAll("null"); + } else try rgb(w, @field(t, name)); try w.writeAll(",\n"); } if (t.palette) |p| { -- cgit v1.3 From 0073d05329335dbb0eb85f5c3044e42805d144e5 Mon Sep 17 00:00:00 2001 From: Gabriel Schneider Date: Fri, 23 Oct 2026 17:10:58 -0300 Subject: Callstack lists every caller and pages a tree too big for one query; memory bounds and resident-size fixes Callstack walks while the server answers, counts on the message row past a second, puts test callers last at each level, and ends a tree past one query's memory (memory.limits.arena.lsp_query, a 256 MiB region taken and given back per query) in a row whose Callstack shows the next page. Grep reads whole files. New hard limits: undo text per file, image decode pixels, the LSP snapshot of other open files, one paste. Safe builds no longer fill worst-case buffers with 0xAA (memory.unfilled), big regions are kept off transparent huge pages, and malloc_trim runs after big results. Also: Look's git show pages into a +Diff, and dropped marks in File. Co-Authored-By: Claude Opus 5.5 --- build.zig.zon | 4 +- docs/typ/guide.typ | 14 +- docs/typ/reference.typ | 9 +- src/9p.zig | 24 -- src/9p_io.zig | 121 +++--- src/File.zig | 141 +++++-- src/Registers.zig | 13 + src/ShaderBuild.zig | 23 +- src/builtins.zig | 3 + src/colors.zig | 2 +- src/config.zig | 66 +--- src/crash.zig | 6 + src/detached/server.zig | 29 +- src/dump.zig | 11 - src/edit.zig | 190 ++++++++-- src/exec.zig | 6 + src/fs.zig | 82 ++-- src/gui/gui.zig | 329 ++++++---------- src/gui/pet.zig | 29 +- src/host_io.zig | 103 ++++-- src/hunks.zig | 19 +- src/image.zig | 8 +- src/locations.zig | 135 +++++-- src/look.zig | 61 ++- src/lsp/lsp.zig | 17 +- src/lsp/lsp_client.zig | 761 ++++++++++++++++++++++++++------------ src/macos.zig | 17 +- src/main.zig | 12 +- src/memory.zig | 98 +++++ src/mini.zig | 16 +- src/modal.zig | 139 +++++-- src/mouse.zig | 11 +- src/ninep/addr.zig | 42 ++- src/ninep/ctl.zig | 4 + src/panes.zig | 3 + src/pardes.zig | 42 ++- src/regexp.zig | 11 +- src/sam_edit.zig | 15 + src/selection_pipe.zig | 11 +- src/surface.zig | 5 - src/terminal.zig | 86 +---- src/tty/tty.zig | 51 ++- test/lspbench.zig | 22 +- test/panes.zig | 3 - test/perf-baseline-gui-debug.json | 2 +- test/perf-baseline-tty-debug.json | 2 +- test/perf-baseline-tty-fast.json | 2 +- test/perf.zig | 21 +- 48 files changed, 1843 insertions(+), 978 deletions(-) (limited to 'src/regexp.zig') diff --git a/build.zig.zon b/build.zig.zon index 292bd479..7885fd4e 100644 --- a/build.zig.zon +++ b/build.zig.zon @@ -9,8 +9,8 @@ // read-only HTTPS URL is what a manifest can carry. Re-pin with // `zig fetch --save=cloud9 git+https://git.sr.ht/~gbrls/cloud9#`. .cloud9 = .{ - .url = "git+https://git.sr.ht/~gbrls/cloud9#d5c37d9122240fe95d308fbe15874e14dd786c5a", - .hash = "cloud9-0.1.0-yt86qnAvHgB1dEWGakukcJ54GZ7J3mv_lm_kApzAuEOY", + .url = "git+https://git.sr.ht/~gbrls/cloud9#0702e33bcfadf0abd9ccbaa8b681a19a84c332fe", + .hash = "cloud9-0.1.0-yt86qkExHgCRI7_eJsztRECamioWguBj2mTFsJ5N-AOn", }, // Pinned to the exact libvaxis commit ghostty is built against, so it // is compatible with ghostty's uucode configuration (the one we share diff --git a/docs/typ/guide.typ b/docs/typ/guide.typ index a5327a44..e60b4231 100644 --- a/docs/typ/guide.typ +++ b/docs/typ/guide.typ @@ -240,11 +240,19 @@ wrapping. Cross-directory Look paths are shortened when possible. pane. #word("LookNext") / #word("LookPrev") (#keys("SPC j n", "SPC j p")) select locations without opening them. The view follows the language server's static call hierarchy; recursive paths are marked `↻`, and a -large hierarchy invites continuing from another function. +hierarchy deeper than the limit invites continuing from another function. +Every caller is listed: the walk runs as long as the server keeps answering, +says how many callers it has found on the message row once it takes more +than a second, and shows what it found if the server stops answering. +Asking anything else stops it. A tree too big for one page (about 100,000 +rows, the memory one query may use) ends in a row saying so: running +#word("Callstack") in `+Callstack` shows the next page, from where this +one stopped, with that caller's own callers above it again. +At each level, callers whose function or file name contains "test" come +after the rest. #word("CallstackDepth") reports the current caller depth; `CallstackDepth 16` changes it for subsequent queries. The default is 8 -levels, and the accepted range is 1–64. The view also caps collection at -64 callers. +levels, and the accepted range is 1–64. `CallstackLayout stacked` (the default) puts each signature and call snippet below its location, aligned with the location text; `CallstackLayout inline` puts it diff --git a/docs/typ/reference.typ b/docs/typ/reference.typ index bb13b011..c14ad659 100644 --- a/docs/typ/reference.typ +++ b/docs/typ/reference.typ @@ -264,7 +264,9 @@ A line written to #file("exec") is #word("Exec"): a declaration; from a use it reaches every file); #word("Callstack") (#key("g c") or #key("SPC l b")) opens and focuses `+Callstack`, with static caller ancestry, tree-sitter signatures and Look locations - for callers and call sites. #word("CallstackDepth") reports or sets the + for callers and call sites, test callers after the rest at each level; run + in `+Callstack` itself, it shows the next page of a tree too big for one. + #word("CallstackDepth") reports or sets the runtime depth (1–64, default 8) for subsequent queries. `CallstackLayout stacked` (the default) aligns snippets below their locations; `CallstackLayout inline` places the snippets beside them. @@ -971,12 +973,11 @@ one a #file("exec") open still holds is never reused. In a command pane, #key("n = Find and Grep #word("Find") matches file names, #word("Grep") the text of lines, both literally and -ignoring ASCII case. #word("Grep") reads at most 256 KiB of a file and stops at 512 +ignoring ASCII case. #word("Grep") reads whole files and stops at 512 hits, #word("Find") at 512 names. The walk stops at 20000 files or 100000 entries, 16 deep, and passes over `.git`, `.jj`, `target`, `node_modules`, `.venv`, `__pycache__`, `.zig-cache` and `zig-out`. A cap hit, or a -directory it could not open, is said at the end: `cut at 512 hits`, `N -files read only in part (first 256 KiB)`, `walk cut at N entries`, `N +directory it could not open, is said at the end: `cut at 512 hits`, `walk cut at N entries`, `N directories skipped: permission denied`. A search that finds nothing and skipped nothing fails, `Grep: text not found`, and leaves the `+Search` as it was. diff --git a/src/9p.zig b/src/9p.zig index 67b3b39a..0979b66a 100644 --- a/src/9p.zig +++ b/src/9p.zig @@ -5,16 +5,10 @@ const cloud9 = @import("cloud9"); pub const fs = cloud9.fs; pub const Server = cloud9.fs.Server; -pub const Error = cloud9.Error; -pub const Type = cloud9.Type; -pub const isT = cloud9.isT; pub const header_len = cloud9.header_len; -pub const qid_len = cloud9.qid_len; -pub const stat_fixed = cloud9.stat_fixed; pub const notag = cloud9.notag; pub const nofid = cloud9.nofid; pub const max_welem = cloud9.max_welem; -pub const iohdrsz = cloud9.iohdrsz; pub const Qid = cloud9.Qid; pub const Stat = cloud9.Stat; pub const Msg = cloud9.Msg; @@ -23,29 +17,11 @@ pub const frameLen = cloud9.frameLen; pub const encode = cloud9.encode; pub const decode = cloud9.decode; pub const qtdir = cloud9.qtdir; -pub const qtappend = cloud9.qtappend; -pub const qtexcl = cloud9.qtexcl; -pub const qtmount = cloud9.qtmount; -pub const qtauth = cloud9.qtauth; -pub const qttmp = cloud9.qttmp; -pub const qtfile = cloud9.qtfile; -pub const dmdir = cloud9.dmdir; -pub const dmappend = cloud9.dmappend; -pub const dmexcl = cloud9.dmexcl; -pub const dmmount = cloud9.dmmount; -pub const dmauth = cloud9.dmauth; -pub const dmtmp = cloud9.dmtmp; -pub const dmperm = cloud9.dmperm; pub const Client = cloud9.Client; -pub const ClientError = cloud9.ClientError; -pub const max_tags = cloud9.max_tags; pub const oread = cloud9.oread; pub const owrite = cloud9.owrite; pub const ordwr = cloud9.ordwr; -pub const oexec = cloud9.oexec; pub const otrunc = cloud9.otrunc; -pub const ocexec = cloud9.ocexec; -pub const orclose = cloud9.orclose; pub const errmax = cloud9.fs.errmax; pub const errString = cloud9.fs.errString; pub const msize_min = cloud9.fs.msize_min; diff --git a/src/9p_io.zig b/src/9p_io.zig index 446c24de..87972b43 100644 --- a/src/9p_io.zig +++ b/src/9p_io.zig @@ -174,8 +174,15 @@ pub const Listener = struct { io: std.Io, /// The core that answers; `reset` puts a replacement in. core: *pardes.Pardes, - runner: Runner = undefined, + /// cloud9's runner, sized for every connection's buffers and parks (some + /// 136 MiB): pages of its own (`allocRunner`), written only as clients + /// take slots, so an idle session does not hold them resident. + runner: *Runner = undefined, stopping: std.atomic.Value(bool) = .init(false), + /// The watchers of held Edit writes (watchEdit): canceled and awaited + /// by `deinit`, once the turn lets a waiting one through, so none wakes + /// into a freed Listener. + edit_watchers: std.Io.Group = .init, /// A connection wrote the Restore and has yet to answer it. restore_writer: std.atomic.Value(bool) = .init(false), /// Clients turned away with every slot taken, since the log last said so. @@ -354,8 +361,7 @@ pub const Listener = struct { const later: pardes.ctlfs.Reply = .{ .tag = req.tag, .status = .again }; const ticket = conn.hold(&later) orelse return; core.fs.edit_hold = .{ .asker = conn, .slot = ticket.slot, .seq = ticket.seq, .tag = req.tag, .node = req.node, .handle = req.handle, .written = reply.written }; - const watcher = std.Thread.spawn(.{}, watchEdit, .{ l, conn, ticket }) catch return; - watcher.detach(); + l.edit_watchers.concurrent(l.io, watchEdit, .{ l, conn, ticket }) catch {}; return; } const restoring = core.restore_req != null; @@ -457,13 +463,13 @@ pub const Listener = struct { /// flush, so it is looked for, every 100 ms; a park out for a retry is /// not waiting either, so only three looks in a row with no write being /// served count. - fn watchEdit(l: *Listener, conn: *Runner.Conn, ticket: cloud9.fs.Ticket) void { - const io = pardes.turn.io orelse return; + fn watchEdit(l: *Listener, conn: *Runner.Conn, ticket: cloud9.fs.Ticket) error{Canceled}!void { var misses: u8 = 0; while (!l.stopping.load(.acquire)) { - io.sleep(.fromMilliseconds(100), .awake) catch return; + try l.io.sleep(.fromMilliseconds(100), .awake); _ = pardes.turn.take(); defer pardes.turn.give(); + if (l.stopping.load(.acquire)) return; const core = l.core; const h = core.fs.edit_hold orelse return; if (h.slot != ticket.slot or h.seq != ticket.seq or h.asker != @as(*anyopaque, conn)) return; @@ -914,6 +920,7 @@ pub const Listener = struct { // A connection task may be waiting for the turn; it answers // nothing more once stopping, but the wait itself must end. pardes.turn.rest(); + l.edit_watchers.cancel(l.io); // An answer given as the editor quit (a Kill written to /ctl) // may still be on its way out: let it go before the cut. const deadline = nowMs() +| 200; @@ -942,10 +949,26 @@ pub const Listener = struct { z[l.path_len] = 0; _ = libc.unlink(z[0..l.path_len :0]); } + freeRunner(l.runner); gpa.destroy(l); } }; +/// The runner's memory, from the page allocator with no fill: an Allocator's +/// `create` writes `undefined` over all of it in a safe build, which would make +/// every page resident at startup. +fn allocRunner() ?*Runner { + if (comptime !supported) return @constCast(&@as(Runner, {})); + const bytes = std.heap.page_allocator.rawAlloc(@sizeOf(Runner), .of(Runner), @returnAddress()) orelse return null; + @import("memory.zig").sparse(bytes[0..@sizeOf(Runner)]); + return @ptrCast(@alignCast(bytes)); +} + +fn freeRunner(runner: *Runner) void { + if (comptime !supported) return; + std.heap.page_allocator.rawFree(@as([*]u8, @ptrCast(runner))[0..@sizeOf(Runner)], .of(Runner), @returnAddress()); +} + /// The socket this process listens on, kept where a fatal signal's handler /// can reach it without allocating (unlinkOwnSocket); empty when none. var own_socket: [sun_path_len:0]u8 = undefined; @@ -1065,6 +1088,10 @@ pub fn listen(io: std.Io, gpa: std.mem.Allocator, core: *pardes.Pardes, named: [ } const l = gpa.create(Listener) catch return null; l.* = .{ .io = io, .core = core }; + l.runner = allocRunner() orelse { + gpa.destroy(l); + return null; + }; pardes.turn.start(io); pardes.turn.wake_parked = Listener.wakeParked; pardes.turn.answer_held = Listener.answerHeld; @@ -1551,7 +1578,7 @@ test "a socket path too long for sun_path is no listener, and leaves the editor' defer if (old_copy) |o| { _ = setenv("XDG_RUNTIME_DIR", o, 1); } else { - _ = unsetenv("XDG_RUNTIME_DIR"); + _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); }; const long = &@as([120]u8, @splat('n')); try testing.expect(listen(testing.io, testing.allocator, p, long, "", null, null) == null); @@ -1572,7 +1599,7 @@ test "a listening editor posts itself into the 9P registry and unposts on stop" if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); gpa.free(v); - } else _ = unsetenv("XDG_RUNTIME_DIR"); + } else _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); } try testing.expectEqual(@as(c_int, 0), setenv("XDG_RUNTIME_DIR", &directory, 1)); @@ -1633,7 +1660,7 @@ test "Unix TCP and QUIC share one listener through reads writes reconnects and r if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); gpa.free(v); - } else _ = unsetenv("XDG_RUNTIME_DIR"); + } else _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); } try testing.expectEqual(@as(c_int, 0), setenv("XDG_RUNTIME_DIR", &directory, 1)); const replacement = try gpa.alloc(u8, 3 * msize + 27); @@ -1754,7 +1781,7 @@ test "a change waits while the editor is out mid-step, a read does not, and the if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); gpa.free(v); - } else _ = unsetenv("XDG_RUNTIME_DIR"); + } else _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); } try testing.expectEqual(@as(c_int, 0), setenv("XDG_RUNTIME_DIR", &directory, 1)); @@ -1829,7 +1856,7 @@ test "a held read is answered when the log has news, and a flushed one spends no if (old_runtime) |v| { _ = setenv("XDG_RUNTIME_DIR", v, 1); gpa.free(v); - } else _ = unsetenv("XDG_RUNTIME_DIR"); + } else _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); } try testing.expectEqual(@as(c_int, 0), setenv("XDG_RUNTIME_DIR", &directory, 1)); @@ -2001,28 +2028,22 @@ pub fn exportSessionEnv(serial: u32, adopts: bool) void { exportEnv(if (own_socket_len > 0) own_socket[0..own_socket_len] else null, serial, adopts); } +/// Told to children through host_io.ChildEnv, never set into this process's +/// own environment (see ChildEnv.exportVar for why). fn exportEnv(socket: ?[]const u8, serial: u32, adopts: bool) void { - var announced = false; - if (adopts) announcing: { - var buf: [16]u8 = undefined; - const text = std.fmt.bufPrintSentinel(&buf, "{d}", .{@as(u32, @intCast(libc.getpid()))}, 0) catch - break :announcing; - announced = setenv("PARDES_PID", text, 1) == 0; - } - if (!announced) _ = unsetenv("PARDES_PID"); - - if (socket) |own| exporting: { - var sock: [sun_path_len]u8 = undefined; - const path = std.fmt.bufPrintSentinel(&sock, "{s}", .{own}, 0) catch break :exporting; + const ChildEnv = @import("host_io.zig").ChildEnv; + var pid_buf: [16]u8 = undefined; + ChildEnv.exportVar("PARDES_PID", if (adopts) std.fmt.bufPrint(&pid_buf, "{d}", .{@as(u32, @intCast(libc.getpid()))}) catch null else null); + if (socket) |own| { var buf: [16]u8 = undefined; - const id = std.fmt.bufPrintSentinel(&buf, "{d}", .{serial}, 0) catch break :exporting; - if (setenv("PARDES_9P", path, 1) != 0) break :exporting; + ChildEnv.exportVar("PARDES_9P", own); exportMount(own); - if (setenv("PARDES_PANE", id, 1) == 0) return; + ChildEnv.exportVar("PARDES_PANE", std.fmt.bufPrint(&buf, "{d}", .{serial}) catch unreachable); + return; } - _ = unsetenv("PARDES_9P"); - _ = unsetenv("PARDES_PANE"); - _ = unsetenv("PARDES_MOUNT"); + ChildEnv.exportVar("PARDES_9P", null); + ChildEnv.exportVar("PARDES_PANE", null); + ChildEnv.exportVar("PARDES_MOUNT", null); } /// $PARDES_MOUNT, the session's own directory, when the environment says @@ -2040,11 +2061,11 @@ fn exportMount(sock_path: []const u8) void { const suffix = ".sock"; if (mntgen and root.len > 0 and std.mem.startsWith(u8, base, sock_prefix) and std.mem.endsWith(u8, base, suffix) and base.len > sock_prefix.len + suffix.len) { var buf: [4096]u8 = undefined; - if (std.fmt.bufPrintSentinel(&buf, "{s}/pardes/{s}", .{ std.mem.trimEnd(u8, root, "/"), base[sock_prefix.len .. base.len - suffix.len] }, 0)) |mount| { - if (setenv("PARDES_MOUNT", mount, 1) == 0) return; + if (std.fmt.bufPrint(&buf, "{s}/pardes/{s}", .{ std.mem.trimEnd(u8, root, "/"), base[sock_prefix.len .. base.len - suffix.len] })) |mount| { + return @import("host_io.zig").ChildEnv.exportVar("PARDES_MOUNT", mount); } else |_| {} } - _ = unsetenv("PARDES_MOUNT"); + @import("host_io.zig").ChildEnv.exportVar("PARDES_MOUNT", null); } test "a pane's PARDES_MOUNT is the session's directory under 9ns --mntgen, and unset when not known" { @@ -2057,16 +2078,16 @@ test "a pane's PARDES_MOUNT is the session's directory under 9ns --mntgen, and u if (value) |old| { _ = setenv(name, old, 1); testing.allocator.free(old); - } else _ = unsetenv(name); + } else _ = @import("host_io.zig").testUnsetenv(name); }; _ = setenv("NINE_MOUNT", "/mnt/9p", 1); _ = setenv("NINE_MNTGEN", "1", 1); exportMount("/run/user/1000/pardes-9p-work.sock"); - try testing.expectEqualStrings("/mnt/9p/pardes/work", std.mem.span(libc.getenv("PARDES_MOUNT").?)); + try testing.expectEqualStrings("/mnt/9p/pardes/work", @import("host_io.zig").ChildEnv.exported("PARDES_MOUNT").?); // A mount of another server (9ns --unix): not this session's. - _ = unsetenv("NINE_MNTGEN"); + _ = @import("host_io.zig").testUnsetenv("NINE_MNTGEN"); exportMount("/run/user/1000/pardes-9p-work.sock"); - try testing.expect(libc.getenv("PARDES_MOUNT") == null); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_MOUNT") == null); } test "9P shell environment states being inside pardes apart from how to reach it" { @@ -2079,16 +2100,16 @@ test "9P shell environment states being inside pardes apart from how to reach it if (value) |old| { _ = setenv(name, old, 1); testing.allocator.free(old); - } else _ = unsetenv(name); + } else _ = @import("host_io.zig").testUnsetenv(name); }; var dir: [sun_path_len:0]u8 = undefined; _ = setenv("XDG_RUNTIME_DIR", "/run/user/1000", 1); try testing.expectEqualStrings("/run/user/1000", socketDir(&dir).?); - _ = unsetenv("XDG_RUNTIME_DIR"); + _ = @import("host_io.zig").testUnsetenv("XDG_RUNTIME_DIR"); _ = setenv("HOME", "/home/example", 1); try testing.expectEqualStrings("/home/example/.local/state/pardes", socketDir(&dir).?); - _ = unsetenv("HOME"); + _ = @import("host_io.zig").testUnsetenv("HOME"); try testing.expect(socketDir(&dir) == null); const listener = try testing.allocator.create(Listener); @@ -2104,25 +2125,25 @@ test "9P shell environment states being inside pardes apart from how to reach it // parent, so a pardes started in one of its shells runs a session of // its own instead of handing its argument over. exportPaneEnv(listener, 7, false); - try testing.expectEqualStrings(path, std.mem.span(libc.getenv("PARDES_9P").?)); - try testing.expectEqualStrings("7", std.mem.span(libc.getenv("PARDES_PANE").?)); - try testing.expect(libc.getenv("PARDES_PID") == null); + try testing.expectEqualStrings(path, @import("host_io.zig").ChildEnv.exported("PARDES_9P").?); + try testing.expectEqualStrings("7", @import("host_io.zig").ChildEnv.exported("PARDES_PANE").?); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_PID") == null); exportPaneEnv(listener, 8, true); - try testing.expectEqualStrings("8", std.mem.span(libc.getenv("PARDES_PANE").?)); - try testing.expectEqualStrings(own_pid, std.mem.span(libc.getenv("PARDES_PID").?)); + try testing.expectEqualStrings("8", @import("host_io.zig").ChildEnv.exported("PARDES_PANE").?); + try testing.expectEqualStrings(own_pid, @import("host_io.zig").ChildEnv.exported("PARDES_PID").?); // A session whose listener never came up is still the session this shell // is inside: the pid stands on its own, and only the address is missing. exportPaneEnv(null, 0, true); - try testing.expect(libc.getenv("PARDES_9P") == null); - try testing.expect(libc.getenv("PARDES_PANE") == null); - try testing.expectEqualStrings(own_pid, std.mem.span(libc.getenv("PARDES_PID").?)); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_9P") == null); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_PANE") == null); + try testing.expectEqualStrings(own_pid, @import("host_io.zig").ChildEnv.exported("PARDES_PID").?); exportPaneEnv(null, 0, false); - try testing.expect(libc.getenv("PARDES_9P") == null); - try testing.expect(libc.getenv("PARDES_PANE") == null); - try testing.expect(libc.getenv("PARDES_PID") == null); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_9P") == null); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_PANE") == null); + try testing.expect(@import("host_io.zig").ChildEnv.exported("PARDES_PID") == null); } pub const Client = struct { diff --git a/src/File.zig b/src/File.zig index 59b04e3f..7e05ac9f 100644 --- a/src/File.zig +++ b/src/File.zig @@ -43,14 +43,14 @@ pub const Snapshot = struct { saved_len: ?usize = null, fn deinit(s: Snapshot, gpa: std.mem.Allocator) void { - gpa.free(s.content); + pardes.memory.freeUnfilled(gpa, s.content); } }; /// The pane's text, place and name now, for an undo or redo step. fn snapshotOf(gpa: std.mem.Allocator, pane: *const Pane, f: *const State) ?Snapshot { return .{ - .content = gpa.dupe(u8, f.content) catch return null, + .content = pardes.memory.dupeUnfilled(gpa, f.content) catch return null, .cur_row = pane.body.cur_row, .cur_col = pane.body.cur_col, .vsel = pane.body.vsel, @@ -105,6 +105,8 @@ pub const History = struct { undo_len: usize = 0, redo: [limits.undo_max]Snapshot = undefined, redo_len: usize = 0, + /// The text `undo` and `redo` hold, kept near limits.undo_bytes. + bytes: usize = 0, /// The names the file had, each with the rename generation it ended /// (Snapshot.gen), and the generation now. @@ -115,6 +117,7 @@ pub const History = struct { const history = try gpa.create(History); history.undo_len = 0; history.redo_len = 0; + history.bytes = 0; history.names = .empty; history.gen = 0; return history; @@ -165,6 +168,9 @@ pub const State = struct { /// edit drops them (setContentSpan); shifting them through the splice, /// as pane.zig's shiftOne moves dot, would keep them. pager_spans: []pardes.pager.Span = &.{}, + /// The rows and spans an edit dropped, with the text they were for: an + /// edit (an undo) back to that very text takes them again. + dropped_marks: ?DroppedMarks = null, /// Seconds of the last content change, for the filesystem's stat; zero until edited. mtime: u32 = 0, /// Non-null for a generated output buffer rather than an on-disk file. @@ -457,10 +463,6 @@ test "an invalid byte in a grapheme is one cell, and never reaches vaxis's width try std.testing.expectEqual(@as(usize, 4), fitEnd(text, 0, 10)); } -pub fn byteDisplayWidth(byte: u8) usize { - return if (byte == '\t') config.tab_now else 1; -} - pub fn displayWidth(text: []const u8) usize { var width: usize = 0; var at: usize = 0; @@ -787,18 +789,9 @@ pub fn lineIndex(gpa: std.mem.Allocator, f: *State) ![]const usize { pub const Span = struct { head: usize, tail: usize }; fn editedLineStarts(gpa: std.mem.Allocator, starts: []const usize, old: []const u8, new: []const u8, known: ?Span) ![]usize { - const n = @min(old.len, new.len); - var head: usize = 0; - var tail: usize = 0; - if (known) |span| { - head = span.head; - tail = span.tail; - } else { - while (head + 64 <= n and std.mem.eql(u8, old[head..][0..64], new[head..][0..64])) head += 64; - while (head < n and old[head] == new[head]) head += 1; - while (tail + 64 <= n - head and std.mem.eql(u8, old[old.len - tail - 64 ..][0..64], new[new.len - tail - 64 ..][0..64])) tail += 64; - while (tail < n - head and old[old.len - tail - 1] == new[new.len - tail - 1]) tail += 1; - } + const span = known orelse spanOf(old, new); + const head = span.head; + const tail = span.tail; // A start s follows the newline at s-1: kept while that newline is in // the common head, shifted while it is in the common tail. const kept = std.sort.upperBound(usize, starts, head, struct { @@ -1123,6 +1116,7 @@ pub fn deinit(p: *Pardes, pane: *Pane, file: *State) void { if (file.highlights.len > 0) p.tree_sitter_gpa.free(file.highlights); p.gpa.free(file.pager_spans); locations.freeRows(p.gpa, file.location_rows); + if (file.dropped_marks) |m| m.deinit(p.gpa); if (file.context_declarations.len > 0) p.tree_sitter_gpa.free(file.context_declarations); if (file.node_tree) |tree| syntax.freeTree(tree); for (file.node_history.items) |step| p.gpa.free(step.ranges); @@ -1140,6 +1134,18 @@ fn reportEdit(p: *Pardes, f: *State, new: []const u8) void { pardes.ctlfs.events.noteReplace(p, id, false, f.content, new); } +pub const DroppedMarks = struct { + len: usize, + hash: u64, + rows: []locations.Row, + spans: []pardes.pager.Span, + + fn deinit(m: DroppedMarks, gpa: std.mem.Allocator) void { + locations.freeRows(gpa, m.rows); + gpa.free(m.spans); + } +}; + pub fn setContent(p: *Pardes, f: *State, new: []u8) void { setContentSpan(p, f, new, null); } @@ -1160,10 +1166,16 @@ pub fn setContentSpan(p: *Pardes, f: *State, new: []u8, span_in: ?Span) void { moved[i] = .{ .cursor = spliced(@min(at, f.content.len), f.content.len, new.len, span.?), .scroll = spliced(lineStart(p.gpa, tf, tf.scroll), f.content.len, new.len, span.?) }; } const old_starts = f.line_starts.ptr; - locations.freeRows(p.gpa, f.location_rows); - f.location_rows = &.{}; - p.gpa.free(f.pager_spans); - f.pager_spans = &.{}; + if (f.location_rows.len > 0 or f.pager_spans.len > 0) { + if (f.dropped_marks) |m| m.deinit(p.gpa); + f.dropped_marks = .{ .len = f.content.len, .hash = std.hash.Wyhash.hash(0, f.content), .rows = f.location_rows, .spans = f.pager_spans }; + f.location_rows = &.{}; + f.pager_spans = &.{}; + } else if (f.dropped_marks) |m| if (m.len == new.len and m.hash == std.hash.Wyhash.hash(0, new)) { + f.location_rows = m.rows; + f.pager_spans = m.spans; + f.dropped_marks = null; + }; for (p.panes) |slot| if (slot) |pane| { if (pane.file) |*file| if (file == f) { for (0..pane.sel.len) |button| { @@ -1182,7 +1194,9 @@ pub fn setContentSpan(p: *Pardes, f: *State, new: []u8, span_in: ?Span) void { f.saved_hash = std.hash.Wyhash.hash(0, f.content); f.saved_len = f.content.len; } - p.gpa.free(f.content); + // Raw: `free` overwrites a safe build's freed memory, a whole file's + // worth per keystroke. + if (f.content.len > 0) p.gpa.rawFree(f.content, .of(u8), @returnAddress()); f.content = new; f.revision +%= 1; if (f.saved_hash) |h| if ((f.saved_len orelse new.len) == new.len and std.hash.Wyhash.hash(0, new) == h) { @@ -1241,14 +1255,43 @@ fn spliced(at: usize, old_len: usize, new_len: usize, span: Span) usize { } /// What `old` and `new` agree on at either end. -fn spanOf(old: []const u8, new: []const u8) Span { +/// What `new` kept of `old` at either end, compared 32 bytes at a time +/// (a byte at a time was most of a `d` in a 7 MB file). +pub fn spanOf(old: []const u8, new: []const u8) Span { + const block = 32; const n = @min(old.len, new.len); - const head = std.mem.indexOfDiff(u8, old[0..n], new[0..n]) orelse n; + var head: usize = 0; + while (head + block <= n and std.mem.eql(u8, old[head..][0..block], new[head..][0..block])) head += block; + while (head < n and old[head] == new[head]) head += 1; var tail: usize = 0; + while (tail + block <= n - head and std.mem.eql(u8, old[old.len - tail - block ..][0..block], new[new.len - tail - block ..][0..block])) tail += block; while (tail < n - head and old[old.len - 1 - tail] == new[new.len - 1 - tail]) tail += 1; return .{ .head = head, .tail = tail }; } +test "spanOf by blocks finds the byte loop's ends" { + var prng: std.Random.DefaultPrng = .init(5); + const random = prng.random(); + var a: [160]u8 = undefined; + var b: [160]u8 = undefined; + for (0..4000) |_| { + const la = random.uintLessThan(usize, a.len); + const lb = random.uintLessThan(usize, b.len); + for (a[0..la]) |*c| c.* = "ab"[random.uintLessThan(usize, 2)]; + const keep = @min(la, lb); + @memcpy(b[0..keep], a[0..keep]); + for (b[keep..lb]) |*c| c.* = "ab"[random.uintLessThan(usize, 2)]; + if (lb > 0 and random.boolean()) b[random.uintLessThan(usize, lb)] ^= 1; + const old = a[0..la]; + const new = b[0..lb]; + const n = @min(old.len, new.len); + const head = std.mem.indexOfDiff(u8, old[0..n], new[0..n]) orelse n; + var tail: usize = 0; + while (tail < n - head and old[old.len - 1 - tail] == new[new.len - 1 - tail]) tail += 1; + try std.testing.expectEqual(Span{ .head = head, .tail = tail }, spanOf(old, new)); + } +} + fn positionIn(gpa: std.mem.Allocator, f: *State, at: usize) struct { row: usize, col: usize } { const index = lineIndex(gpa, f) catch { const c = modal.positionAt(f.content, at); @@ -1492,14 +1535,18 @@ pub fn restoreSnap(pane: *Pane, f: *State, snap: Snapshot) void { pane.ensureCursorVisible(); } -fn pushHistory(gpa: std.mem.Allocator, slots: []Snapshot, len: *usize, snap: Snapshot) void { - if (len.* == slots.len) { +/// Bounded in steps and in bytes: the oldest go first, though the one +/// pushed always stays. +fn pushHistory(gpa: std.mem.Allocator, history: *History, slots: []Snapshot, len: *usize, snap: Snapshot) void { + while (len.* > 0 and (len.* == slots.len or history.bytes + snap.content.len > limits.undo_bytes)) { + history.bytes -= slots[0].content.len; slots[0].deinit(gpa); - std.mem.copyForwards(Snapshot, slots[0 .. slots.len - 1], slots[1..]); + std.mem.copyForwards(Snapshot, slots[0 .. len.* - 1], slots[1..len.*]); len.* -= 1; } slots[len.*] = snap; len.* += 1; + history.bytes += snap.content.len; } pub fn pushUndo(p: *Pardes, pane: *Pane) void { @@ -1510,9 +1557,12 @@ pub fn pushUndo(p: *Pardes, pane: *Pane) void { if (std.mem.eql(u8, last.content, f.content) and last.gen == history.gen) return; } const snap = snapshotOf(p.gpa, pane, f) orelse return; - pushHistory(p.gpa, &history.undo, &history.undo_len, snap); - for (history.redo[0..history.redo_len]) |item| item.deinit(p.gpa); + for (history.redo[0..history.redo_len]) |item| { + history.bytes -= item.content.len; + item.deinit(p.gpa); + } history.redo_len = 0; + pushHistory(p.gpa, history, &history.undo, &history.undo_len, snap); } pub fn undo(p: *Pardes, pane: *Pane) void { @@ -1520,9 +1570,10 @@ pub fn undo(p: *Pardes, pane: *Pane) void { const history = f.history; if (history.undo_len == 0) return; const current = snapshotOf(p.gpa, pane, f) orelse return; - pushHistory(p.gpa, &history.redo, &history.redo_len, current); + pushHistory(p.gpa, history, &history.redo, &history.redo_len, current); history.undo_len -= 1; const previous = history.undo[history.undo_len]; + history.bytes -= previous.content.len; // the file's text now setContent(p, f, previous.content); restoreName(p, pane, f, previous); restoreSnap(pane, f, previous); @@ -1533,9 +1584,10 @@ pub fn redo(p: *Pardes, pane: *Pane) void { const history = f.history; if (history.redo_len == 0) return; const current = snapshotOf(p.gpa, pane, f) orelse return; - pushHistory(p.gpa, &history.undo, &history.undo_len, current); + pushHistory(p.gpa, history, &history.undo, &history.undo_len, current); history.redo_len -= 1; const next = history.redo[history.redo_len]; + history.bytes -= next.content.len; // the file's text now setContent(p, f, next.content); restoreName(p, pane, f, next); restoreSnap(pane, f, next); @@ -2266,3 +2318,28 @@ pub fn paintWordSelection( } } } + +test "undo keeps its steps within a byte bound, the newest always" { + const gpa = std.testing.allocator; + const p = try Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 24 }); + defer p.deinit(); + const pane = try p.setTestFile(""); + const f = &pane.file.?; + const step = limits.undo_bytes / 4 + 1; + const text = try gpa.alloc(u8, step); + defer gpa.free(text); + for (0..8) |i| { + @memset(text, 'a' + @as(u8, @intCast(i))); + pushUndo(p, pane); + setContent(p, f, try gpa.dupe(u8, text)); + } + pushUndo(p, pane); + const history = f.history; + try std.testing.expect(history.undo_len > 0 and history.undo_len < 8); + try std.testing.expect(history.bytes <= limits.undo_bytes + step); + var sum: usize = 0; + for (history.undo[0..history.undo_len]) |snap| sum += snap.content.len; + try std.testing.expectEqual(sum, history.bytes); + undo(p, pane); + try std.testing.expectEqualStrings(text[0..1], f.content[0..1]); +} diff --git a/src/Registers.zig b/src/Registers.zig index 3f2486cd..5e23da5b 100644 --- a/src/Registers.zig +++ b/src/Registers.zig @@ -81,6 +81,7 @@ pub fn text(r: *Registers, gpa: std.mem.Allocator, name: u21) ?[]const u8 { const reg = r.get(name) orelse return null; if (reg.joined) |j| return j; var out: std.ArrayList(u8) = .empty; + defer out.deinit(gpa); // what a failed append left; empty once owned for (reg.values, 0..) |v, i| { if (i > 0 and !std.mem.endsWith(u8, reg.values[i - 1], "\n")) out.append(gpa, '\n') catch return null; out.appendSlice(gpa, v) catch return null; @@ -105,3 +106,15 @@ test "a register holds a value per range and joins them for one-text readers" { try std.testing.expectEqualStrings("x", r.text(gpa, 'a').?); try std.testing.expectEqualStrings("one\ntwo", r.text(gpa, default).?); } + +test "joining a register that runs out of memory partway leaks nothing" { + var r: Registers = .{}; + defer r.deinit(std.testing.allocator); + r.put(std.testing.allocator, default, "one", 0, 2, true); + r.put(std.testing.allocator, default, "two", 1, 2, false); + r.put(std.testing.allocator, default, &(@as([4096]u8, @splat('x'))), 1, 2, false); + // The first append allocates; growing for the 4096 bytes fails. + var failing = std.testing.FailingAllocator.init(std.testing.allocator, .{ .fail_index = 1, .resize_fail_index = 0 }); + try std.testing.expect(r.text(failing.allocator(), default) == null); + try std.testing.expectEqual(failing.allocated_bytes, failing.freed_bytes); +} diff --git a/src/ShaderBuild.zig b/src/ShaderBuild.zig index b50999a5..e5b6680b 100644 --- a/src/ShaderBuild.zig +++ b/src/ShaderBuild.zig @@ -76,7 +76,8 @@ const Result = union(enum) { spirv: struct { bytes: []u8, source: u64, animated: bool }, }; -/// One compile of the stale files, on its own thread. +/// One compile of the stale files, a task of its own: `deinit` cancels it +/// (glslc's wait is a cancelation point) rather than waiting out a glslc. const Job = struct { io: std.Io, gpa: std.mem.Allocator, @@ -86,7 +87,10 @@ const Job = struct { results: [max]Result = @splat(.none), len: usize = 0, done: std.atomic.Value(bool) = .init(false), - thread: std.Thread = undefined, + /// Set before the cancel: the files after the one it stopped are not + /// compiled either (a cancel is said once, to the wait it lands in). + stop: std.atomic.Value(bool) = .init(false), + future: std.Io.Future(void) = undefined, }; files: [max]File = @splat(.{}), @@ -105,7 +109,7 @@ pub fn sync(b: *Build, gpa: std.mem.Allocator, io: std.Io, core: *pardes.Pardes, var collected = false; if (b.job) |job| { if (!job.done.load(.acquire)) return false; - job.thread.join(); + job.future.await(job.io); b.job = null; b.collect(gpa, core, job); collected = true; @@ -148,7 +152,7 @@ pub fn sync(b: *Build, gpa: std.mem.Allocator, io: std.Io, core: *pardes.Pardes, job.len += 1; file.stale = false; } - job.thread = std.Thread.spawn(.{}, compile, .{job}) catch { + job.future = io.concurrent(compile, .{job}) catch { for (job.paths[0..job.len]) |path| gpa.free(path); gpa.destroy(job); return collected; @@ -174,7 +178,8 @@ pub fn view(b: *const Build, chain: *const Chain, out: *[max]Pass) []const Pass pub fn deinit(b: *Build, gpa: std.mem.Allocator, fd: c_int, watches: *file_watch.Table) void { if (b.job) |job| { - job.thread.join(); + job.stop.store(true, .release); + job.future.cancel(job.io); freeJob(gpa, job); } for (b.files[0..b.len]) |file| free(gpa, file); @@ -245,7 +250,10 @@ fn expand(buf: *[4096]u8, path: []const u8) []const u8 { /// build's own flags (build.zig compileGlsl), source on stdin, SPIR-V on /// stdout. fn compile(job: *Job) void { - for (job.paths[0..job.len], 0..) |path, i| job.results[i] = compileOne(job.gpa, job.io, path, job.known[i]); + for (job.paths[0..job.len], 0..) |path, i| { + if (job.stop.load(.acquire)) break; + job.results[i] = compileOne(job.gpa, job.io, path, job.known[i]); + } job.done.store(true, .release); if (job.wake.call) |call| call(job.wake.ctx); } @@ -378,7 +386,8 @@ test "a save that changes a file compiles it again, a failed one keeps the last fn run(build: *Build, c: *pardes.Pardes, f: c_int, w: *file_watch.Table) void { _ = build.sync(std.testing.allocator, std.testing.io, c, f, w, .{}); while (build.job) |job| { - while (!job.done.load(.acquire)) std.Thread.yield() catch {}; + job.future.await(std.testing.io); + job.done.store(true, .release); _ = build.sync(std.testing.allocator, std.testing.io, c, f, w, .{}); } } diff --git a/src/builtins.zig b/src/builtins.zig index cc7d1d4a..fdce534b 100644 --- a/src/builtins.zig +++ b/src/builtins.zig @@ -1971,6 +1971,9 @@ pub const WsDiagnostics = struct { /// Show outermost callers first, with signatures and call sites (gc). pub const Callstack = struct { pub fn run(c: Ctx) void { + // In +Callstack itself: the next page of what it shows. + if (c.pane.file) |f| if (f.output) |o| if (std.meta.eql(o.from, panes.Output.Origin{ .query = .callstack })) + return c.p.callstackMore(c.id); c.p.lspRequest(c.id, .callstack, ""); } }; diff --git a/src/colors.zig b/src/colors.zig index dd0d8938..28136975 100644 --- a/src/colors.zig +++ b/src/colors.zig @@ -840,7 +840,7 @@ pub fn loadThemeFile(p: *Pardes, generation: u32, bytes: []const u8, animate: bo @memcpy(settled[0..request.path.len], request.path); _ = p.theme_file_loaded.set(settled[0..request.path.len]); invalidateThemeDependentRasters(p); - _ = p.scratch.reset(.retain_capacity); + _ = p.scratch.reset(.{ .retain_with_limit = pardes.memory.limits.scratch_kept }); p.sync(); return true; } diff --git a/src/config.zig b/src/config.zig index a92aabb6..5251b1e5 100644 --- a/src/config.zig +++ b/src/config.zig @@ -1230,27 +1230,12 @@ pub const Runtime = struct { if (!state.shell.requested.set(value)) return false; state.shell.pending = true; }, - .tagline_size => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - const percent = std.fmt.parseInt(u16, text, 10) catch return false; - if (percent < tagline_percent_min or percent > tagline_percent_max) return false; - state.font.tagline_percent = @intCast(percent); - }, + .tagline_size => state.font.tagline_percent = @intCast(number(u16, argument, " \t\r\n", tagline_percent_min, tagline_percent_max) orelse return false), .window_opacity => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - for (text) |byte| if (!std.ascii.isDigit(byte)) return false; - const percent = std.fmt.parseInt(u8, text, 10) catch return false; - if (percent > 100) return false; - state.window_opacity = percent; + state.window_opacity = number(u8, argument, " \t\r\n", 0, 100) orelse return false; state.window_opacity_pending = true; }, - .window_blur => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - for (text) |byte| if (!std.ascii.isDigit(byte)) return false; - const percent = std.fmt.parseInt(u8, text, 10) catch return false; - if (percent > 100) return false; - state.window_blur = percent; - }, + .window_blur => state.window_blur = number(u8, argument, " \t\r\n", 0, 100) orelse return false, // Bare, it returns to the default directory. Else an absolute or // `~` path, to a directory there or to be made, that may be // written: a Dump there could never be, refused now. @@ -1264,10 +1249,7 @@ pub const Runtime = struct { if (!state.dump_dir.set(text)) return false; }, .message_ms => |which| { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - for (text) |byte| if (!std.ascii.isDigit(byte)) return false; - const ms = std.fmt.parseInt(u16, text, 10) catch return false; - if (ms > message_ms_max) return false; + const ms = number(u16, argument, " \t\r\n", 0, message_ms_max) orelse return false; switch (which) { inline else => |w| @field(state, "message_" ++ @tagName(w) ++ "_ms") = ms, } @@ -1330,30 +1312,10 @@ pub const Runtime = struct { std.meta.stringToEnum(LiftStyle, text) orelse return false; }, .motion => state.motion = stepped(animation.Motion.Flavour, state.motion, argument) orelse return false, - .inactive_dim => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n%"); - const percent = std.fmt.parseInt(u8, text, 10) catch return false; - if (percent > 100) return false; - state.inactive_dim = percent; - }, - .grip_width => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n%"); - const percent = std.fmt.parseInt(u16, text, 10) catch return false; - if (percent < 50 or percent > 300) return false; - state.grip_width = percent; - }, - .callstack_depth => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - const depth = std.fmt.parseInt(u8, text, 10) catch return false; - if (depth < 1 or depth > @import("lsp/lsp.zig").callstack_max_depth) return false; - state.callstack_depth = depth; - }, - .tab => { - const text = std.mem.trim(u8, argument orelse return false, " \t\r\n"); - const width = std.fmt.parseInt(u16, text, 10) catch return false; - if (width < 1 or width > 16) return false; - state.tab = width; - }, + .inactive_dim => state.inactive_dim = number(u8, argument, " \t\r\n%", 0, 100) orelse return false, + .grip_width => state.grip_width = number(u16, argument, " \t\r\n%", 50, 300) orelse return false, + .callstack_depth => state.callstack_depth = number(u8, argument, " \t\r\n", 1, @import("lsp/lsp.zig").callstack_max_depth) orelse return false, + .tab => state.tab = number(u16, argument, " \t\r\n", 1, 16) orelse return false, .theme, .font => return false, } return true; @@ -1411,6 +1373,16 @@ pub const Runtime = struct { return all[(at + 1) % all.len]; } + /// A setting's number: digits only (parseInt alone takes a sign and `_`), + /// trimmed of `trim`, within lo..hi. + fn number(comptime T: type, argument: ?[]const u8, trim: []const u8, lo: T, hi: T) ?T { + const text = std.mem.trim(u8, argument orelse return null, trim); + if (text.len == 0) return null; + for (text) |byte| if (!std.ascii.isDigit(byte)) return null; + const n = std.fmt.parseInt(T, text, 10) catch return null; + return if (n < lo or n > hi) null else n; + } + /// What `on`, `off` or nothing after a switch word asks for: null flips /// it, as a bare tagline word does. fn switchedTo(argument: ?[]const u8) error{NotOnOrOff}!?bool { @@ -2523,7 +2495,7 @@ test "CallstackDepth is bounded and keeps the last valid runtime limit" { try std.testing.expect(Runtime.apply(&state, Runtime.find("CallstackDepth").?, value)); try std.testing.expectEqual(try std.fmt.parseInt(u8, value, 10), state.callstack_depth); } - for ([_][]const u8{ "0", "65", "256", "-1", "eight", "" }) |value| { + for ([_][]const u8{ "0", "65", "256", "-1", "+8", "1_6", "eight", "" }) |value| { try std.testing.expect(!Runtime.apply(&state, Runtime.find("CallstackDepth").?, value)); try std.testing.expectEqual(@as(u8, 64), state.callstack_depth); } diff --git a/src/crash.zig b/src/crash.zig index 34dc1b3d..a432fb87 100644 --- a/src/crash.zig +++ b/src/crash.zig @@ -34,6 +34,9 @@ pub const name = "crashes"; /// A panic handler is the one caller that cannot check whether its input is /// still alive, so it does not borrow. var dir_buf: [1024]u8 = undefined; + +/// The crash file's bound (see `record`). +const max_log_bytes = 256 * 1024; var dir_len: usize = 0; /// Called where the launcher sets `Options.config_dir`. Ignores a path too long @@ -109,6 +112,9 @@ pub fn record(msg: []const u8) void { }, @as(std.c.mode_t, 0o600)); if (fd < 0) return; defer _ = std.c.close(fd); + // Bounded: a file past `max_log_bytes` starts over, so it keeps the + // newest records rather than every crash since it was first made. + if (std.c.lseek(fd, 0, std.c.SEEK.END) > max_log_bytes) _ = std.c.ftruncate(fd, 0); var w: std.Io.Writer = .fixed(&scratch); diff --git a/src/detached/server.zig b/src/detached/server.zig index d6a1f553..6c8d4352 100644 --- a/src/detached/server.zig +++ b/src/detached/server.zig @@ -107,17 +107,19 @@ const Mailbox = struct { status: ?[]u8 = null, }; - mutex: std.atomic.Mutex = .unlocked, + /// Taken by the workers and the loop; `io` is the session's (initAsync). + mutex: std.Io.Mutex = .init, + io: std.Io = undefined, batch: Batch = .{}, wake: [2]c_int = .{ -1, -1 }, fn post(box: *Mailbox, completion: Completion) void { - while (!box.mutex.tryLock()) std.atomic.spinLoopHint(); + box.mutex.lockUncancelable(box.io); // Tasks retain their slot until the owner consumes their one completion. std.debug.assert(box.batch.len < capacity); box.batch.items[box.batch.len] = completion; box.batch.len += 1; - box.mutex.unlock(); + box.mutex.unlock(box.io); box.signal(); } @@ -130,8 +132,8 @@ const Mailbox = struct { fn take(box: *Mailbox) Batch { var bytes: [128]u8 = undefined; while (libc.recv(box.wake[0], &bytes, bytes.len, 0) > 0) {} - while (!box.mutex.tryLock()) std.atomic.spinLoopHint(); - defer box.mutex.unlock(); + box.mutex.lockUncancelable(box.io); + defer box.mutex.unlock(box.io); const batch = box.batch; box.batch.len = 0; box.batch.status = null; @@ -184,6 +186,7 @@ pub const Session = struct { accept_paused_ms: i64 = 0, pub fn initAsync(s: *Session) !void { + s.mailbox.io = s.io; var pair: [2]c_int = undefined; if (libc.socketpair(libc.AF.UNIX, libc.SOCK.STREAM, 0, &pair) != 0) return error.SocketFailed; errdefer for (pair) |fd| { @@ -410,7 +413,7 @@ pub const Session = struct { s.core.update(.{ .lsp_resp = .{ .id = request.id, .rows = null } }); return s.core.reportError(request.pane, "lsp", err); }; - const future = s.io.concurrent(lspWorker, .{ s.worker_gpa, job, &s.mailbox }) catch |err| { + const future = s.io.concurrent(lspWorker, .{ s.io, s.worker_gpa, job, &s.mailbox }) catch |err| { job.free(s.worker_gpa); s.core.update(.{ .lsp_resp = .{ .id = request.id, .rows = null } }); return s.core.reportError(request.pane, "lsp", err); @@ -418,8 +421,8 @@ pub const Session = struct { s.lsp_task = .{ .id = request.id, .future = future }; } - fn lspWorker(gpa: std.mem.Allocator, job: *host_io.Lsp.Job, box: *Mailbox) anyerror!void { - host_io.Lsp.work(gpa, job, box, deliverLspRows); + fn lspWorker(io: std.Io, gpa: std.mem.Allocator, job: *host_io.Lsp.Job, box: *Mailbox) anyerror!void { + host_io.Lsp.work(io, gpa, job, box, deliverLspRows); } fn deliverLspRows(ctx: ?*anyopaque, id: u32, rows: ?[]u8) void { @@ -430,10 +433,10 @@ pub const Session = struct { fn lspStatus(ctx: ?*anyopaque, text: []const u8) void { const s = of(ctx); const copy = s.worker_gpa.dupe(u8, text) catch return; - while (!s.mailbox.mutex.tryLock()) std.atomic.spinLoopHint(); + s.mailbox.mutex.lockUncancelable(s.io); if (s.mailbox.batch.status) |old| s.worker_gpa.free(old); s.mailbox.batch.status = copy; - s.mailbox.mutex.unlock(); + s.mailbox.mutex.unlock(s.io); s.mailbox.signal(); } @@ -529,7 +532,8 @@ pub const Session = struct { if (pane >= s.ptys.len) return; const pt = &s.ptys[pane]; if (pt.fd < 0) return; - if (pt.out.items.len > pty_backlog) { + // The bound holds with this write in: queued and new together. + if (pt.out.items.len + bytes.len > pty_backlog) { var mbuf: [256]u8 = undefined; const text = std.fmt.bufPrint( &mbuf, @@ -1145,6 +1149,9 @@ pub const Session = struct { s.queue(c, bytes); } + /// A client's queued output is at most `out_backlog + wire.max_payload`: + /// past the backlog it is closed, and one frame lands on top of it first + /// (a frame bigger than the backlog must still go out whole). fn queue(s: *Session, c: *Client, bytes: []const u8) void { if (c.out.items.len > out_backlog) return s.close(c, .backlog); s.account(); diff --git a/src/dump.zig b/src/dump.zig index 76a6de85..67934e0a 100644 --- a/src/dump.zig +++ b/src/dump.zig @@ -394,17 +394,6 @@ test "oversized recognized output argument is rejected before restore" { ); } -pub fn writeFile(io: std.Io, gpa: std.mem.Allocator, path: []const u8, state: State) !void { - try validate(state); - var out: std.Io.Writer.Allocating = .init(gpa); - defer out.deinit(); - try std.zon.stringify.serialize(state, .{ .whitespace = true }, &out.writer); - - var file = try std.Io.Dir.createFileAbsolute(io, path, .{}); - defer file.close(io); - try file.writeStreamingAll(io, out.written()); -} - pub const Parsed = struct { value: State, arena: std.heap.ArenaAllocator, diff --git a/src/edit.zig b/src/edit.zig index 466b5cdb..55365692 100644 --- a/src/edit.zig +++ b/src/edit.zig @@ -119,7 +119,15 @@ pub fn typeToTty(p: *Pardes, id: usize, pane: *const Pane, text: []const u8) voi for (panes.terminal.encodePaste(pane, cp)) |part| if (part.len > 0) p.emitWrite(id, part); } -pub fn applyPaste(p: *Pardes, bytes: []const u8) void { +pub fn applyPaste(p: *Pardes, whole: []const u8) void { + // Every shell's paste comes through here, bounded: what is past the + // limit is left out, at a character's start, and said. + const limit = @import("memory.zig").limits.max_paste_bytes; + const bytes = if (whole.len > limit) whole[0..modal.graphemeStart(whole, limit)] else whole; + if (whole.len > limit) { + var said: [96]u8 = undefined; + p.setMessage(p.active, std.fmt.bufPrint(&said, "paste cut at its limit, {d} KiB: the rest left out", .{limit / 1024}) catch "paste cut at its limit"); + } // A prompt answering for a header takes the paste, as it takes keys. const asking = if (p.panes[p.active]) |active| active.prompt != .none else false; if (!asking) if (tagline.focusedHeader(p)) |t| { @@ -466,8 +474,10 @@ fn editText(p: *Pardes, t: *Text, lo: i32, hi: i32, col: i32) ?panes.EditText { fn editTextEol(p: *Pardes, t: *Text, b: Bounds) ?panes.EditText { const eb = editText(p, t, b.lo_row, b.hi_row, -1) orelse return null; const r: usize = @intCast(@max(0, b.hi_row - eb.row0)); - if (r + 1 < modal.lineCount(eb.text)) return eb; - if (b.hi_col < @as(i32, @intCast(modal.lineSlice(eb.text, r).len))) return eb; + // A file's own line count and line, from its index. + const lines = if (indexedLine(p, t, eb.text, 0) != null) panes.File.nlines(p.gpa, &t.pane().?.file.?) else modal.lineCount(eb.text); + if (r + 1 < lines) return eb; + if (b.hi_col < @as(i32, @intCast(lineAt(p, t, eb.text, r).line.len))) return eb; return editText(p, t, b.lo_row, b.hi_row + 1, -1); } @@ -476,6 +486,47 @@ pub fn setEditText(p: *Pardes, t: *Text, new: []u8) void { setEditTextEnd(p, t, new, null); } +const LineAt = struct { start: usize, line: []const u8 }; + +/// `text`'s line count: the file's index when `text` is the pane's file. +fn lineCountOf(p: *Pardes, t: *Text, text: []const u8) usize { + if (indexedLine(p, t, text, 0) != null) return panes.File.nlines(p.gpa, &t.pane().?.file.?); + return modal.cursorLineCount(text); +} + +/// The row offset `off` is on: a search of the file's index when `text` is +/// the pane's file, else the newlines before it counted (modal.lineAtOffset). +fn rowOf(p: *Pardes, t: *Text, text: []const u8, off: usize) usize { + if (indexedLine(p, t, text, 0) != null) { + const index = panes.File.lineIndex(p.gpa, &t.pane().?.file.?) catch return modal.lineAtOffset(text, off); + return std.sort.upperBound(usize, index, @min(off, text.len), struct { + fn order(key: usize, item: usize) std.math.Order { + return std.math.order(key, item); + } + }.order) -| 1; + } + return modal.lineAtOffset(text, off); +} + +/// Line `row` of `text` and where it starts: through the file's line index +/// when `text` is the pane's file (a lookup), else counted from the top. +/// Finding line 94000 by counting was most of a keystroke in a 7 MB file. +fn lineAt(p: *Pardes, t: *Text, text: []const u8, row: usize) LineAt { + return indexedLine(p, t, text, row) orelse .{ .start = modal.lineStartOffset(text, row), .line = modal.lineSlice(text, row) }; +} + +/// lineAt's lookup, null when `text` is not the pane's file or `row` is past its last line. +fn indexedLine(p: *Pardes, t: *Text, text: []const u8, row: usize) ?LineAt { + if (t.what != .body) return null; + const pane = t.pane() orelse return null; + const f = if (pane.file) |*f| f else return null; + if (f.content.ptr != text.ptr or f.content.len != text.len) return null; + if (row >= panes.File.nlines(p.gpa, f)) return null; + const start = @min(panes.File.lineStart(p.gpa, f, row), text.len); + const end = std.mem.indexOfScalarPos(u8, text, start, '\n') orelse text.len; + return .{ .start = start, .line = text[start..end] }; +} + /// Where text `s` put in at `c` ends: `g.`'s place after typing it. fn insertedEnd(c: modal.Cursor, s: []const u8) modal.Cursor { const nl = std.mem.lastIndexOfScalar(u8, s, '\n') orelse return .{ .row = c.row, .col = c.col + s.len }; @@ -486,6 +537,13 @@ fn insertedEnd(c: modal.Cursor, s: []const u8) modal.Cursor { /// rows) when the caller knows it, as typing does: then nothing scans the /// whole text to find it. Null, it is found by comparing the texts. pub fn setEditTextEnd(p: *Pardes, t: *Text, new: []u8, known_end: ?modal.Cursor) void { + setEditTextSpan(p, t, new, known_end, null); +} + +/// `setEditTextEnd` with what a file's splice left alone at either end, when +/// the caller knows it (typing): the line index is then not found again by +/// comparing the whole old and new texts, which a keystroke in a 7 MB file did. +fn setEditTextSpan(p: *Pardes, t: *Text, new: []u8, known_end: ?modal.Cursor, span: ?panes.File.Span) void { if (t.what == .tag) { // A tag is edited as it is shown, but only what follows its // prefix is its own: an edit that changed the prefix is refused. @@ -511,16 +569,13 @@ pub fn setEditTextEnd(p: *Pardes, t: *Text, new: []u8, known_end: ?modal.Cursor) const pane = t.pane() orelse return retireEditText(p, new); // what changed ended where the texts agree again from their ends const old: []const u8, const row0: i32 = if (pane.file) |f| .{ f.content, 0 } else if (pane.ovl) |o| .{ o.text, o.row } else .{ "", 0 }; - const end = known_end orelse found: { - // ponytail: O(file), for the edits that do not say where they ended - // (paste, a pipe, s): typing passes its end and never comes here. - const front = std.mem.indexOfDiff(u8, old, new) orelse @min(old.len, new.len); - var back: usize = 0; - while (back < @min(old.len, new.len) - front and old[old.len - 1 - back] == new[new.len - 1 - back]) back += 1; - break :found modal.positionAt(new, new.len - back); - }; + // ponytail: O(file), for the edits that do not say what they changed + // (paste, a pipe, d, s): compared once, block by block, here; the same + // span then spares the line index a second comparison. + const found = span orelse if (known_end == null or pane.file != null) panes.File.spanOf(old, new) else null; + const end = known_end orelse modal.positionAt(new, new.len - found.?.tail); t.last_edit = .{ .row = @as(i32, @intCast(end.row)) + row0, .col = @intCast(end.col) }; - if (pane.file) |*f| return panes.File.setContent(p, f, new); + if (pane.file) |*f| return panes.File.setContentSpan(p, f, new, found); panes.terminal.setEditText(p, pane, new); } @@ -815,8 +870,10 @@ fn insertEdit(p: *Pardes, t: *Text, key: Key) void { const text = eb.text; const c = modal.Cursor{ .row = @intCast(@max(0, t.cur_row - eb.row0)), .col = @intCast(@max(0, t.cur_col)) }; if (key.text.len > 0 and !key.ctrl and !key.alt) { - const new = modal.insertAt(p.gpa, text, c, key.text) catch return; - setEditTextEnd(p, t, new, insertedEnd(c, key.text)); + // insertAt's offset (col clamped to the line), the line found once. + const at = if (indexedLine(p, t, text, c.row)) |l| l.start + @min(c.col, l.line.len) else modal.insertOffset(text, c); + const new = modal.spliceAlloc(p.gpa, text, at, at, key.text) catch return; + setEditTextSpan(p, t, new, insertedEnd(c, key.text), .{ .head = at, .tail = text.len - at }); for (key.text) |ch| { if (ch == '\n') { t.cur_row += 1; @@ -918,7 +975,8 @@ fn insertEdit(p: *Pardes, t: *Text, key: Key) void { }, Key.backspace => { if (t.cur_col > 0) { - const line = modal.lineSlice(text, c.row); + const found = lineAt(p, t, text, c.row); + const line = found.line; var prev = modal.prevGrapheme(line, c.col); // in leading blanks, back to the previous indent stop (helix // dedent): a whole unit when on one, else the part past it @@ -931,8 +989,12 @@ fn insertEdit(p: *Pardes, t: *Text, key: Key) void { prev = c.col; while (drop > 0 and prev > 0 and line[prev - 1] == ' ') : (drop -= 1) prev -= 1; } - const new = modal.deleteSpan(p.gpa, text, .{ .row = c.row, .col = prev }, c) catch return; - setEditTextEnd(p, t, new, .{ .row = c.row, .col = prev }); + // deleteSpan's splice, the line found once. + const s = found.start + @min(prev, line.len); + const e = found.start + @min(c.col, line.len); + const new = if (e > s) modal.spliceAlloc(p.gpa, text, s, e, "") catch return else p.gpa.dupe(u8, text) catch return; + const span: panes.File.Span = if (e > s) .{ .head = s, .tail = text.len - e } else .{ .head = text.len, .tail = 0 }; + setEditTextSpan(p, t, new, .{ .row = c.row, .col = prev }, span); t.cur_col = @intCast(prev); t.cur_pinned = true; t.show(); @@ -1479,26 +1541,33 @@ pub fn normalJoin(p: *Pardes, t: *Text, select_space: bool) void { const text = eb.text; var rs: [Text.max_selections]modal.Selection = undefined; const got = t.ranges(text, eb.row0, &rs); - const nlines = modal.cursorLineCount(text); + // Lines, their count and rows from the file's index when it has one: + // counted from the top, a J in a 7 MB file was 8 ms. + const nlines = lineCountOf(p, t, text); const arena = p.scratch.allocator(); - const chs = arena.alloc(TextChange, nlines) catch return; - var chs_len: usize = 0; + var changes: std.ArrayList(TextChange) = .empty; var next: usize = 0; // lines below this one are joined already for (rs[0..got.n]) |r| { - const span = Text.rangeLineSpan(text, r); + // Text.rangeLineSpan, through rowOf + const lo = @min(r.anchor, r.head); + const hi = @max(r.anchor, r.head); + const span = .{ .start = rowOf(p, t, text, lo), .end = rowOf(p, t, text, if (lo == hi) hi else @max(modal.prevGrapheme(text, hi), lo)) }; var end = span.end; if (span.start == end) end = @min(end + 1, nlines - 1); var l = @max(span.start, next); while (l < end) : (l += 1) { - const from = modal.lineEndOffset(text, l); - var to = if (l + 1 >= nlines) text.len else modal.lineStartOffset(text, l + 1); + const here = lineAt(p, t, text, l); + const from = here.start + here.line.len; + var to = if (l + 1 >= nlines) text.len else lineAt(p, t, text, l + 1).start; while (to < text.len and (text[to] == ' ' or text[to] == '\t')) to += 1; - const sep: []const u8 = if (to == modal.lineEndOffset(text, @min(l + 1, nlines - 1))) "" else " "; - chs[chs_len] = .{ .from = from, .to = to, .ins = sep }; - chs_len += 1; + const last = lineAt(p, t, text, @min(l + 1, nlines - 1)); + const sep: []const u8 = if (to == last.start + last.line.len) "" else " "; + changes.append(arena, .{ .from = from, .to = to, .ins = sep }) catch return; } next = @max(next, end); } + const chs = changes.items; + const chs_len = chs.len; if (chs_len == 0) return; pushUndo(p, t); const new = applyChanges(p, text, chs[0..chs_len]) catch return; @@ -2459,3 +2528,72 @@ test "a paste into a terminal is xterm's: bracketed when the program asked, cont p.update(.{ .paste = "x\x1b[201~y\n" }); try std.testing.expectEqualStrings("\x1b[200~x [201~y\n\x1b[201~", try take(p, &out)); } + +test "a paste past the limit is cut at it, and says so" { + const gpa = std.testing.allocator; + const p = try Pardes.init(gpa, .{ .tty_only = true, .cols = 80, .rows = 24 }); + defer p.deinit(); + const pane = try p.setTestFile(""); + pane.body.mode = .insert; + const limit = @import("memory.zig").limits.max_paste_bytes; + const big = try gpa.alloc(u8, limit + 100); + defer gpa.free(big); + @memset(big, 'x'); + applyPaste(p, big); + try std.testing.expect(pane.file.?.content.len <= limit); + try std.testing.expect(pane.file.?.content.len > 0); + try std.testing.expect(std.mem.indexOf(u8, pane.msg[0..pane.msg_len], "paste cut") != null); +} + +test "typing and backspace through the line index keep the text and the index as a rebuild has them" { + const gpa = std.testing.allocator; + const p = try Pardes.init(gpa, .{ .tty_only = true }); + defer p.deinit(); + while (p.nextEffect()) |_| {} + const pane = try p.setTestFile("one\n two\n\nthree four\n"); + pane.body.mode = .insert; + var prng: std.Random.DefaultPrng = .init(11); + const random = prng.random(); + for (0..300) |_| { + _ = panes.File.lineIndex(gpa, &pane.file.?) catch unreachable; + const f = &pane.file.?; + const row = random.uintLessThan(usize, panes.File.nlines(gpa, f)); + const line = modal.lineSlice(f.content, row); + pane.body.cur_row = @intCast(row); + pane.body.cur_col = @intCast(random.uintLessThan(usize, line.len + 1)); + pane.body.cur_pinned = true; + if (random.boolean()) { + insertKey(p, &pane.body, .{ .cp = 'x', .text = "x" }); + } else { + insertKey(p, &pane.body, .{ .cp = Key.backspace }); + } + // the index the edits kept, against one built from the text + const kept = try gpa.dupe(usize, try panes.File.lineIndex(gpa, &pane.file.?)); + defer gpa.free(kept); + gpa.free(pane.file.?.line_starts); + pane.file.?.line_starts = &.{}; + try std.testing.expectEqualSlices(usize, try panes.File.lineIndex(gpa, &pane.file.?), kept); + } +} + +test "J joins a line with the next, its indent dropped, a space between unless the next is empty" { + const gpa = std.testing.allocator; + const cases = [_]struct { text: []const u8, row: i32, rows: i32 = 0, want: []const u8 }{ + .{ .text = "a\n b\nc\n", .row = 0, .want = "a b\nc\n" }, + .{ .text = "a\n\nc\n", .row = 0, .want = "a\nc\n" }, + .{ .text = "a\nb", .row = 1, .want = "a\nb" }, + .{ .text = "one\n\ttwo\nthree\nfour\n", .row = 0, .rows = 2, .want = "one two three\nfour\n" }, + }; + for (cases) |case| { + const p = try Pardes.init(gpa, .{ .tty_only = true }); + defer p.deinit(); + while (p.nextEffect()) |_| {} + const pane = try p.setTestFile(case.text); + pane.body.mode = .normal; + pane.body.cur_row = case.row; + pane.body.cur_col = 0; + if (case.rows > 0) pane.body.vsel = .{ .active = true, .row = case.row + case.rows, .col = 0, .explicit = true }; + normalJoin(p, &pane.body, false); + try std.testing.expectEqualStrings(case.want, pane.file.?.content); + } +} diff --git a/src/exec.zig b/src/exec.zig index d03a4a48..c9f304b0 100644 --- a/src/exec.zig +++ b/src/exec.zig @@ -913,6 +913,12 @@ pub fn commandDone(p: *Pardes, id: usize, status: ?u8) void { pane.repl = null; pane.command_status = status; pane.body.mode = .normal; + // Look's `git show`: its output went to a +Pager, the +Diff now; only a + // failure is worth the pane. + if (pane.command_pages) { + pane.command_pages = false; + if (status == 0) return p.removePane(id, null) catch {}; + } var buf: [16]u8 = undefined; const code = if (status) |n| std.fmt.bufPrint(&buf, "{d}", .{n}) catch "?" else "?"; var said: [32]u8 = undefined; diff --git a/src/fs.zig b/src/fs.zig index 061d4225..032ab9c0 100644 --- a/src/fs.zig +++ b/src/fs.zig @@ -1087,17 +1087,17 @@ pub const platform_has_fs = !pardes.isolated and switch (pardes.platform) { const find_max_hits = 512; /// What a Find or Grep walk passed over, said under its hits: directories -/// it might not open, files read only in part, and caps it stopped at. +/// it might not open, and caps it stopped at. pub const WalkNotes = struct { skipped: usize = 0, - /// Files longer than grep_max_bytes, searched in their first part only. - partial: usize = 0, hits_cut: bool = false, + /// A pane search's matching lines it stopped at, its rows filling + /// look.search_text_output_bytes; 0 if none. + rows_cut: usize = 0, /// The cap on entries walked it stopped at, 0 if none. walk_cut: usize = 0, pub const max_hits = find_max_hits; - pub const max_file_bytes = grep_max_bytes; pub const max_files = grep_max_files; pub const max_depth = find_max_depth; @@ -1105,7 +1105,7 @@ pub const WalkNotes = struct { pub fn write(n: WalkNotes, out: []u8) usize { var w = std.Io.Writer.fixed(out); if (n.hits_cut) w.print("cut at {d} hits\n", .{max_hits}) catch {}; - if (n.partial > 0) w.print("{d} file{s} read only in part (first {d} KiB)\n", .{ n.partial, if (n.partial == 1) "" else "s", max_file_bytes / 1024 }) catch {}; + if (n.rows_cut > 0) w.print("cut at {d} lines\n", .{n.rows_cut}) catch {}; if (n.walk_cut > 0) w.print("walk cut at {d} entries\n", .{n.walk_cut}) catch {}; if (n.skipped > 0) w.print("{d} director{s} skipped: permission denied\n", .{ n.skipped, if (n.skipped == 1) "y" else "ies" }) catch {}; return w.buffered().len; @@ -1124,7 +1124,8 @@ pub const WalkNotes = struct { }; const find_max_depth = 16; const find_max_steps = 100_000; -pub const search_max_output_bytes = pardes.MAX_PANES * find_max_hits * (4096 + 320); +/// One root's Find or Grep rows at most: its hits, each a path and a line. +pub const search_root_output_bytes = find_max_hits * (4096 + 320); const find_skip = [_][]const u8{ ".git", ".jj", "target", "node_modules", @@ -1194,9 +1195,10 @@ const grep_max_files = 20_000; const GrepResult = struct { bytes: usize, hits: usize }; -fn grepText(path: []const u8, text: []const u8, pat: []const u8, out: []u8, budget: usize) GrepResult { +/// `text` starts at line `first_line` + 1 of its file. +fn grepText(path: []const u8, text: []const u8, pat: []const u8, out: []u8, budget: usize, first_line: usize) GrepResult { var result: GrepResult = .{ .bytes = 0, .hits = 0 }; - var line: usize = 0; + var line: usize = first_line; var it = std.mem.splitScalar(u8, text, '\n'); while (it.next()) |raw| { line += 1; @@ -1220,7 +1222,7 @@ pub fn grep(arena: std.mem.Allocator, gpa: std.mem.Allocator, dir: []const u8, b if (!platform_has_fs) { for (sources) |s| { if (hits >= find_max_hits or written == out.len) break; - const result = grepText(s.path, s.contents, pat, out[written..], find_max_hits - hits); + const result = grepText(s.path, s.contents, pat, out[written..], find_max_hits - hits, 0); hits += result.hits; written += result.bytes; } @@ -1276,30 +1278,39 @@ pub fn grep(arena: std.mem.Allocator, gpa: std.mem.Allocator, dir: []const u8, b const path_z = std.fmt.bufPrintSentinel(&pathbuf, "{s}", .{path}, 0) catch return error.PathTooLong; const fd = libc.open(path_z, .{ .ACCMODE = .RDONLY, .CLOEXEC = true, .NONBLOCK = true }); if (fd < 0) continue; + defer _ = libc.close(fd); + // The whole file, a buffer at a time: each searched through its last + // whole line, the rest carried into the next. + const shown = lsp.rel(home, path); var len: usize = 0; - var readable = true; - while (len < buf.len) { - const n = libc.read(fd, buf[len..].ptr, buf.len - len); - if (n < 0) { - if (libc.errno(n) == .INTR) continue; - readable = false; - break; + var lines_before: usize = 0; + var first = true; + file: while (true) { + var eof = false; + while (len < buf.len) { + const n = libc.read(fd, buf[len..].ptr, buf.len - len); + if (n < 0) { + if (libc.errno(n) == .INTR) continue; + break :file; + } + if (n == 0) { + eof = true; + break; + } + len += @intCast(n); } - if (n == 0) break; - len += @intCast(n); + if (first and std.mem.indexOfScalar(u8, buf[0..@min(len, 1024)], 0) != null) break; + first = false; + // A line longer than the buffer is searched in buffer-sized parts. + const end = if (eof) len else if (std.mem.lastIndexOfScalar(u8, buf[0..len], '\n')) |nl| nl + 1 else len; + const result = grepText(shown, buf[0..end], pat, out[written..], find_max_hits - hits, lines_before); + hits += result.hits; + written += result.bytes; + if (eof or hits >= find_max_hits or written == out.len) break; + lines_before += std.mem.count(u8, buf[0..end], "\n"); + std.mem.copyForwards(u8, buf[0 .. len - end], buf[end..len]); + len -= end; } - // Read to its cap with more behind it: searched in part, and said. - var more: [1]u8 = undefined; - const longer = readable and len == buf.len and libc.read(fd, &more, 1) > 0; - _ = libc.close(fd); - if (!readable) continue; - const text = buf[0..len]; - if (std.mem.indexOfScalar(u8, text[0..@min(len, 1024)], 0) != null) continue; - if (longer) notes.partial += 1; - const shown = lsp.rel(home, path); - const result = grepText(shown, text, pat, out[written..], find_max_hits - hits); - hits += result.hits; - written += result.bytes; if (hits >= find_max_hits) notes.hits_cut = true; } return written; @@ -1366,14 +1377,14 @@ test "a directory the walk may not open is skipped and counted: Grep and Find ke try std.testing.expectEqual(@as(usize, 1), notes.skipped); } -test "Grep says when a cap hid what it might have found: a file read in part, the hits cut" { +test "Grep reads a long file to its end, and says when its hits were cut" { if (!platform_has_fs) return; const gpa = std.testing.allocator; var tmp = std.testing.tmpDir(.{}); defer tmp.cleanup(); var base_buf: [std.fs.max_path_bytes]u8 = undefined; const dir = base_buf[0..try tmp.dir.realPath(std.testing.io, &base_buf)]; - // The match past the first 256 KiB of a long file. + // A match past the first buffer of a long file, on its own line number. const long = try gpa.alloc(u8, grep_max_bytes + 64); defer gpa.free(long); @memset(long, 'x'); @@ -1386,8 +1397,9 @@ test "Grep says when a cap hid what it might have found: a file read in part, th defer gpa.free(out); var notes: WalkNotes = .{}; var n = try grep(arena.allocator(), gpa, dir, dir, "needle", out, ¬es); - try std.testing.expectEqual(@as(usize, 0), n); - try std.testing.expectEqual(@as(usize, 1), notes.partial); + var want_buf: [64]u8 = undefined; + const want = try std.fmt.bufPrint(&want_buf, "long.txt:{d}:", .{std.mem.count(u8, long, "\n")}); + try std.testing.expect(std.mem.startsWith(u8, out[0..n], want)); // More hits than the cap. var many: std.ArrayList(u8) = .empty; defer many.deinit(gpa); @@ -1398,7 +1410,7 @@ test "Grep says when a cap hid what it might have found: a file read in part, th try std.testing.expect(notes.hits_cut); var said: [256]u8 = undefined; const lines = said[0..notes.write(&said)]; - try std.testing.expectEqualStrings("cut at 512 hits\n1 file read only in part (first 256 KiB)\n", lines); + try std.testing.expectEqualStrings("cut at 512 hits\n", lines); try std.testing.expect(WalkNotes.isNote("cut at 512 hits") and WalkNotes.isNote("2 directories skipped: permission denied") and !WalkNotes.isNote("src/2 directories.txt")); } diff --git a/src/gui/gui.zig b/src/gui/gui.zig index f9921111..9cd37b6b 100644 --- a/src/gui/gui.zig +++ b/src/gui/gui.zig @@ -828,18 +828,13 @@ fn emitTouchClick(t: *Touch, in: *Input, point: TouchNormPoint, button: pardes.M in.post(.{ .mouse = .{ .button = button, .kind = .release, .col = cell.col, .row = cell.row, .tag_hit = cell.tag_hit, .body_hit = cell.body_hit } }); } -fn normCell(norm: f32, win: f32, cell: f32) u16 { - const px = std.math.clamp(norm, 0.0, 1.0) * win; - return @intFromFloat(@max(0, @floor(px / cell))); -} - const Pty = struct { fd: c_int, pid: libc.pid_t, serial: u32, kill_at: i64 = 0, - reader: ?std.Thread = null, - stop: [2]c_int = .{ -1, -1 }, + /// Its reader, blocked in an Io read: `closePty` cancels it. + reader: ?std.Io.Future(void) = null, /// A command pane's child, watched to its exit (host_io.watchExit); /// the pty stays open until both its end of file and its exit. cmd: host_io.CommandWatch = .{}, @@ -876,31 +871,6 @@ const Msg = union(enum) { } }; -const LspWorkers = struct { - const capacity = 16; - active: std.atomic.Value(usize) = .init(0), - - fn start(workers: *LspWorkers) bool { - var count = workers.active.load(.monotonic); - while (count < capacity) - count = workers.active.cmpxchgWeak(count, count + 1, .monotonic, .monotonic) orelse return true; - return false; - } - - fn finish(workers: *LspWorkers) void { - const previous = workers.active.fetchSub(1, .release); - std.debug.assert(previous > 0); - } - - fn wait(workers: *LspWorkers) void { - var spins: u8 = 0; - while (workers.active.load(.acquire) != 0) { - spins +%= 1; - if (spins == 0) std.Thread.yield() catch {} else std.atomic.spinLoopHint(); - } - } -}; - const PipeTasks = selection_pipe.Tasks; const queue_capacity = 512; @@ -918,10 +888,13 @@ const MessageBatch = struct { const Queue = struct { gpa: std.mem.Allocator, lsp_allocator: std.mem.Allocator, - lsp_workers: *LspWorkers, + /// The query in flight; the UI thread's alone (spawnLsp, teardown). + lsp_task: ?host_io.Lsp.Task = null, sdl_wake: bool, // wake a blocking SDL_WaitEventTimeout on cross-thread push - mutex: libc.pthread_mutex_t = .{}, - space: libc.pthread_cond_t = .{}, + /// The process's Io: its terminal readers wait for space on `space`. + io: std.Io, + mutex: std.Io.Mutex = .init, + space: std.Io.Condition = .init, items: [queue_capacity]Msg = undefined, head: usize = 0, len: usize = 0, @@ -934,13 +907,11 @@ const Queue = struct { completion_len: usize = 0, fn lock(q: *Queue) void { - // unreachable failures: pthread calls on this queue's own initialised, non-recursive - // mutex and condition fail only on misuse (EINVAL, EDEADLK), never on input. - std.debug.assert(libc.pthread_mutex_lock(&q.mutex) == .SUCCESS); + q.mutex.lockUncancelable(q.io); } fn unlock(q: *Queue) void { - std.debug.assert(libc.pthread_mutex_unlock(&q.mutex) == .SUCCESS); + q.mutex.unlock(q.io); } fn wake(q: *Queue) void { @@ -973,14 +944,14 @@ const Queue = struct { }; if (matches) q.removeAt(i).deinit(q.gpa, q.lsp_allocator) else i += 1; } - std.debug.assert(libc.pthread_cond_broadcast(&q.space) == .SUCCESS); + q.space.broadcast(q.io); } fn pushOutput(q: *Queue, pane: u8, gen: u32, bytes: []u8) bool { q.lock(); while (!q.closed and q.readers[pane] == gen and q.len >= output_capacity) { q.waiting += 1; - std.debug.assert(libc.pthread_cond_wait(&q.space, &q.mutex) == .SUCCESS); + q.space.waitUncancelable(q.io, &q.mutex); q.waiting -= 1; } if (q.closed or q.readers[pane] != gen) { @@ -1088,7 +1059,7 @@ const Queue = struct { @memcpy(batch.items[batch.len..][0..q.completion_len], q.completions[0..q.completion_len]); batch.len += q.completion_len; q.completion_len = 0; - std.debug.assert(libc.pthread_cond_broadcast(&q.space) == .SUCCESS); + q.space.broadcast(q.io); } fn discardCompletions(q: *Queue) void { @@ -1104,7 +1075,7 @@ const Queue = struct { q.closed = true; q.files_changed = false; q.readers = @splat(null); - std.debug.assert(libc.pthread_cond_broadcast(&q.space) == .SUCCESS); + q.space.broadcast(q.io); q.unlock(); q.discardCompletions(); q.lock(); @@ -1121,15 +1092,12 @@ const Queue = struct { fn deinit(q: *Queue) void { q.close(); std.debug.assert(q.waiting == 0); - std.debug.assert(libc.pthread_cond_destroy(&q.space) == .SUCCESS); - std.debug.assert(libc.pthread_mutex_destroy(&q.mutex) == .SUCCESS); } }; test "GUI completion survives a full output queue" { const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false, .lsp_id = 7 }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false, .lsp_id = 7 }; defer queue.deinit(); queue.acceptReader(0, 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ @@ -1185,8 +1153,7 @@ const PtyTests = struct { test "GUI PTY backpressure retains byte order and tail before EOF" { const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); queue.acceptReader(0, 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ .pane = 0, .gen = 1, .bytes = try gpa.dupe(u8, "before") } }); @@ -1220,8 +1187,7 @@ test "GUI PTY backpressure retains byte order and tail before EOF" { test "GUI PTY backpressure cancellation and close release owned producers" { const gpa = std.testing.allocator; for ([_]bool{ false, true }) |close| { - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); queue.acceptReader(0, 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ .pane = 0, .gen = 1, .bytes = try gpa.dupe(u8, "before") } }); @@ -1252,8 +1218,7 @@ test "GUI PTY backpressure cancellation and close release owned producers" { test "GUI PTY EOF reserve cannot evict terminal output" { const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); for (0..pardes.MAX_PANES) |pane| queue.acceptReader(@intCast(pane), 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ .pane = 0, .gen = 1, .bytes = try gpa.dupe(u8, "before") } }); @@ -1269,8 +1234,7 @@ test "GUI PTY reader delivers a real shell tail before EOF and joins on close" { const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true }); defer core.deinit(); - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); var ptys: [pardes.MAX_PANES]?Pty = @splat(null); var gens: [pardes.MAX_PANES]u32 = @splat(1); @@ -1320,8 +1284,7 @@ test "GUI PTY Restore joins a real reader waiting for queue space" { var core = try pardes.Pardes.init(gpa, .{ .tty_only = true }); defer core.deinit(); try dump.dumpState(core); - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); queue.acceptReader(1, 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ .pane = 1, .gen = 1, .bytes = try gpa.dupe(u8, "other pane") } }); @@ -1365,8 +1328,7 @@ test "GUI PTY deletion joins an idle reader and retains only its owned child for const gpa = std.testing.allocator; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true }); defer core.deinit(); - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); var ptys: [pardes.MAX_PANES]?Pty = @splat(null); var gens: [pardes.MAX_PANES]u32 = @splat(1); @@ -1397,8 +1359,7 @@ test "GUI PTY deletion joins an idle reader and retains only its owned child for test "GUI PTY file watcher stops and joins without closing its watched descriptor" { if (!file_watch.supported) return error.SkipZigTest; const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); const fd = file_watch.init(true); if (fd < 0) return error.WatchInitFailed; @@ -1422,8 +1383,7 @@ test "GUI completion failure survives backlog and leaves the document unchanged" pane.body.cur_col = 1; core.lspRequest(core.active, .completion, ""); const id = core.lsp_wait.?.id; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false, .lsp_id = id }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false, .lsp_id = id }; defer queue.deinit(); queue.acceptReader(0, 1); for (0..output_capacity) |_| queue.push(.{ .output = .{ @@ -1452,8 +1412,7 @@ test "GUI completion failure survives backlog and leaves the document unchanged" test "GUI completions retain their arrival order across LSP and pipes" { const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false, .lsp_id = 2 }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false, .lsp_id = 2 }; defer queue.deinit(); queue.push(.{ .pipe = .{ .id = 1, .success = true, .outputs = &.{} } }); queue.push(.{ .lsp = .{ .id = 2, .rows = try gpa.dupe(u8, "second") } }); @@ -1469,8 +1428,7 @@ test "GUI completions retain their arrival order across LSP and pipes" { test "GUI completion reset frees queued payloads and rejects late LSP workers" { const gpa = std.testing.allocator; - var workers: LspWorkers = .{}; - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = gpa, .lsp_workers = &workers, .sdl_wake = false, .lsp_id = 9 }; + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false, .lsp_id = 9 }; defer queue.deinit(); queue.acceptReader(0, 1); queue.push(.{ .output = .{ .pane = 0, .gen = 1, .bytes = try gpa.dupe(u8, "retained") } }); @@ -1523,29 +1481,22 @@ fn lspStatusSink(ctx: ?*anyopaque, text: []const u8) void { q.push(.{ .lsp_status = copy }); } -fn readPtyThread(gpa: std.mem.Allocator, fd: c_int, stop: c_int, pane: u8, gen: u32, q: *Queue) void { +fn readPty(io: std.Io, gpa: std.mem.Allocator, fd: c_int, pane: u8, gen: u32, q: *Queue) void { + const pty: std.Io.File = .{ .handle = fd, .flags = .{ .nonblocking = false } }; var buf: [0x10000]u8 = undefined; var failure: ?anyerror = null; while (true) { - var fds = [_]libc.pollfd{ - .{ .fd = stop, .events = libc.POLL.IN, .revents = 0 }, - .{ .fd = fd, .events = libc.POLL.IN, .revents = 0 }, + const n = pty.readStreaming(io, &.{&buf}) catch |err| switch (err) { + // A pty whose child has gone answers EIO: its end of file. + error.EndOfStream, error.InputOutput => break, + error.Canceled => return, + else => { + failure = err; + break; + }, }; - if (libc.poll(&fds, fds.len, -1) < 0) { - if (libc.errno(-1) == .INTR) continue; - failure = error.PollFailed; - break; - } - if (fds[0].revents != 0) return; - if (fds[1].revents == 0) continue; - const n = libc.read(fd, &buf, buf.len); - if (n < 0) { - if (libc.errno(n) == .INTR) continue; - if (libc.errno(n) != .IO) failure = error.ReadFailed; - break; - } - if (n == 0) break; - const bytes = gpa.dupe(u8, buf[0..@intCast(n)]) catch |err| { + if (n == 0) continue; + const bytes = gpa.dupe(u8, buf[0..n]) catch |err| { failure = err; break; }; @@ -1566,14 +1517,9 @@ fn stopPipe() ![2]c_int { fn spawnReader(gpa: std.mem.Allocator, pt: *Pty, pane: u8, gen: u32, q: *Queue) !void { std.debug.assert(pt.reader == null and pt.fd >= 0); - const stop = try stopPipe(); - errdefer for (stop) |fd| { - _ = libc.close(fd); - }; q.acceptReader(pane, gen); errdefer q.cancelReader(pane, gen); - pt.reader = try std.Thread.spawn(.{}, readPtyThread, .{ gpa, pt.fd, stop[0], pane, gen, q }); - pt.stop = stop; + pt.reader = try q.io.concurrent(readPty, .{ q.io, gpa, pt.fd, pane, gen, q }); } fn watchThread(fd: c_int, stop: c_int, q: *Queue) void { @@ -1592,9 +1538,8 @@ fn watchThread(fd: c_int, stop: c_int, q: *Queue) void { } } -fn lspThread(lsp_allocator: std.mem.Allocator, workers: *LspWorkers, job: *host_io.Lsp.Job, q: *Queue) void { - defer workers.finish(); - host_io.Lsp.work(lsp_allocator, job, q, pushLspRows); +fn lspThread(io: std.Io, lsp_allocator: std.mem.Allocator, job: *host_io.Lsp.Job, q: *Queue) anyerror!void { + host_io.Lsp.work(io, lsp_allocator, job, q, pushLspRows); } fn pushLspRows(ctx: ?*anyopaque, id: u32, rows: ?[]u8) void { @@ -1602,107 +1547,52 @@ fn pushLspRows(ctx: ?*anyopaque, id: u32, rows: ?[]u8) void { q.push(.{ .lsp = .{ .id = id, .rows = rows } }); } -fn spawnLsp(core: *pardes.Pardes, q: *Queue, e: host_io.Lsp.Request) void { +fn spawnLsp(core: *pardes.Pardes, io: std.Io, q: *Queue, e: host_io.Lsp.Request) void { const lsp_allocator = q.lsp_allocator; q.lock(); q.lsp_id = e.id; q.discardLsp(); q.unlock(); - if (!q.lsp_workers.start()) { - core.update(.{ .lsp_resp = .{ .id = e.id, .rows = null } }); - return core.reportError(e.pane, "lsp", error.WorkersBusy); + // Its waits are cancelation points: this join is short, and what it + // still delivers carries an id the queue now rejects. + if (q.lsp_task) |*old| { + old.future.cancel(io) catch {}; + q.lsp_task = null; } const job = host_io.Lsp.snapshot(lsp_allocator, core, e) catch |err| { - q.lsp_workers.finish(); core.update(.{ .lsp_resp = .{ .id = e.id, .rows = null } }); return core.reportError(e.pane, "lsp", err); }; - const th = std.Thread.spawn(.{}, lspThread, .{ lsp_allocator, q.lsp_workers, job, q }) catch |err| { - q.lsp_workers.finish(); + const future = io.concurrent(lspThread, .{ io, lsp_allocator, job, q }) catch |err| { job.free(lsp_allocator); core.update(.{ .lsp_resp = .{ .id = e.id, .rows = null } }); return core.reportError(e.pane, "lsp", err); }; - th.detach(); + q.lsp_task = .{ .id = e.id, .future = future }; } -test "GUI LSP worker limit rejects before allocation and recovers after owned workers exit" { +test "GUI LSP query answers, and a newer one cancels it" { const gpa = std.testing.allocator; + const io = std.testing.io; const core = try pardes.Pardes.init(gpa, .{ .tty_only = true }); defer core.deinit(); - const pane = try core.setTestFile("abc"); - pane.body.cur_col = 1; - var workers: LspWorkers = .{}; - var gate: std.Io.Event = .unset; - var threads: [LspWorkers.capacity]?std.Thread = @splat(null); - defer { - gate.set(std.testing.io); - for (threads) |thread| if (thread) |owned| owned.join(); - workers.wait(); - } - for (&threads) |*thread| { - try std.testing.expect(workers.start()); - thread.* = std.Thread.spawn(.{}, struct { - fn run(active: *LspWorkers, ready: *std.Io.Event) void { - defer active.finish(); - ready.waitUncancelable(std.testing.io); - } - }.run, .{ &workers, &gate }) catch |err| { - workers.finish(); - return err; - }; - } - try std.testing.expectEqual(@as(usize, LspWorkers.capacity), workers.active.load(.monotonic)); - try std.testing.expect(!workers.start()); - var failing = std.testing.FailingAllocator.init(gpa, .{ .fail_index = 0 }); - var queue: Queue = .{ .gpa = gpa, .lsp_allocator = failing.allocator(), .lsp_workers = &workers, .sdl_wake = false }; + _ = try core.setTestFile("abc"); + var queue: Queue = .{ .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, .sdl_wake = false }; defer queue.deinit(); - core.lspRequest(core.active, .completion, ""); - const rejected = core.lsp_wait.?.id; - queue.lsp_id = rejected -% 1; - queue.push(.{ .lsp = .{ .id = queue.lsp_id.?, .rows = try gpa.dupe(u8, "obsolete queued result") } }); - spawnLsp(core, &queue, .{ - .id = rejected, - .kind = .completion, - .pane = @intCast(core.active), - .offset = 1, - .arg = "", - }); - try std.testing.expect(core.lsp_wait == null); - try std.testing.expectEqualStrings("abc", pane.file.?.content); - try std.testing.expectEqual(@as(i32, 1), pane.body.cur_col); - try std.testing.expect(std.mem.indexOf(u8, pane.msg[0..pane.msg_len], "workers busy") != null); - try std.testing.expectEqual(@as(usize, 0), failing.alloc_index); - try std.testing.expect(!failing.has_induced_failure); - try std.testing.expectEqual(rejected, queue.lsp_id.?); - try std.testing.expectEqual(@as(usize, 0), queue.completion_len); - queue.push(.{ .lsp = .{ .id = rejected -% 1, .rows = try gpa.dupe(u8, "obsolete late result") } }); - try std.testing.expectEqual(@as(usize, 0), queue.completion_len); - try std.testing.expectEqual(@as(usize, LspWorkers.capacity), workers.active.load(.monotonic)); - - gate.set(std.testing.io); - workers.wait(); - try std.testing.expectEqual(@as(usize, 0), workers.active.load(.monotonic)); - queue.lsp_allocator = gpa; - core.lspRequest(core.active, .status, ""); - const accepted = core.lsp_wait.?.id; - spawnLsp(core, &queue, .{ - .id = accepted, - .kind = .status, - .pane = @intCast(core.active), - .offset = 0, - .arg = "", - }); - workers.wait(); + for (0..2) |_| { + core.lspRequest(core.active, .status, ""); + spawnLsp(core, io, &queue, .{ .id = core.lsp_wait.?.id, .kind = .status, .pane = @intCast(core.active), .offset = 0, .arg = "" }); + } + const last = core.lsp_wait.?.id; + try queue.lsp_task.?.future.await(io); + queue.lsp_task = null; var batch = queue.take(); defer for (batch.slice()) |msg| msg.deinit(gpa, gpa); try std.testing.expectEqual(@as(usize, 1), batch.len); - try std.testing.expectEqual(accepted, batch.items[0].lsp.id); + try std.testing.expectEqual(last, batch.items[0].lsp.id); try std.testing.expect(batch.items[0].lsp.rows != null); - core.update(.{ .lsp_resp = .{ .id = accepted, .rows = batch.items[0].lsp.rows } }); + core.update(.{ .lsp_resp = .{ .id = last, .rows = batch.items[0].lsp.rows } }); try std.testing.expect(core.lsp_wait == null); - try std.testing.expectEqual(@as(usize, 0), workers.active.load(.monotonic)); - try std.testing.expectEqualStrings("abc", pane.file.?.content); } fn pipeThread(io: std.Io, gpa: std.mem.Allocator, job: *selection_pipe.Job, q: *Queue) anyerror!void { @@ -1748,11 +1638,10 @@ test "GUI worker setup failures finish matching LSP and pipe requests" { var failing = std.testing.FailingAllocator.init(gpa, .{ .fail_index = if (failure == 0) 0 else std.math.maxInt(usize), }); - var workers: LspWorkers = .{}; var queue: Queue = .{ + .io = std.testing.io, .gpa = gpa, .lsp_allocator = failing.allocator(), - .lsp_workers = &workers, .sdl_wake = false, }; defer queue.deinit(); @@ -1768,7 +1657,7 @@ test "GUI worker setup failures finish matching LSP and pipe requests" { if (failure == 0) { queue.lsp_id = req.id -% 1; queue.push(.{ .lsp = .{ .id = queue.lsp_id.?, .rows = try gpa.dupe(u8, "old queued result") } }); - spawnLsp(core, &queue, req); + spawnLsp(core, std.testing.io, &queue, req); try std.testing.expectEqual(req.id, queue.lsp_id.?); try std.testing.expectEqual(@as(usize, 0), queue.completion_len); queue.push(.{ .lsp = .{ .id = req.id -% 1, .rows = try gpa.dupe(u8, "late old result") } }); @@ -1791,7 +1680,7 @@ test "GUI worker setup failures finish matching LSP and pipe requests" { spawnPipe(core, failing_io, failing.allocator(), &queue, &tasks, pipe_id); try std.testing.expect(core.pipe.wait == null); try std.testing.expectEqualStrings("abc", pane.file.?.content); - try std.testing.expectEqual(@as(usize, 0), workers.active.load(.monotonic)); + try std.testing.expect(queue.lsp_task == null); try std.testing.expectEqual(@as(usize, 0), tasks.len); try std.testing.expectEqual(@as(usize, 0), queue.len); } @@ -2071,22 +1960,6 @@ fn fixedCellLayout(g: *const Gui) CellLayout { return .{ .w = @floatFromInt(g.cell_w), .h = @floatFromInt(g.cell_h), .x_off = 0, .y_off = 0 }; } -fn compactTaglineLayout(g: *const Gui, origin_col: f32) CellLayout { - const body_w: f32 = @floatFromInt(g.cell_w); - const tag_w: f32 = @floatFromInt(g.tagline_width); - return .{ - .w = tag_w, - .h = @floatFromInt(g.cell_h), - .x_off = origin_col * (body_w - tag_w), - .y_off = 0, - }; -} - -fn taglineLayoutForCell(g: *const Gui, col: u16) CellLayout { - // Canonical tag cells are physical grips; text is drawn by TagLayer. - return compactTaglineLayout(g, @floatFromInt(col)); -} - const boxContains = pardes.animation.Box.contains; /// What one tier draws, or one track in tiers 2 and 3 (docs/render-pipeline.md @@ -2663,15 +2536,14 @@ fn localSession( var ptys: [pardes.MAX_PANES]?Pty = @splat(null); var gens: [pardes.MAX_PANES]u32 = @splat(0); - var lsp_workers: LspWorkers = .{}; var queue: Queue = .{ + .io = io, .gpa = gpa, .lsp_allocator = allocs.lsp, - .lsp_workers = &lsp_workers, .sdl_wake = true, }; defer { - lsp_workers.wait(); + if (queue.lsp_task) |*task| task.future.cancel(io) catch {}; queue.deinit(); } var pipe_tasks: PipeTasks = .{}; @@ -3035,15 +2907,14 @@ fn runGrid(init: std.process.Init, opts_in: pardes.Options) !void { var ptys: [pardes.MAX_PANES]?Pty = @splat(null); var gens: [pardes.MAX_PANES]u32 = @splat(0); - var lsp_workers: LspWorkers = .{}; var queue: Queue = .{ + .io = io, .gpa = gpa, .lsp_allocator = allocs.lsp, - .lsp_workers = &lsp_workers, .sdl_wake = false, }; defer { - lsp_workers.wait(); + if (queue.lsp_task) |*task| task.future.cancel(io) catch {}; queue.deinit(); } var pipe_tasks: PipeTasks = .{}; @@ -4195,13 +4066,12 @@ const Shell = struct { fn closePty(s: *Shell, pane: u8) void { const pt = if (s.ptys[pane]) |*pt| pt else return; if (pt.fd < 0) return; + // Out of the queue first, which lets go a reader waiting for room + // in it; then its read is canceled (or, at its end, it is joined). s.queue.cancelReader(pane, s.gens[pane]); - if (pt.reader) |thread| { - _ = host_io.writeFd(pt.stop[1], "x"); - thread.join(); + if (pt.reader) |*reader| { + reader.cancel(s.queue.io); // the Io it was started on (spawnReader) pt.reader = null; - for (pt.stop) |fd| _ = libc.close(fd); - pt.stop = .{ -1, -1 }; } _ = libc.close(pt.fd); pt.fd = -1; @@ -4848,11 +4718,10 @@ test "the headless grid host round-trips a yank back as a paste" { var ptys: [pardes.MAX_PANES]?Pty = @splat(null); var gens: [pardes.MAX_PANES]u32 = @splat(0); - var lsp_workers: LspWorkers = .{}; var queue: Queue = .{ + .io = std.testing.io, .gpa = gpa, .lsp_allocator = gpa, - .lsp_workers = &lsp_workers, .sdl_wake = false, }; defer queue.deinit(); @@ -4933,7 +4802,7 @@ fn openLink(ctx: ?*anyopaque, url: []const u8) void { fn lsp(ctx: ?*anyopaque, req: host_io.Lsp.Request) void { const s = shellOf(ctx); - if (s.threads_ok) return spawnLsp(s.core, s.queue, req); + if (s.threads_ok) return spawnLsp(s.core, s.io, s.queue, req); s.core.update(.{ .lsp_resp = .{ .id = req.id, .rows = null } }); s.core.reportError(req.pane, "lsp", error.WorkersUnavailable); } @@ -5198,6 +5067,12 @@ const Glide = struct { below: []pardes.Cell = &.{}, nabove: u16 = 0, nbelow: u16 = 0, + /// The core's lines a scroll moved, not yet measured; and what the + /// glide in flight has moved, in lines and in the picture's rows. A + /// wrapped line is several rows, and the thumb counts lines. + pending_lines: f32 = 0, + lines_moved: f32 = 0, + rows_moved: f32 = 0, /// The Motion flavour the last loop step read. motion: pardes.animation.Motion = .{}, @@ -5211,6 +5086,16 @@ const Glide = struct { glide.pending = false; glide.nabove = 0; glide.nbelow = 0; + glide.pending_lines = 0; + glide.lines_moved = 0; + glide.rows_moved = 0; + } + + /// The glide's lag in the core's lines: its rows, scaled by the lines + /// they carried. + fn lagLines(glide: *const Glide) f32 { + if (glide.rows_moved == 0) return glide.offset; + return glide.offset * glide.lines_moved / glide.rows_moved; } fn deinit(glide: *Glide, gpa: std.mem.Allocator) void { @@ -5270,6 +5155,9 @@ const Glide = struct { break :found kk; } else 0; if (shift == 0) return glide.stop(); + glide.rows_moved += @floatFromInt(shift); + glide.lines_moved += glide.pending_lines; + glide.pending_lines = 0; // The rows that left: past the top going down, the bottom going up. if (!glide.ensure(gpa)) return glide.stop(); const keep = if (glide.dir > 0) &glide.above else &glide.below; @@ -5326,6 +5214,8 @@ fn stepGlide(g: *Gui, core: *pardes.Pardes, gpa: std.mem.Allocator) void { glide.offset = 0; glide.nabove = 0; glide.nbelow = 0; + glide.lines_moved = 0; + glide.rows_moved = 0; } } const notch = g.scroll_notch; @@ -5357,6 +5247,8 @@ fn stepGlide(g: *Gui, core: *pardes.Pardes, gpa: std.mem.Allocator) void { } var left = notch; var moved = false; + const first = pane.scroll(); + defer glide.pending_lines += @floatFromInt(@abs(pane.scroll() - first)); while (left != 0) { const down = left > 0; left += if (down) -1 else 1; @@ -9347,7 +9239,7 @@ fn paneDecor(g: *Gui, gpa: std.mem.Allocator, regions: []const pardes.Region, gr // In pixels, not rows, and with the picture while it glides // or lags behind the core's rows (both in rows of the body). const lag: f32 = if (g.glide.offset != 0 and g.glide.pane == rail.owner) - g.glide.offset + g.glide.lagLines() else if (g.scroll_pane != null and g.scroll_pane.? == rail.owner) g.scroll_lag else 0; const track_px = body_y1 - y0; const size = @min(track_px, @max(layout.h, rail.thumb_span * track_px)); @@ -9983,6 +9875,12 @@ test "the scrollbar thumb moves by the pixel and follows a glide" { g.glide.pane = core.active; g.glide.offset = -1; try std.testing.expectApproxEqAbs(at10, try thumb.top(&g, core), 1e-3); + // A wrapped line: the picture moved two rows for the one line, and the + // thumb lags the line, not two. + g.glide.lines_moved = 1; + g.glide.rows_moved = 2; + g.glide.offset = -2; + try std.testing.expectApproxEqAbs(at10, try thumb.top(&g, core), 1e-3); } test "the rule between stacked panes stays in the tag band's slack, at any tagline size" { @@ -10581,12 +10479,11 @@ fn appendWorkspacePet(g: *Gui, core: ?*const pardes.Pardes, surface: *const pard return; } // Read only the rendered global row. No pane/column tags, status text or - // mouse targets are changed, and editing the workspace parks the pet. + // mouse targets are changed; editing the workspace tag leaves it be. var occupied: usize = 0; for (surface.cells[0..surface.cols], 0..) |*cell, i| { if (cellCodepoint(cell) != ' ') occupied = i + 1; } - const editing = p.header_focus and p.header_column == null; const band_top: f32 = @floatFromInt(taglineBandOffset(0, builder.win_h, g.cell_h, g.tagline_height)); const region = if (occupied >= surface.cols) null else pet.lane( occupied, @@ -10594,7 +10491,6 @@ fn appendWorkspacePet(g: *Gui, core: ?*const pardes.Pardes, surface: *const pard builder.win_w, band_top, @floatFromInt(g.tagline_height), - editing, ); // SDL's regular pump already wakes at least every 16 ms. Quantization in // sample keeps this independent of input, render rate and scene effects, @@ -10602,16 +10498,24 @@ fn appendWorkspacePet(g: *Gui, core: ?*const pardes.Pardes, surface: *const pard const pose = g.pet_state.sample(kind, c.SDL_GetTicks(), region) orelse return; const ink = overlayRgb(p.chromeTheme().tag_fg); const eye = overlayRgb(p.chromeTheme().tag_bg); + // Behind the cursor: the overlay is over the cells, so the pixels a + // cursor covers are left out. + var boxes: [max_cursors]CursorBox = undefined; + const cursors = boxes[0..cursorBoxes(g, surface, fixedCellLayout(g), builder.win_h, &boxes)]; for (0..pet.height) |y| for (0..pet.width) |x| { const color = pet.pixel(kind, pose, x, y); if (color == 0) continue; const x0 = pose.x + @as(f32, @floatFromInt(x)) * pose.scale; const y0 = pose.y + @as(f32, @floatFromInt(y)) * pose.scale; + const covered = for (cursors) |box| { + if (x0 < box.x1 and x0 + pose.scale > box.x0 and y0 < box.y1 and y0 + pose.scale > box.y0) break true; + } else false; + if (covered) continue; builder.addRect(x0, y0, x0 + pose.scale, y0 + pose.scale, if (color == 2) eye else ink); }; } -test "pet overlay stays beyond workspace text and vanishes while its tag is edited" { +test "pet overlay stays beyond workspace text and keeps going while its tag is edited" { std.testing.refAllDecls(pet); const core = try pardes.Pardes.init(std.testing.allocator, .{ .cols = 80, .rows = 24 }); defer core.deinit(); @@ -10638,10 +10542,13 @@ test "pet overlay stays beyond workspace text and vanishes while its tag is edit try std.testing.expect(x >= 20 * 8 - 0.01); try std.testing.expect(y >= 0 and y < 20); } + const before = g.pet_state; pardes.tagline.enterHeader(core, null); builder.len = 0; appendWorkspacePet(&g, core, &surface, &builder); - try std.testing.expectEqual(@as(usize, 0), builder.len); + try std.testing.expect(builder.len > 0); + try std.testing.expectEqual(before.tick, g.pet_state.tick); + try std.testing.expectEqual(before.x, g.pet_state.x); } fn cursorColor(chrome: *const pardes.Chrome, cell: *const pardes.Cell, role: pardes.FontRole) [3]u8 { diff --git a/src/gui/pet.zig b/src/gui/pet.zig index a249d70f..7a13ca91 100644 --- a/src/gui/pet.zig +++ b/src/gui/pet.zig @@ -13,8 +13,8 @@ const hop_ticks = 8; /// Only the trailing blank part of the global tag is eligible. The caller /// supplies its last occupied cell, including any wide-grapheme continuation. -pub fn lane(occupied: usize, cell_w: f32, screen_w: f32, top: f32, band_h: f32, editing: bool) ?Lane { - if (editing or cell_w <= 0 or band_h < height + 2) return null; +pub fn lane(occupied: usize, cell_w: f32, screen_w: f32, top: f32, band_h: f32) ?Lane { + if (cell_w <= 0 or band_h < height + 2) return null; const scale = @min(3, @floor((band_h - 2) / height)); const left = @as(f32, @floatFromInt(occupied + 2)) * cell_w; const right = screen_w - 2 * cell_w - width * scale; @@ -32,11 +32,10 @@ pub const State = struct { frame: u8 = 0, rng: std.Random.DefaultPrng = undefined, + /// No lane (the tag's text has filled it) hides the pet but keeps + /// where it was and what it was doing. pub fn sample(s: *State, kind: Kind, now_ms: u64, area: ?Lane) ?Pose { - const region = area orelse { - s.* = .{}; - return null; - }; + const region = area orelse return null; if (kind == .off) { s.* = .{}; return null; @@ -142,11 +141,10 @@ pub fn pixel(kind: Kind, pose: Pose, x_in: usize, y: usize) u2 { }; } -test "pet stays in trailing workspace blanks and disappears during editing" { - try std.testing.expect(lane(20, 8, 200, 0, 16, false) == null); - try std.testing.expect(lane(3, 8, 200, 0, 16, true) == null); - try std.testing.expect(lane(3, 8, 200, 0, 10, false) == null); - const region = lane(3, 8, 200, 2, 20, false).?; +test "pet stays in trailing workspace blanks, and a lost lane hides it without a reset" { + try std.testing.expect(lane(20, 8, 200, 0, 16) == null); + try std.testing.expect(lane(3, 8, 200, 0, 10) == null); + const region = lane(3, 8, 200, 2, 20).?; try std.testing.expect(region.left >= 5 * 8); var state: State = .{}; var moved = false; @@ -159,12 +157,15 @@ test "pet stays in trailing workspace blanks and disappears during editing" { reversed = reversed or pose.left; } try std.testing.expect(moved and reversed); + const kept = state; try std.testing.expect(state.sample(.cat, 300000, null) == null); + try std.testing.expectEqual(kept.x, state.x); + try std.testing.expectEqual(kept.tick, state.tick); try std.testing.expect(state.sample(.off, 300100, region) == null); } test "pet clock is fixed rate and sprites bounded in both directions" { - const region = lane(1, 8, 400, 0, 16, false).?; + const region = lane(1, 8, 400, 0, 16).?; var state: State = .{}; const start = state.sample(.frog, 1000, region).?; try std.testing.expectEqualDeep(start, state.sample(.frog, 1099, region).?); @@ -180,7 +181,7 @@ test "pet clock is fixed rate and sprites bounded in both directions" { } test "pet rests at random spots and the frog travels only in the air" { - const region = lane(1, 8, 800, 0, 16, false).?; + const region = lane(1, 8, 800, 0, 16).?; var state: State = .{}; var last = state.sample(.frog, 0, region).?; var rested_midway = false; @@ -195,7 +196,7 @@ test "pet rests at random spots and the frog travels only in the air" { } test "pet initial idle visibly animates without walking or leaving its tile" { - const region = lane(1, 8, 400, 0, 16, false).?; + const region = lane(1, 8, 400, 0, 16).?; for ([_]Kind{ .cat, .frog }) |kind| { var state: State = .{}; const initial = state.sample(kind, 0, region).?; diff --git a/src/host_io.zig b/src/host_io.zig index c593551b..2c736e20 100644 --- a/src/host_io.zig +++ b/src/host_io.zig @@ -203,11 +203,14 @@ pub const Lsp = struct { } others.deinit(gpa); } + var copied: usize = 0; for (core.panes, 0..) |slot, i| { const other = slot orelse continue; if (i == req.pane) continue; const f = other.file orelse continue; if (f.output != null) continue; + if (copied + f.content.len > pardes.memory.limits.lsp_snapshot_bytes) continue; + copied += f.content.len; const local = filesystem.localPath(f.path) orelse continue; const other_path = try gpa.dupe(u8, local); const other_source = gpa.dupeSentinel(u8, f.content, 0) catch |err| { @@ -224,13 +227,19 @@ pub const Lsp = struct { // Null is failure; the receiver frees non-null rows with the job's allocator. pub const Deliver = *const fn (ctx: ?*anyopaque, id: u32, rows: ?[]u8) void; - pub fn work(gpa: std.mem.Allocator, job: *Job, ctx: ?*anyopaque, deliver: Deliver) void { + pub fn work(io: std.Io, gpa: std.mem.Allocator, job: *Job, ctx: ?*anyopaque, deliver: Deliver) void { defer job.free(gpa); - var arena: std.heap.ArenaAllocator = .init(gpa); - defer arena.deinit(); + // The query's memory is the fixed region memory.zig reserved at + // startup: a hard bound, reused by each query in turn. + const region = pardes.memory.takeLspQuery() orelse { + deliver(ctx, job.id, null); + return; + }; + var arena: std.heap.FixedBufferAllocator = .init(region); + defer pardes.memory.giveLspQuery(arena.end_index); var out: std.Io.Writer.Allocating = .init(gpa); defer out.deinit(); - pardes.lsp.query(gpa, arena.allocator(), .{ + pardes.lsp.query(io, gpa, arena.allocator(), .{ .kind = job.kind, .path = job.path, .source = job.source, @@ -240,6 +249,7 @@ pub const Lsp = struct { .others = job.others, .callstack_depth = job.callstack_depth, .callstack_corners = job.callstack_corners, + .budget = &arena, }, &out.writer) catch { deliver(ctx, job.id, null); return; @@ -367,7 +377,7 @@ pub const Lsp = struct { .offset = 0, .arg = "", }); - work(gpa, job, null, Sink.take); + work(std.testing.io, gpa, job, null, Sink.take); // `status` is the one kind that answers with no file and no cursor, which // is what makes it assertable here without a language server on the box. @@ -416,7 +426,7 @@ pub const Lsp = struct { .arg = "renamed", }); if (fail) failing.fail_index = failing.alloc_index; - work(failing.allocator(), job, &sink, Sink.take); + work(std.testing.io, failing.allocator(), job, &sink, Sink.take); try std.testing.expectEqual(fail, sink.failed); try std.testing.expect(core.lsp_wait == null); if (fail) { @@ -482,7 +492,7 @@ pub const Lsp = struct { defer core.deinit(); const job = try snapshot(gpa, core, .{ .id = 37, .kind = .status, .pane = @intCast(core.active), .offset = 0, .arg = "" }); var sink: Sink = .{}; - work(gpa, job, &sink, Sink.take); + work(std.testing.io, gpa, job, &sink, Sink.take); if (sink.rows) |rows| gpa.free(rows); try std.testing.expect(allocator.failed); try std.testing.expectEqual(@as(usize, 1), sink.calls); @@ -1179,7 +1189,7 @@ pub const Shell = struct { if (builtin.os.tag == .windows) return; var saved_buf: [std.fs.max_path_bytes:0]u8 = @splat(0); const saved: ?[:0]const u8 = if (std.c.getenv("XDG_RUNTIME_DIR")) |v| std.fmt.bufPrintSentinel(&saved_buf, "{s}", .{std.mem.span(v)}, 0) catch null else null; - defer _ = if (saved) |v| setenv("XDG_RUNTIME_DIR", v.ptr, 1) else unsetenv("XDG_RUNTIME_DIR"); + defer _ = if (saved) |v| setenv("XDG_RUNTIME_DIR", v.ptr, 1) else testUnsetenv("XDG_RUNTIME_DIR"); var tmp = std.testing.tmpDir(.{}); defer tmp.cleanup(); @@ -1213,7 +1223,7 @@ pub const Shell = struct { try std.testing.expectEqualStrings(bash_rc, readSmall(c.bashPath().?, &buf) orelse return error.ReadFailed); // No runtime dir: private files, one per owner, gone at teardown. - _ = unsetenv("XDG_RUNTIME_DIR"); + _ = testUnsetenv("XDG_RUNTIME_DIR"); var d = PromptFiles.init(); var e = PromptFiles.init(); defer e.deinit(); @@ -1294,6 +1304,44 @@ pub const ChildEnv = struct { "TERM_PROGRAM=" ++ pardes.config.child_term_program, }; + /// What this editor tells its children of itself (exportPaneEnv, `$winid`), + /// kept here rather than set into this process's environment: std's Io read + /// that block at startup, and setenv/unsetenv rewrite it in place under it + /// (a spawn then scans a null inside it). `build` puts each exported one in + /// place of an inherited variable of its name; a cleared one hides it; one + /// never exported is inherited as it is. + pub const exported_names = [_][]const u8{ "PARDES_PID", "PARDES_9P", "PARDES_PANE", "PARDES_MOUNT", "winid" }; + var exported_state: [exported_names.len]enum { inherited, cleared, set } = @splat(.inherited); + var exported_line: [exported_names.len][pardes.memory.limits.host_path_cap + 64:0]u8 = undefined; + + /// Set (or, null, clear) what children are told for `var_name`. + pub fn exportVar(var_name: []const u8, value: ?[]const u8) void { + const i = exportedIndex(var_name).?; + const v = value orelse { + exported_state[i] = .cleared; + return; + }; + const line = std.fmt.bufPrintSentinel(&exported_line[i], "{s}={s}", .{ var_name, v }, 0) catch { + exported_state[i] = .cleared; + return; + }; + exported_state[i] = .set; + _ = line; + } + + /// What children are told for `var_name`, as `exportVar` left it; null + /// when cleared or never exported. + pub fn exported(var_name: []const u8) ?[]const u8 { + const i = exportedIndex(var_name) orelse return null; + if (exported_state[i] != .set) return null; + return std.mem.span(@as([*:0]const u8, &exported_line[i]))[var_name.len + 1 ..]; + } + + fn exportedIndex(var_name: []const u8) ?usize { + for (exported_names, 0..) |n, i| if (std.mem.eql(u8, n, var_name)) return i; + return null; + } + /// Wide enough for any real environment; `build` answers null rather than /// truncate, because a child missing half its variables is a worse bug /// than one missing TERM, and the caller then execs with `environ` intact. @@ -1318,7 +1366,8 @@ pub const ChildEnv = struct { if (std.mem.eql(u8, inherited, name(mine))) break true; } else false; if (shadowed) continue; - if (n + own.len + 1 > slots.len) return null; + if (exportedIndex(inherited)) |e| if (exported_state[e] != .inherited) continue; + if (n + own.len + exported_names.len + 1 > slots.len) return null; slots[n] = entry; n += 1; } @@ -1336,10 +1385,14 @@ pub const ChildEnv = struct { if (std.mem.eql(u8, name(entry.?), name(default))) break true; } else false; if (set) continue; - if (n + own.len + 4 > slots.len) return null; + if (n + own.len + exported_names.len + 4 > slots.len) return null; slots[n] = default; n += 1; } + for (&exported_line, exported_state) |*line, state| if (state == .set) { + slots[n] = line; + n += 1; + }; for (own) |mine| { slots[n] = mine; n += 1; @@ -1411,14 +1464,14 @@ test "the child environment replaces the launcher's terminal identity exactly on defer if (restore) |t| { _ = Shell.setenv("TERM", t.ptr, 1); } else { - _ = unsetenv("TERM"); + _ = testUnsetenv("TERM"); }; // A launcher TERM must be shadowed rather than duplicated, and the rest of // the environment must arrive untouched. try std.testing.expectEqual(@as(c_int, 0), Shell.setenv("TERM", "dumb", 1)); try std.testing.expectEqual(@as(c_int, 0), Shell.setenv("PARDES_CHILD_ENV_PROBE", "1", 1)); - defer _ = unsetenv("PARDES_CHILD_ENV_PROBE"); + defer _ = testUnsetenv("PARDES_CHILD_ENV_PROBE"); var slots: ChildEnv.Slots = undefined; // Where the user named no pager: a terminal's shell pages into a @@ -1481,6 +1534,16 @@ test "the child environment replaces the launcher's terminal identity exactly on extern "c" fn unsetenv(name: [*:0]const u8) c_int; +/// For tests: unset `name` without compacting the environment block std +/// captured at startup (std.testing.environ and Io's spawns read it later). A +/// new variable first moves glibc's `environ` to a heap copy, which unsetenv +/// then compacts instead. Runtime code never changes this process's +/// environment (ChildEnv.exportVar). +pub fn testUnsetenv(name: [*:0]const u8) c_int { + _ = Shell.setenv("PARDES_TEST_ENVIRON_COPIED", "1", 1); + return unsetenv(name); +} + /// The environment a pane's command is given (exportPaneEnv, `$winid`, /// ChildEnv), as `NAME=value` lines in `arena`, for a command that is not a /// pane's: an Edit's filters, run from pane `serial`. @@ -1497,10 +1560,7 @@ pub fn commandEnv(arena: std.mem.Allocator, serial: u32, adopts: bool) ![]const fn exportWinid(serial: u32) void { var buf: [16]u8 = undefined; - if (serial != 0) if (std.fmt.bufPrintSentinel(&buf, "{d}", .{serial}, 0)) |text| { - if (Shell.setenv("winid", text, 1) == 0) return; - } else |_| {}; - _ = unsetenv("winid"); + ChildEnv.exportVar("winid", if (serial == 0) null else std.fmt.bufPrint(&buf, "{d}", .{serial}) catch null); } /// The longest line a command pane runs (exec.zig refuses a longer one). @@ -1587,17 +1647,18 @@ pub fn forkShell( if (helper) |path| { const c = core.?; const pn = c.panes[pane].?; - const socket = libc.getenv("PARDES_9P") orelse return error.SessionSocketRequired; + const socket = ChildEnv.exported("PARDES_9P") orelse return error.SessionSocketRequired; var command: std.Io.Writer.Allocating = .init(c.gpa); defer command.deinit(); - try @import("linux/v9fs.zig").writeLaunchCommand(&command.writer, path, std.mem.span(socket), &spawn.argv); + try @import("linux/v9fs.zig").writeLaunchCommand(&command.writer, path, socket, &spawn.argv); if (!try pardes.panes.terminal.queuePendingCommand(pn, command.written())) return error.ShellAlreadyStarted; } // Built here and not after the fork: between fork and exec the child may // not call setenv, whose malloc can deadlock against a pty reader thread // that held the heap when the fork took its snapshot. var env_slots: ChildEnv.Slots = undefined; - const paging: ChildEnv.Paging = if (one_line != null) + const pages = if (core) |c| if (c.panes[pane]) |pn| pn.command_pages else false else false; + const paging: ChildEnv.Paging = if (one_line != null and !pages) .command else if (core) |c| (if (c.settings.pager == .off) .terminal_off else .terminal) else .terminal; const envp = ChildEnv.build(&env_slots, paging); @@ -1695,7 +1756,7 @@ test "a forked shell is told which terminal it is talking to" { defer if (restore) |t| { _ = Shell.setenv("TERM", t.ptr, 1); } else { - _ = unsetenv("TERM"); + _ = testUnsetenv("TERM"); }; _ = Shell.setenv("TERM", "dumb", 1); diff --git a/src/hunks.zig b/src/hunks.zig index 73c0654a..24aaf63d 100644 --- a/src/hunks.zig +++ b/src/hunks.zig @@ -368,21 +368,34 @@ const max_line_distance = 0.6; /// rather than aligned at a quadratic cost (delta aligns them all). const max_run = 256; const max_cells = 1 << 16; +/// Pairs one hunk compares (each a table of up to `max_cells`): past it, the +/// rest of its lines are shown unemphasised. And the scratch each pair is +/// aligned in, reused by the next: its table and the ranges it marks. +const max_pairs = 4096; +const pair_scratch_bytes = 8 << 20; /// Pair each removed line with the first added line after the last one /// paired that is near enough to it, in order, and emphasise in both the /// tokens one has and the other has not. fn inferEdits(arena: std.mem.Allocator, minus: []HunkLine, plus: []HunkLine) !void { if (minus.len == 0 or plus.len == 0 or minus.len > max_run or plus.len > max_run) return; + const plus_tokens = try arena.alloc([]const []const u8, plus.len); + for (plus, plus_tokens) |line, *tokens| tokens.* = try tokenize(arena, line.code); + var scratch: std.heap.FixedBufferAllocator = .init(try @import("memory.zig").unfilled(arena, pair_scratch_bytes)); + var pairs: usize = 0; var plus_index: usize = 0; for (minus) |*m| { const m_tokens = try tokenize(arena, m.code); var j = plus_index; while (j < plus.len) : (j += 1) { - const p_tokens = try tokenize(arena, plus[j].code); + const p_tokens = plus_tokens[j]; if ((m_tokens.len + 1) * (p_tokens.len + 1) > max_cells) continue; - var alignment = try Alignment.init(arena, m_tokens, p_tokens); - const marks = try annotate(arena, &alignment, m.code, plus[j].code); + if (pairs == max_pairs) return; + pairs += 1; + scratch.reset(); + const pair = scratch.allocator(); + var alignment = Alignment.init(pair, m_tokens, p_tokens) catch continue; + const marks = annotate(pair, &alignment, m.code, plus[j].code) catch continue; if (marks.distance > max_line_distance) continue; for (marks.minus) |r| for (m.styles[r[0]..r[1]]) |*b| { b.* |= syntax.diff_emph; diff --git a/src/image.zig b/src/image.zig index 8bf9de08..5797ce9e 100644 --- a/src/image.zig +++ b/src/image.zig @@ -768,7 +768,13 @@ pub fn readable(bytes: []const u8) bool { /// decode + downscale to RGBA, gpa-owned. Returns null on any failure — the /// pane then simply shows a blank body. pub fn decode(gpa: std.mem.Allocator, bytes: []const u8) ?struct { rgba: []u8, w: usize, h: usize } { - if (bytes.len == 0) return null; + if (bytes.len == 0 or bytes.len > std.math.maxInt(c_int)) return null; + // Bounded before decoding: a few KB of PNG can name gigapixels. + var iw: c_int = 0; + var ih: c_int = 0; + var ic: c_int = 0; + if (stbi_info_from_memory(bytes.ptr, @intCast(bytes.len), &iw, &ih, &ic) != 1) return null; + if (@as(usize, @intCast(@max(0, iw))) * @as(usize, @intCast(@max(0, ih))) > @import("memory.zig").limits.image_decode_pixels) return null; var img = zstbi.Image.loadFromMemory(bytes, 4) catch return null; // 4 = RGBA defer img.deinit(); var scaled: ?zstbi.Image = null; diff --git a/src/locations.zig b/src/locations.zig index 2828566d..1e4a20cf 100644 --- a/src/locations.zig +++ b/src/locations.zig @@ -217,6 +217,98 @@ pub fn treeContinuation(line: []const u8, buffer: []u8, max_cells: usize) []cons return out.buffered()[0..end]; } +const Resolved = struct { code: []const u8 = "", colors: []const u8 = &.{}, source_row: usize = 0 }; + +/// Each Callstack row's code (its call site's line, or a caller's header) and +/// colors, by input line. A page spans thousands of files, so they are taken +/// a file at a time: grouped by path, each file read, line-indexed and +/// analyzed once, its rows' code and colors copied out, and the file let go +/// before the next. Row by row, the cache compared the whole source each +/// time and past its 64 entries parsed it again, every lookup walked the +/// file from its top, and every source stayed in the arena to the end. +fn resolveCallstack(p: *pardes.Pardes, arena: std.mem.Allocator, dir: []const u8, input: []const u8) ![]Resolved { + const mark = @import("lsp/lsp.zig").callstack_match; + var all: std.ArrayList([]const u8) = .empty; + var by_path: std.StringArrayHashMapUnmanaged(std.ArrayList(u32)) = .empty; + var lines = std.mem.splitScalar(u8, std.mem.trimEnd(u8, input, "\n"), '\n'); + while (lines.next()) |raw| { + const line = if (std.mem.startsWith(u8, raw, mark)) raw[mark.len..] else raw; + const index: u32 = @intCast(all.items.len); + try all.append(arena, line); + const target = parseRaw(line[treePrefix(line).end..]); + if (target.at.line == 0) continue; + const entry = try by_path.getOrPut(arena, target.path); + if (!entry.found_existing) entry.value_ptr.* = .empty; + try entry.value_ptr.append(arena, index); + } + const resolved = try arena.alloc(Resolved, all.items.len); + @memset(resolved, .{}); + var file_arena: std.heap.ArenaAllocator = .init(p.gpa); + defer file_arena.deinit(); + for (by_path.keys(), by_path.values()) |path, indexes| { + _ = file_arena.reset(.retain_capacity); + const fa = file_arena.allocator(); + const source = sourceText(p, fa, dir, path); + var cached: Cache.Hit = .{ .analysis = .{} }; + defer cached.deinit(p.tree_sitter_gpa); + // Where each line starts, the file's length closing the last. + var starts: std.ArrayList(usize) = .empty; + if (source) |text| { + cached = p.locations_cache.get(p.tree_sitter_gpa, text.path, text.bytes, true, true) catch .{ .analysis = .{} }; + try starts.append(fa, 0); + var at: usize = 0; + while (std.mem.indexOfScalarPos(u8, text.bytes, at, '\n')) |nl| : (at = nl + 1) try starts.append(fa, nl + 1); + } + for (indexes.items) |index| { + const line = all.items[index]; + const prefix = treePrefix(line); + const target = parseRaw(line[prefix.end..]); + var code = line[prefix.end + target.code_start ..]; + var colors: []const u8 = &.{}; + var source_row = target.at.line - 1; + if (source) |text| { + const row = target.at.line - 1; + if (prefix.branch) { + // Choose the innermost header containing the server's name range. + var chosen: ?syntax.ContextDeclaration = null; + for (cached.analysis.declarations) |declaration| { + if (declaration.start_line <= row and row <= declaration.header_end_line) { + if (chosen == null or declaration.end_byte - declaration.start_byte < chosen.?.end_byte - chosen.?.start_byte) + chosen = declaration; + } + } + if (chosen) |header| { + code = text.bytes[header.start_byte..header.header_end_byte]; + source_row = header.start_line; + } else { + const fallback = indexedLine(text.bytes, starts.items, row); + if (fallback.len > 0) code = fallback; + } + } else if (code.len == 0) { + // A call site the language worker sent bare (it reads no + // sources for Callstack): its line, from the source here. + code = indexedLine(text.bytes, starts.items, row); + } + const offset = if (source_row < starts.items.len) starts.items[source_row] else text.bytes.len; + if (offset + code.len <= cached.analysis.colors.len and + std.mem.eql(u8, code, text.bytes[offset..][0..code.len])) + colors = cached.analysis.colors[offset..][0..code.len]; + code = try arena.dupe(u8, code); + colors = try arena.dupe(u8, colors); + } + resolved[index] = .{ .code = code, .colors = colors, .source_row = source_row }; + } + } + return resolved; +} + +/// modal.lineSlice through a line index (`starts`, as resolveCallstack builds it). +fn indexedLine(text: []const u8, starts: []const usize, row: usize) []const u8 { + if (row >= starts.len or starts[row] >= text.len) return ""; + const end = if (row + 1 < starts.len) starts[row + 1] - 1 else text.len; + return text[starts[row]..end]; +} + /// Enrich caller nodes on the UI thread, using the same source/declaration cache /// as LocationsContext. The LSP worker only supplies relationships and addresses. pub fn formatCallstack(p: *pardes.Pardes, dir: []const u8, input: []const u8) !Result { @@ -232,9 +324,10 @@ pub fn formatCallstack(p: *pardes.Pardes, dir: []const u8, input: []const u8) !R rows.deinit(p.gpa); } const stacked = p.settings.callstack_layout == .stacked; - var sources: std.StringHashMap(?Source) = .init(arena); + const resolved = try resolveCallstack(p, arena, dir, input); var lines = std.mem.splitScalar(u8, std.mem.trimEnd(u8, input, "\n"), '\n'); - while (lines.next()) |raw| { + var index: usize = 0; + while (lines.next()) |raw| : (index += 1) { const mark = @import("lsp/lsp.zig").callstack_match; const focus = std.mem.startsWith(u8, raw, mark); const line = if (focus) raw[mark.len..] else raw; @@ -245,41 +338,9 @@ pub fn formatCallstack(p: *pardes.Pardes, dir: []const u8, input: []const u8) !R try appendMetadata(p.gpa, &rows, .{ .path = "", .at = .{}, .label = "", .code = "", .original = null }, .context, 0, line.len, false); continue; } - const entry = try sources.getOrPut(target.path); - if (!entry.found_existing) entry.value_ptr.* = sourceText(p, arena, dir, target.path); - const source = entry.value_ptr.*; - var cached: Cache.Hit = .{ .analysis = .{} }; - defer cached.deinit(p.tree_sitter_gpa); - if (source) |text| cached = p.locations_cache.get(p.tree_sitter_gpa, text.path, text.bytes, true, true) catch .{ .analysis = .{} }; - var code = line[prefix.end + target.code_start ..]; - var colors: []const u8 = &.{}; - var source_row = target.at.line - 1; - if (source) |text| { - const row = target.at.line - 1; - if (prefix.branch) { - // Choose the innermost header containing the server's name range. - var chosen: ?syntax.ContextDeclaration = null; - for (cached.analysis.declarations) |declaration| { - if (declaration.start_line <= row and row <= declaration.header_end_line) { - if (chosen == null or declaration.end_byte - declaration.start_byte < chosen.?.end_byte - chosen.?.start_byte) - chosen = declaration; - } - } - if (chosen) |header| { - code = text.bytes[header.start_byte..header.header_end_byte]; - source_row = header.start_line; - } else { - const fallback = @import("modal.zig").lineSlice(text.bytes, row); - if (fallback.len > 0) code = fallback; - } - } - { - const offset = @import("modal.zig").lineStartOffset(text.bytes, source_row); - if (offset + code.len <= cached.analysis.colors.len and - std.mem.eql(u8, code, text.bytes[offset..][0..code.len])) - colors = cached.analysis.colors[offset..][0..code.len]; - } - } + const code = resolved[index].code; + const colors = resolved[index].colors; + var source_row = resolved[index].source_row; const recursion = prefix.branch and std.mem.endsWith(u8, line, " ↻ recursion"); const first_line = code[0 .. std.mem.indexOfScalar(u8, code, '\n') orelse code.len]; const source_indent = if (stacked) first_line.len - std.mem.trimStart(u8, first_line, " \t").len else 0; diff --git a/src/look.zig b/src/look.zig index f4fbe52e..02754fff 100644 --- a/src/look.zig +++ b/src/look.zig @@ -587,6 +587,10 @@ fn lookFirstHit(p: *Pardes, id: usize) void { /// Fills the results buffer, and answers how many rows a text or PDF /// search found (0 for Find and Grep, which do not count). With /// `open_empty` false a text search that found nothing opens no buffer. +/// A search of one pane's text (or its PDF) writes at most this many bytes +/// of rows: past it, the rows stop and say where. +pub const search_text_output_bytes = 16 << 20; + pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start: SearchStart, open_empty: bool) !usize { const pane = p.panes[id] orelse return 0; // Blanks at either end are the pattern's (a look of an indented line); @@ -598,7 +602,9 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start // root, and the directory the results buffer is named in. From a // column's or the workspace's tag, the session's. const dir = if (p.exec_header) p.sessionDir() else Pardes.paneDir(pane); - const out = try arena.alloc(u8, filesystem.search_max_output_bytes); + // As much as the search can write and no more: a root's hits for each + // root Grep walks, one for Find, `search_text_output_bytes` in a pane. + var out: []u8 = &.{}; var out_len: usize = 0; var nrows: usize = 0; var anchor: ?usize = null; @@ -631,6 +637,7 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start roots[nroots] = d; nroots += 1; } + out = try @import("memory.zig").unfilled(arena, @max(1, nroots) * filesystem.search_root_output_bytes); // A pane's directory that is not there (a scratch named into one // not made yet, a peer's) has nothing to grep: passed over, not the // search's failure. @@ -641,9 +648,11 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start }; nrows = std.mem.count(u8, out[0..out_len], "\n"); } else if (kind == .find) { + out = try @import("memory.zig").unfilled(arena, filesystem.search_root_output_bytes); out_len = try filesystem.find(arena, dir, pat, out, ¬es); nrows = std.mem.count(u8, out[0..out_len], "\n"); } else if (pane.hasPdf()) { + out = try @import("memory.zig").unfilled(arena, search_text_output_bytes); if (comptime pdf_enabled) { const found = try panes.Pdf.searchRows( &pane.pdf.?, @@ -658,6 +667,7 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start anchor = found.anchor; } } else { + out = try @import("memory.zig").unfilled(arena, search_text_output_bytes); const lines = try edit.paneCursorLines(p, &pane.body); // a real file names itself; a terminal or an output buffer has no path const has_path = if (pane.file) |f| f.output == null else false; @@ -675,7 +685,10 @@ pub fn runSearch(p: *Pardes, id: usize, pat_raw: []const u8, kind: Search, start const row = try std.fmt.allocPrint(arena, "{s}:{d}:{d}{c}{d} {s}\n", .{ loc, i + 1, at + 1, config.range_sep, at + pat.len, std.mem.trimEnd(u8, ln, " \t"), }); - if (row.len > out.len - out_len) break; + if (row.len > out.len - out_len) { + notes.rows_cut = nrows; + break; + } if (start == .cursor and (i < cl or (i == cl and at <= cc))) anchor = nrows; @memcpy(out[out_len..][0..row.len], row); out_len += row.len; @@ -1451,6 +1464,7 @@ fn lookUnresolved(p: *Pardes, id: usize, pane: *Pane, trimmed: []const u8, text: // 9P is no keyboard input, and searches whatever the key mode. if (pane.image != null or (pane.body.mode == .tty and !p.fs.serving)) return; if (text.len == 0) return; + if (lookCommit(p, id, pane, trimmed)) return; if (p.settings.look_word == .search and pane.file != null) return lookWord(p, id, pane, text); p.word_look = true; defer p.word_look = false; @@ -1471,6 +1485,49 @@ fn lookUnresolved(p: *Pardes, id: usize, pane: *Pane, trimmed: []const u8, text: } } +/// A commit hash (7 to 40 hex digits, a letter and a digit among them) +/// looked at inside a git repository: `git show` it, paged into a +Pager, +/// which stacks its hunks in a +Diff (ctl.fillPager); the command pane +/// goes when it succeeds (exec.commandDone). +// ponytail: a hex word that names no commit leaves git's error in a command pane, not a search; check the object first if that bites. +fn lookCommit(p: *Pardes, id: usize, pane: *Pane, word: []const u8) bool { + if (comptime !pardes.hosted) return false; + if (!isCommitHash(word)) return false; + var dir = filesystem.localPath(Pardes.paneDir(pane)) orelse return false; + while (true) { + var buf: [4096]u8 = undefined; + const dot_git = std.fmt.bufPrint(&buf, "{s}/.git", .{dir}) catch return false; + if (exec.kindOf(dot_git) != null) break; + dir = std.fs.path.dirname(dir) orelse return false; + } + var line: [64]u8 = undefined; + const dst = exec.execute(p, id, std.fmt.bufPrint(&line, "git show {s}", .{word}) catch return false) orelse return true; + if (p.panes[dst]) |c| c.command_pages = true; + return true; +} + +fn isCommitHash(word: []const u8) bool { + if (word.len < 7 or word.len > 40) return false; + var digit = false; + var letter = false; + for (word) |c| switch (c) { + '0'...'9' => digit = true, + 'a'...'f' => letter = true, + else => return false, + }; + return digit and letter; +} + +test "a commit hash is 7 to 40 lowercase hex digits with a letter and a digit" { + try std.testing.expect(isCommitHash("9bffa7041c")); + try std.testing.expect(isCommitHash("9bffa7041cd2e8a0b6b1f7c3e4d5a6b7c8d9e0f1")); + try std.testing.expect(!isCommitHash("9bffa7")); + try std.testing.expect(!isCommitHash("defaced")); + try std.testing.expect(!isCommitHash("1234567")); + try std.testing.expect(!isCommitHash("0x9bffa7041c")); + try std.testing.expect(!isCommitHash("9BFFA7041C")); +} + /// Opens (or finds) `path` as Look would, `id`'s own pane when it is empty, /// and selects what `expr` addresses there from its dot. An address that /// does not evaluate moves nothing and says why (acme: no jump). diff --git a/src/lsp/lsp.zig b/src/lsp/lsp.zig index 4a70efe5..1c1b57a2 100644 --- a/src/lsp/lsp.zig +++ b/src/lsp/lsp.zig @@ -26,6 +26,10 @@ pub const Kind = enum { }; pub const callstack_match = "@callstack-match "; +/// The last row of a Callstack page that stopped full: ` `, +/// the entry indices from the top the next page starts at (`3.0.17`), and +/// how many callers this page holds. +pub const callstack_more = "@callstack-more "; pub const CallstackCorners = enum { rounded, square }; pub const callstack_default_depth: u8 = 8; pub const callstack_max_depth: u8 = 64; @@ -46,6 +50,9 @@ pub const Req = struct { /// For `explain`: the query it narrates, read from `arg` (` /// `) by `explaining`; a definition when none was asked before. explains: Kind = .definition, + /// The fixed region `arena` allocates from, when it is one: a walk that + /// can grow (Callstack) reads how much is left and stops in time. + budget: ?*const std.heap.FixedBufferAllocator = null, }; pub const Doc = struct { path: []const u8, source: [:0]const u8 }; @@ -102,18 +109,18 @@ pub fn speaks(path: []const u8) bool { return false; } -pub fn query(gpa: std.mem.Allocator, arena: std.mem.Allocator, req: Req, out: *std.Io.Writer) !void { +pub fn query(io: std.Io, gpa: std.mem.Allocator, arena: std.mem.Allocator, req: Req, out: *std.Io.Writer) !void { if (req.kind == .status) { try out.print("backend: {s}\n", .{backend_name}); - inline for (backends) |b| try b.query(gpa, arena, req, out); + inline for (backends) |b| try b.query(io, gpa, arena, req, out); return; } inline for (backends) |b| { if (b.speaks(req.path) and b.supports.contains(req.kind)) - return b.query(gpa, arena, req, out); + return b.query(io, gpa, arena, req, out); } if (req.kind == .explain and backends.len > 0) - return backends[0].query(gpa, arena, req, out); + return backends[0].query(io, gpa, arena, req, out); // Nothing asked: said why, where an empty answer would read as nothing // found. inline for (backends) |b| if (b.speaks(req.path)) @@ -217,6 +224,6 @@ test "a question no backend answers for a file says why rather than find nothing var buf: [256]u8 = undefined; var out: std.Io.Writer = .fixed(&buf); // A file no server speaks for is answered, not left empty. - try query(std.testing.allocator, arena.allocator(), .{ .kind = .incoming_calls, .path = "/x/notes.txt", .source = "", .offset = 0 }, &out); + try query(std.testing.io, std.testing.allocator, arena.allocator(), .{ .kind = .incoming_calls, .path = "/x/notes.txt", .source = "", .offset = 0 }, &out); try std.testing.expectStringStartsWith(out.buffered(), "@none "); } diff --git a/src/lsp/lsp_client.zig b/src/lsp/lsp_client.zig index b396879b..e356c979 100644 --- a/src/lsp/lsp_client.zig +++ b/src/lsp/lsp_client.zig @@ -8,9 +8,10 @@ //! worker thread, and a language server costs tens of milliseconds to start //! and MINUTES to index a large workspace. So a server is spawned lazily on //! the first query that needs it, kept for the life of the editor, and each -//! spec gets at most one — `conns[i]` guards itself with a pthread mutex -//! because the gui shell detaches its workers and a superseded query can still -//! be inside `run` when the next arrives. +//! spec gets at most one — `conns[i]` guards itself with an `Io.Mutex` +//! because a canceled query can still be inside `run` when the next arrives. +//! Every wait is an `Io` cancelation point: a shell cancels the worker of a +//! query it no longer wants, and the worker stops there. //! //! ONE READER THREAD PER SERVER, and it is not optional. The first design //! pumped the socket only while a query waited, which works for a server that @@ -28,7 +29,8 @@ //! under a thread still polling it. //! //! Nothing here may wedge the editor: -//! - every write and every mailbox wait is bounded by a deadline, +//! - every write is bounded by a deadline, and every mailbox wait by a +//! deadline or, while the server narrates progress, by cancelation, //! - a query the server does not answer in time returns no rows and sends //! `$/cancelRequest`; the server is NOT killed for being busy (an indexing //! server is busy for minutes and the status row says so) — but three @@ -58,20 +60,10 @@ extern "c" fn access(path: [*:0]const u8, mode: c_int) c_int; extern "c" fn getenv(name: [*:0]const u8) ?[*:0]const u8; extern "c" fn usleep(usec: c_uint) c_int; extern "c" fn atexit(f: *const fn () callconv(.c) void) c_int; -extern "c" fn realpath(path: [*:0]const u8, resolved: [*]u8) ?[*:0]u8; // test-only libc (the seam's own tests build a real directory tree) extern "c" fn mkdtemp(template: [*:0]u8) ?[*:0]u8; extern "c" fn mkdir(path: [*:0]const u8, mode: libc.mode_t) c_int; extern "c" fn system(cmd: [*:0]const u8) c_int; -// pthread condattr surface, absent from std.c: what makes the mailbox waits -// tick on CLOCK_MONOTONIC (linux) or a relative timeout (darwin). -const pthread_condattr_t = extern struct { data: [8]u8 align(@alignOf(usize)) = @splat(0) }; -extern "c" fn pthread_condattr_init(attr: *pthread_condattr_t) c_int; -extern "c" fn pthread_condattr_setclock(attr: *pthread_condattr_t, clock: c_int) c_int; -extern "c" fn pthread_condattr_destroy(attr: *pthread_condattr_t) c_int; -extern "c" fn pthread_cond_init(cond: *libc.pthread_cond_t, attr: ?*const pthread_condattr_t) c_int; -extern "c" fn pthread_cond_timedwait_relative_np(cond: *libc.pthread_cond_t, mutex: *libc.pthread_mutex_t, reltime: *const libc.timespec) c_int; -const CLOCK_MONOTONIC: c_int = 1; // linux ABI; the setclock call is linux-only const X_OK: c_int = 1; const POLLIN: i16 = 0x001; @@ -182,21 +174,35 @@ fn specFor(path: []const u8) ?usize { return null; } -// Deadlines. Requests are bounded because tty.zig cancels-and-joins the -// previous worker on a new keypress, so the worst UI stall a wedged wait can -// cause is one req_ms. Timeouts do NOT kill the server — an indexing -// rust-analyzer legitimately sits on a `gd` for longer than anyone will wait, -// and the message row is already narrating why. +// Deadlines. A shell cancels-and-joins the previous worker on a new query, +// and its waits are cancelation points: so a request needs no deadline to +// keep the editor responsive, and waits for its answer. Past req_ms the +// message row says the server is still at it; only a server silent and idle +// (no $/progress) for wedge_ms has failed to answer. Timeouts do NOT kill the +// server — an indexing rust-analyzer legitimately sits on a `gd` for minutes. +// The handshake runs protected from cancelation, so it keeps its deadline. const init_ms = 8_000; // spawn + initialize handshake -const req_ms = 4_000; // one request/response round trip +const req_ms = 4_000; // a request past this is said to be still working +const wedge_ms = 300_000; // silent and idle this long: not answering (rust-analyzer takes ~1 min on `default`'s callers in agave) const diag_ms = 1_200; // wait for a publishDiagnostics push after didChange const reply_ms = 2_000; // our answers to server-to-client requests +const reply_queue_bytes = 64 << 10; // replies waiting for the socket (Conn.replies) +const parse_kept = 1 << 20; // what the reader's parse arena keeps between messages +const max_diag_files = 256; // files a server's pushed diagnostics are kept for +const max_diag_bytes = 32 << 20; // and their bytes const wedged_strikes = 3; // consecutive timeouts before a restart -const max_rows = 2000; +/// Rows one answer may hold, and its writer's bytes: carved from the query's +/// fixed region (memory.limits.arena.lsp_query) with the rest of its memory. +const max_rows = 100_000; +/// What a Callstack walk leaves of its query's region for rendering the +/// tree it found: past this it stops and says so. +const callstack_reserve = 32 << 20; +/// What rendering leaves of the answer's writer for its closing line. +const row_room = 4096; const max_doc_bytes = 8 << 20; -const Err = error{ Dead, Timeout, Protocol, OutOfMemory, NoServer, Backoff, WriteFailed }; +const Err = error{ Dead, Timeout, Protocol, OutOfMemory, NoServer, Backoff, WriteFailed, Canceled }; /// Long-lived state outlives every query arena and cannot borrow the caller's /// gpa (a different one shows up in the harness than in the shell), so @@ -251,8 +257,22 @@ const State = enum(u8) { }; const Conn = struct { - mu: libc.pthread_mutex_t = .{}, - cond: libc.pthread_cond_t = .{}, + mu: std.Io.Mutex = .init, + cond: std.Io.Condition = .init, + /// Frames go out whole: a query's requests and the reader's replies + /// share the socket, and interleaved halves are a parse error that ends + /// the server. + write_mu: std.Io.Mutex = .init, + /// Replies the reader made while a query's write held the socket, sent + /// by that writer once its own message is out (releaseWrite): the reader + /// never waits to write, since a server blocked on its output reads + /// nothing more of ours. + /// Two halves: the reader appends to `replies[filling]`, and the writer + /// flips them and sends the other, with no copy and no wait. + replies: [2][reply_queue_bytes]u8 = undefined, + replies_len: [2]usize = .{ 0, 0 }, + filling: u1 = 0, + replies_mu: std.Io.Mutex = .init, state: State = .off, /// bumped per spawn. A reader thread that sees a different gen than its /// own is reading a corpse and exits; a waiter that sees one stops waiting. @@ -264,9 +284,6 @@ const Conn = struct { /// monotonic ms when the last handshake completed — a crash shortly /// after "ready" counts as a handshake failure for backoff purposes ready_at_ms: i64 = 0, - /// linux: the zeroed condvar defaults to REALTIME; re-initialized with a - /// monotonic condattr before its first wait (under the mutex) - cond_monotonic: bool = false, timeouts: u8 = 0, pid: libc.pid_t = -1, sock: c_int = -1, @@ -278,6 +295,8 @@ const Conn = struct { extra_roots: std.ArrayList([]u8) = .empty, docs: std.ArrayList(Doc) = .empty, diags: std.ArrayList(DiagSet) = .empty, + /// The bytes `diags` holds, kept under max_diag_bytes. + diag_bytes: usize = 0, /// the response mailbox: one request outstanding per connection want_id: u32 = 0, resp: ?[]u8 = null, @@ -285,13 +304,12 @@ const Conn = struct { progress: i32 = 0, title: [48]u8 = @splat(0), title_len: u8 = 0, + /// The server says it is still loading its workspace (rust-analyzer's + /// serverStatus, not quiescent): an empty answer now is not one. + loading: bool = false, + /// its index in `conns`, for what it posts + si: usize = 0, - fn lock(c: *Conn) void { - _ = libc.pthread_mutex_lock(&c.mu); - } - fn unlock(c: *Conn) void { - _ = libc.pthread_mutex_unlock(&c.mu); - } fn alive(c: *const Conn) bool { return c.state == .starting or c.state == .ready; } @@ -406,10 +424,10 @@ const Trace = struct { // ------------------------------------------------------------------- query -pub fn query(gpa: std.mem.Allocator, arena: std.mem.Allocator, asked: lsp.Req, out: *std.Io.Writer) !void { +pub fn query(io: std.Io, gpa: std.mem.Allocator, arena: std.mem.Allocator, asked: lsp.Req, out: *std.Io.Writer) !void { _ = gpa; const req = lsp.explaining(asked); - if (req.kind == .status) return status(req, out); + if (req.kind == .status) return status(io, req, out); var tr: Trace = .{ .on = req.kind == .explain }; const si = specFor(req.path) orelse { @@ -421,13 +439,14 @@ pub fn query(gpa: std.mem.Allocator, arena: std.mem.Allocator, asked: lsp.Req, o }; tr.note("file {s} -> {s} (languageId {s})", .{ std.fs.path.basename(req.path), specs[si].name, specs[si].lang }); - const scratch_buf = try arena.alloc(u8, max_rows * 512); + // Raw: `alloc` fills a safe build's new memory, all 51 MB of it, per query. + const scratch_buf = (arena.rawAlloc(max_rows * 512, .of(u8), @returnAddress()) orelse return error.OutOfMemory)[0 .. max_rows * 512]; var scratch: std.Io.Writer = .fixed(scratch_buf); const t0 = nowUs(); var err_name: []const u8 = ""; var failure: ?anyerror = null; - answer(arena, si, req, &scratch, &tr) catch |e| { + answer(io, arena, si, req, &scratch, &tr) catch |e| { failure = e; err_name = @errorName(e); tr.note("ERROR: {s}", .{err_name}); @@ -443,9 +462,9 @@ pub fn query(gpa: std.mem.Allocator, arena: std.mem.Allocator, asked: lsp.Req, o // again, not that it is not there. if (failure) |err| if (err == error.Backoff) { const c = &conns[si]; - c.lock(); + c.mu.lockUncancelable(io); const left_ms = c.retry_after_ms -| nowMs(); - c.unlock(); + c.mu.unlock(io); return lsp.none(out, "{s} failed recently; retry in {d}s", .{ specs[si].name, @divFloor(left_ms + 999, 1000) }); }; if (failure) |err| return err; @@ -464,17 +483,34 @@ fn traceOut(tr: *const Trace, req: lsp.Req, out: *std.Io.Writer, rows: usize, us try out.print("\n{d} row(s) in {d}us\n", .{ rows, us }); } -fn answer(arena: std.mem.Allocator, si: usize, req: lsp.Req, out: *std.Io.Writer, tr: *Trace) Err!void { +fn answer(io: std.Io, arena: std.mem.Allocator, si: usize, req: lsp.Req, out: *std.Io.Writer, tr: *Trace) Err!void { const c = &conns[si]; - c.lock(); - defer c.unlock(); - const g = try ensure(c, si, arena, req, tr); - run(c, si, arena, req, out, tr) catch |e| { + try c.mu.lock(io); + defer c.mu.unlock(io); + const g = try ensure(c, io, si, arena, req, tr); + const before = out.end; + try attempt(c, io, si, arena, req, out, tr, g); + // A server still loading its workspace answers at once, with nothing + // (rust-analyzer before its index): that is no answer. Wait until it says + // it is loaded, the message row saying so, and ask again. + const answered = out.buffered()[before..]; + if (c.loading and c.gen == g and (answered.len == 0 or std.mem.startsWith(u8, answered, "@none "))) { + tr.note("{s} is still loading: waiting for it, then asking again", .{specs[si].name}); + post(si, .always, "{s} is still loading the workspace; the answer comes when it is done\u{2026}", .{specs[si].name}); + while (c.loading and c.gen == g and c.alive()) _ = try timedWait(c, io, nowMs() + 500); + out.end = before; + try attempt(c, io, si, arena, req, out, tr, g); + } +} + +/// One run of a query, and what its failure says of the server. +fn attempt(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io.Writer, tr: *Trace, g: u32) Err!void { + run(c, io, si, arena, req, out, tr) catch |e| { switch (e) { // transport-level: this server is gone; forget it, and the next // query respawns. Gen-checked so a respawn that happened while we // waited is not the one we kill. - error.Dead, error.Protocol => shutdownIf(c, g), + error.Dead, error.Protocol => shutdownIf(c, io, g), error.Timeout => { // Slow is only wedged when the server is IDLE. One with // active `$/progress` work (rust-analyzer running cargo @@ -487,18 +523,19 @@ fn answer(arena: std.mem.Allocator, si: usize, req: lsp.Req, out: *std.Io.Writer c.timeouts +|= 1; if (c.timeouts >= wedged_strikes and c.gen == g) { post(si, .always, "{s} not answering — restarting", .{specs[si].name}); - shutdownIf(c, g); + shutdownIf(c, io, g); } } }, - error.OutOfMemory, error.NoServer, error.Backoff, error.WriteFailed => {}, + // Canceled: the shell asked something newer, not a wedged server. + error.OutOfMemory, error.NoServer, error.Backoff, error.WriteFailed, error.Canceled => {}, } return e; }; c.timeouts = 0; } -fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io.Writer, tr: *Trace) Err!void { +fn run(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io.Writer, tr: *Trace) Err!void { // `explain` narrates the last query asked (a definition, `gd`, when // none was) — dispatch below on the effective kind so the trace // follows the code that query really runs. @@ -510,18 +547,18 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io var uri: std.ArrayList(u8) = .empty; try uriOf(&uri, arena, req.path); - try syncDoc(c, si, arena, uri.items, req.source); + try syncDoc(c, io, si, arena, uri.items, req.source); // Every other open file this server answers for, before a query whose // answer reaches them: zls renames only in the documents it has. if (lsp.reachesOtherFiles(kind)) for (req.others) |doc| if (specFor(doc.path) == si and !std.mem.eql(u8, doc.path, req.path)) { var other: std.ArrayList(u8) = .empty; try uriOf(&other, arena, doc.path); - try syncDoc(c, si, arena, other.items, doc.source); + try syncDoc(c, io, si, arena, other.items, doc.source); tr.note("synced {s} first", .{std.fs.path.basename(doc.path)}); }; const pos = posOf(req.source, req.offset, c.caps.enc); - var cx: Cx = .{ .arena = arena, .base = req.root, .cur_path = req.path, .cur_src = req.source, .out = out, .relative_parents = req.kind == .callstack, .callstack_corners = req.callstack_corners }; + var cx: Cx = .{ .arena = arena, .base = req.root, .cur_path = req.path, .cur_src = req.source, .out = out, .fixed_out = true, .relative_parents = req.kind == .callstack, .callstack_corners = req.callstack_corners }; for (req.others) |doc| try cx.files.append(arena, .{ .path = doc.path, .text = doc.source }); tr.note("server {s} (pid {d}), root {s}, {s} columns", .{ @tagName(c.state), c.pid, c.root, @tagName(c.caps.enc), @@ -540,7 +577,7 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io var b = try atPos(arena, uri.items, pos); if (kind == .references) try app(&b, arena, ",\"context\":{\"includeDeclaration\":true}"); tr.note("-> {s} @ {d}:{d}", .{ method, pos.line + 1, pos.ch + 1 }); - const result = (try call(c, arena, method, b.items, deadline)) orelse { + const result = (try call(c, io, arena, method, b.items, deadline)) orelse { tr.note("<- empty (no result, an error response, or null)", .{}); return; }; @@ -549,12 +586,12 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io }, .select_refs => { const b = try atPos(arena, uri.items, pos); - const result = (try call(c, arena, "textDocument/documentHighlight", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/documentHighlight", b.items, deadline)) orelse return; for (items(result)) |h| try emitRange(&cx, c.caps.enc, uri.items, get(h, "range"), ""); }, .hover => { const b = try atPos(arena, uri.items, pos); - const result = (try call(c, arena, "textDocument/hover", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/hover", b.items, deadline)) orelse return; const contents = get(result, "contents") orelse return; const text = str(contents) orelse str(get(contents, "value")) orelse blk: { for (items(contents)) |m| if (str(m) orelse str(get(m, "value"))) |t| break :blk t; @@ -569,26 +606,26 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io try app(&b, arena, "\"textDocument\":{\"uri\":"); try jstr(&b, arena, uri.items); try app(&b, arena, "}"); - const result = (try call(c, arena, "textDocument/documentSymbol", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/documentSymbol", b.items, deadline)) orelse return; try walkSymbols(&cx, c.caps.enc, uri.items, result, 0); }, .workspace_symbols => { var b: std.ArrayList(u8) = .empty; try app(&b, arena, "\"query\":"); try jstr(&b, arena, req.arg); - const result = (try call(c, arena, "workspace/symbol", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "workspace/symbol", b.items, deadline)) orelse return; for (items(result)) |sym| { const loc = get(sym, "location") orelse continue; try emitRange(&cx, c.caps.enc, str(get(loc, "uri")) orelse continue, get(loc, "range"), str(get(sym, "name")) orelse ""); } }, - .diagnostics => try diagnostics(c, arena, &cx, uri.items, deadline), - .workspace_diagnostics => try workspaceDiagnostics(c, arena, &cx, deadline), + .diagnostics => try diagnostics(c, io, arena, &cx, uri.items, deadline), + .workspace_diagnostics => try workspaceDiagnostics(c, io, arena, &cx, deadline), .rename => { var b = try atPos(arena, uri.items, pos); try app(&b, arena, ",\"newName\":"); try jstr(&b, arena, req.arg); - const result = (try call(c, arena, "textDocument/rename", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/rename", b.items, deadline)) orelse return; // Other open files that say the old name, should the server // answer for this file alone (zls renaming at a declaration does, // whatever it was told): applied here, they would no longer build. @@ -603,7 +640,7 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io try app(&b, arena, "\"textDocument\":{\"uri\":"); try jstr(&b, arena, uri.items); try app(&b, arena, "},\"options\":{\"tabSize\":4,\"insertSpaces\":true}"); - const result = (try call(c, arena, "textDocument/formatting", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/formatting", b.items, deadline)) orelse return; try formatEdits(&cx, c.caps.enc, result, req.source); }, .code_action => { @@ -611,7 +648,7 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io try app(&b, arena, "\"textDocument\":{\"uri\":"); try jstr(&b, arena, uri.items); try b.print(arena, "}},\"range\":{{\"start\":{{\"line\":{d},\"character\":{d}}},\"end\":{{\"line\":{d},\"character\":{d}}}}},\"context\":{{\"diagnostics\":[]}}", .{ pos.line, pos.ch, pos.line, pos.ch }); - const result = (try call(c, arena, "textDocument/codeAction", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/codeAction", b.items, deadline)) orelse return; for (items(result)) |ca| { const title = str(get(ca, "title")) orelse continue; try lsp.row(cx.out, lsp.rel(cx.base, cx.cur_path), pos.line, 0, flat(arena, title)); @@ -621,7 +658,7 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io .completion => { var b = try atPos(arena, uri.items, pos); try app(&b, arena, ",\"context\":{\"triggerKind\":1}"); - const result = (try call(c, arena, "textDocument/completion", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/completion", b.items, deadline)) orelse return; const list = if (get(result, "items")) |it| items(it) else items(result); const here = lsp.rel(cx.base, cx.cur_path); var n: usize = 0; @@ -643,11 +680,11 @@ fn run(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, out: *std.Io n += 1; } }, - .callstack => try callstack(c, si, arena, &cx, uri.items, pos, deadline, tr, req.callstack_depth), - .incoming_calls => try hierarchy(c, si, arena, &cx, uri.items, pos, deadline, .incoming, tr), - .outgoing_calls => try hierarchy(c, si, arena, &cx, uri.items, pos, deadline, .outgoing, tr), - .supertypes => try hierarchy(c, si, arena, &cx, uri.items, pos, deadline, .supers, tr), - .subtypes => try hierarchy(c, si, arena, &cx, uri.items, pos, deadline, .subs, tr), + .callstack => try callstack(c, io, si, arena, &cx, uri.items, pos, deadline, tr, req.callstack_depth, req.budget, req.arg), + .incoming_calls => try hierarchy(c, io, si, arena, &cx, uri.items, pos, deadline, .incoming, tr), + .outgoing_calls => try hierarchy(c, io, si, arena, &cx, uri.items, pos, deadline, .outgoing, tr), + .supertypes => try hierarchy(c, io, si, arena, &cx, uri.items, pos, deadline, .supers, tr), + .subtypes => try hierarchy(c, io, si, arena, &cx, uri.items, pos, deadline, .subs, tr), .status, .explain => unreachable, } } @@ -687,13 +724,13 @@ fn walkSymbols(cx: *Cx, enc: Enc, uri: []const u8, node: std.json.Value, depth: /// Pull when the server does (LSP 3.17), the push store otherwise. The store /// path waits briefly for a publish that postdates the didChange we just /// sent, so `]d` right after an edit sees the new truth, not the old one. -fn diagnostics(c: *Conn, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, deadline: i64) Err!void { +fn diagnostics(c: *Conn, io: std.Io, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, deadline: i64) Err!void { if (c.caps.pull) { var b: std.ArrayList(u8) = .empty; try app(&b, arena, "\"textDocument\":{\"uri\":"); try jstr(&b, arena, uri); try app(&b, arena, "}"); - const result = (try call(c, arena, "textDocument/diagnostic", b.items, deadline)) orelse return; + const result = (try call(c, io, arena, "textDocument/diagnostic", b.items, deadline)) orelse return; for (items(get(result, "items"))) |dg| try emitDiag(cx, c.caps.enc, uri, dg, arena); return; } @@ -702,13 +739,13 @@ fn diagnostics(c: *Conn, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, dea stale = d.stale; break; }; - if (stale) waitFresh(c, uri, nowMs() + diag_ms); + if (stale) try waitFresh(c, io, uri, nowMs() + diag_ms); try renderStore(c, arena, cx, uri); } -fn workspaceDiagnostics(c: *Conn, arena: std.mem.Allocator, cx: *Cx, deadline: i64) Err!void { +fn workspaceDiagnostics(c: *Conn, io: std.Io, arena: std.mem.Allocator, cx: *Cx, deadline: i64) Err!void { if (c.caps.pull_workspace) { - const result = (try call(c, arena, "workspace/diagnostic", "\"previousResultIds\":[]", deadline)) orelse return; + const result = (try call(c, io, arena, "workspace/diagnostic", "\"previousResultIds\":[]", deadline)) orelse return; for (items(get(result, "items"))) |per| { const uri = str(get(per, "uri")) orelse continue; for (items(get(per, "items"))) |dg| try emitDiag(cx, c.caps.enc, uri, dg, arena); @@ -741,7 +778,7 @@ fn renderStore(c: *Conn, arena: std.mem.Allocator, cx: *Cx, only_uri: ?[]const u /// Block (mutex released) until the reader marks `uri` fresh or the deadline /// passes. The reader broadcasts on every publishDiagnostics. -fn waitFresh(c: *Conn, uri: []const u8, deadline: i64) void { +fn waitFresh(c: *Conn, io: std.Io, uri: []const u8, deadline: i64) error{Canceled}!void { const g = c.gen; while (c.gen == g and c.alive()) { var fresh = false; @@ -750,7 +787,7 @@ fn waitFresh(c: *Conn, uri: []const u8, deadline: i64) void { break; }; if (fresh) return; - if (!timedWait(c, deadline)) return; + if (!try timedWait(c, io, deadline)) return; } } @@ -873,18 +910,21 @@ fn sortEdits(edits: []PutEdit) void { /// A bounded static hierarchy. Reuse the prepared item verbatim (including /// server data), and retain every call-site range rather than just a declaration. -fn hierarchyResult(c: *Conn, arena: std.mem.Allocator, item: std.json.Value, method: []const u8, deadline: i64) Err!?std.json.Value { +fn hierarchyResult(c: *Conn, io: std.Io, arena: std.mem.Allocator, item: std.json.Value, method: []const u8, deadline: i64) Err!?std.json.Value { if (jsonDepth(item, 0) > 96) return null; var body: std.ArrayList(u8) = .empty; try app(&body, arena, "\"item\":"); try app(&body, arena, std.json.Stringify.valueAlloc(arena, item, .{}) catch return error.OutOfMemory); - return call(c, arena, method, body.items, deadline); + return call(c, io, arena, method, body.items, deadline); } const CallTree = struct { - const Site = struct { range: std.json.Value, focus: bool }; + /// What a node keeps of the server's item: its own copies, so the + /// response it came in is freed once its subtree is walked. + const Call = struct { uri: []const u8, name: []const u8, at: Range }; + const Site = struct { range: Range, focus: bool }; const Node = struct { - item: std.json.Value, + call: Call, ranges: std.ArrayList(Site) = .empty, children: std.ArrayList(usize) = .empty, cycle: bool = false, @@ -893,10 +933,39 @@ const CallTree = struct { roots: std.ArrayList(usize) = .empty, path: [lsp.callstack_max_depth]std.json.Value = undefined, depth_limit: usize = lsp.callstack_default_depth, - visited: usize = 0, limited: bool = false, + out_of_time: bool = false, + out_of_memory: bool = false, + budget: ?*const std.heap.FixedBufferAllocator = null, + si: usize = 0, + collected: usize = 0, + started_ms: i64 = 0, + narrated: bool = false, siblings: [lsp.callstack_max_depth]bool = @splat(false), rendered: usize = 0, + /// Paging. A page starts at `start`, a path of entry indices from the + /// top (empty: the start), skipping what is before it at each level; + /// `rows` and `bytes` count what rendering takes, and the walk stops + /// before their caps at `cut`, which the next page starts at (`more`). + /// The ancestors on the path are shown again on it, as context. + start: [lsp.callstack_max_depth]u32 = undefined, + start_len: usize = 0, + at: [lsp.callstack_max_depth]u32 = undefined, + cut: [lsp.callstack_max_depth]u32 = undefined, + cut_len: usize = 0, + more: bool = false, + rows: usize = 0, + row_cap: usize = max_rows - 16, + /// The bytes those rows take, over-estimated (the uri for the path it + /// becomes), kept under the answer writer's room. + bytes: usize = 0, + byte_cap: usize = std.math.maxInt(usize), + + /// A row of a caller at `depth`: its tree columns, the uri the path is + /// made from, the name, and room for the address. + fn rowBytes(depth: usize, caller: Call) usize { + return 5 * (depth + 2) + caller.uri.len + caller.name.len + 40; + } /// Incoming paths arrive nearest-first. Insert them outermost-first, /// sharing common caller prefixes and retaining all their call sites. @@ -909,11 +978,14 @@ const CallTree = struct { const item = get(entry, "from").?; const children = if (parent) |id| tree.nodes.items[id].children.items else tree.roots.items; const found = for (children) |id| { - if (sameCall(tree.nodes.items[id].item, item)) break id; + if (sameCall(tree.nodes.items[id].call, item)) break id; } else null; + const depth = path.len - 1 - at; const id = found orelse blk: { const id = tree.nodes.items.len; - try tree.nodes.append(arena, .{ .item = item }); + try tree.nodes.append(arena, .{ .call = try callOf(arena, item) }); + tree.rows += 1; + tree.bytes += rowBytes(depth, tree.nodes.items[id].call); if (parent) |pid| { try tree.nodes.items[pid].children.append(arena, id); } else try tree.roots.append(arena, id); @@ -924,23 +996,39 @@ const CallTree = struct { for (items(get(entry, "fromRanges"))) |range| { const parsed = rangeOf(range) orelse continue; for (node.ranges.items) |*other| { - if (std.meta.eql(parsed, rangeOf(other.range).?)) { + if (std.meta.eql(parsed, other.range)) { other.focus = other.focus or at == 0; break; } - } else try node.ranges.append(arena, .{ .range = range, .focus = at == 0 }); + } else { + try node.ranges.append(arena, .{ .range = parsed, .focus = at == 0 }); + tree.rows += 1; + tree.bytes += rowBytes(depth + 1, .{ .uri = node.call.uri, .name = "", .at = parsed }); + } } parent = id; } } }; -fn sameCall(a: std.json.Value, b: std.json.Value) bool { - const au = str(get(a, "uri")) orelse return false; +fn callOf(arena: std.mem.Allocator, item: std.json.Value) Err!CallTree.Call { + return .{ + .uri = try arena.dupe(u8, str(get(item, "uri")).?), + .name = try arena.dupe(u8, str(get(item, "name")) orelse "(unnamed)"), + .at = rangeOf(get(item, "selectionRange") orelse get(item, "range")).?, + }; +} + +fn sameCall(a: CallTree.Call, b: std.json.Value) bool { const bu = str(get(b, "uri")) orelse return false; - const ar = rangeOf(get(a, "selectionRange") orelse get(a, "range")) orelse return false; const br = rangeOf(get(b, "selectionRange") orelse get(b, "range")) orelse return false; - return std.mem.eql(u8, au, bu) and std.meta.eql(ar, br); + return std.mem.eql(u8, a.uri, bu) and std.meta.eql(a.at, br); +} + +fn sameItem(a: std.json.Value, b: std.json.Value) bool { + const au = str(get(a, "uri")) orelse return false; + const ar = rangeOf(get(a, "selectionRange") orelse get(a, "range")) orelse return false; + return sameCall(.{ .uri = au, .name = "", .at = ar }, b); } fn callPrefix(out: *std.Io.Writer, tree: *const CallTree, depth: usize) Err!void { @@ -954,44 +1042,115 @@ fn validCaller(entry: std.json.Value) bool { rangeOf(get(from, "selectionRange") orelse get(from, "range")) != null; } -fn callstack(c: *Conn, si: usize, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, pos: Pos, deadline: i64, tr: *Trace, max_depth: u8) Err!void { +fn callstack(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, pos: Pos, deadline: i64, tr: *Trace, max_depth: u8, budget: ?*const std.heap.FixedBufferAllocator, arg: []const u8) Err!void { if (!c.caps.call_hier) return lsp.none(cx.out, "{s} does not answer call hierarchy (no callHierarchyProvider)", .{specs[si].name}); const b = try atPos(arena, uri, pos); - const prepared = (try call(c, arena, "textDocument/prepareCallHierarchy", b.items, deadline)) orelse return; + const prepared = (try call(c, io, arena, "textDocument/prepareCallHierarchy", b.items, deadline)) orelse return; if (items(prepared).len == 0) return lsp.none(cx.out, "no function at this position", .{}); + // The rows come from the tree alone: the core reads the sources. + cx.no_text = true; + // Each response is parsed into its own part of the region, freed once + // its subtree is walked; only the path being walked is live JSON. + // Raw, as query()'s scratch: `alloc` would fill all ~100 MB of it. + const parse_len = if (budget) |fixed| (fixed.buffer.len - fixed.end_index) / 2 else 0; + var parse_fba: std.heap.FixedBufferAllocator = .init(if (budget != null) + (arena.rawAlloc(parse_len, .of(u8), @returnAddress()) orelse return error.OutOfMemory)[0..parse_len] + else + &.{}); + const parse = if (budget != null) parse_fba.allocator() else arena; for (items(prepared)) |item| { - var tree: CallTree = .{ .depth_limit = std.math.clamp(max_depth, 1, lsp.callstack_max_depth) }; - const incoming = try hierarchyResult(c, arena, item, "callHierarchy/incomingCalls", deadline); - const count = try collectCallers(c, arena, item, incoming, deadline, &tree, 0); - if (count == 0) try cx.out.writeAll("No callers reported\n"); + var tree: CallTree = .{ + .depth_limit = std.math.clamp(max_depth, 1, lsp.callstack_max_depth), + .si = si, + .started_ms = nowMs(), + .budget = budget, + // The page ends while the answer's writer has room for the rest. + .byte_cap = if (cx.fixed_out) (cx.out.buffer.len - cx.out.end) -| (1 << 20) else std.math.maxInt(usize), + }; + var steps = std.mem.tokenizeScalar(u8, std.mem.trim(u8, arg, " \n"), '.'); + while (steps.next()) |step| { + if (tree.start_len == tree.start.len) break; + tree.start[tree.start_len] = std.fmt.parseInt(u32, step, 10) catch break; + tree.start_len += 1; + } + const incoming = try hierarchyResult(c, io, parse, item, "callHierarchy/incomingCalls", deadline); + const count = try collectCallers(c, io, arena, if (budget != null) &parse_fba else null, item, incoming, &tree, 0, tree.start_len > 0); + if (tree.narrated) post(si, .always, "Callstack: {d} callers", .{tree.collected}); + if (count == 0 and tree.start_len == 0) try cx.out.writeAll("No callers reported\n"); try renderCallers(cx, c.caps.enc, &tree, tree.roots.items, 0); - if (tree.limited) try cx.out.writeAll("… continue from a function above with Callstack\n"); + if (tree.more) { + try cx.out.writeAll(lsp.callstack_more); + for (tree.cut[0..tree.cut_len], 0..) |index, level| try cx.out.print("{s}{d}", .{ if (level == 0) "" else ".", index }); + try cx.out.print(" {d}\n", .{tree.collected}); + } else if (tree.out_of_time) { + try cx.out.writeAll("… the language server ran out of time; continue from a function above with Callstack\n"); + } else if (tree.out_of_memory) { + try cx.out.writeAll("… stopped at the query's memory limit; continue from a function above with Callstack\n"); + } else if (tree.limited) try cx.out.writeAll("… continue from a function above with Callstack\n"); tr.note("callstack: {d} callers", .{tree.rendered}); } } -fn collectCallers(c: *Conn, arena: std.mem.Allocator, focus: std.json.Value, incoming: ?std.json.Value, deadline: i64, tree: *CallTree, depth: usize) Err!usize { +/// No caller count and no overall deadline: the walk narrates itself on the +/// message row after a second, and it stops when a request goes unanswered +/// (what was collected is shown), when the page is full (the rest is the +/// next page), or when the shell cancels it for a newer query. +fn collectCallers(c: *Conn, io: std.Io, arena: std.mem.Allocator, parse: ?*std.heap.FixedBufferAllocator, focus: std.json.Value, incoming: ?std.json.Value, tree: *CallTree, depth: usize, on_path: bool) Err!usize { var count: usize = 0; - for (items(incoming)) |entry| { - if (tree.visited >= 64) { - tree.limited = true; - break; + for (items(incoming), 0..) |entry, index| { + // A later page skips what the pages before it walked. + const resuming = on_path and depth < tree.start_len; + if (resuming and index < tree.start[depth]) continue; + tree.at[depth] = @intCast(index); + if (tree.budget) |b| if (b.buffer.len - b.end_index < callstack_reserve) { + tree.out_of_memory = true; + }; + if (!tree.more and (tree.rows >= tree.row_cap or tree.bytes >= tree.byte_cap or tree.out_of_memory)) { + // The page is full: the next goes on here. Never where this + // one began, which would page in place: past that entry. + tree.more = true; + tree.cut_len = depth + 1; + @memcpy(tree.cut[0..tree.cut_len], tree.at[0..tree.cut_len]); + if (tree.cut_len == tree.start_len and std.mem.eql(u32, tree.cut[0..tree.cut_len], tree.start[0..tree.start_len])) + tree.cut[depth] += 1; } + if (tree.out_of_time or tree.more) break; if (!validCaller(entry)) continue; const from = get(entry, "from").?; - const cycle = sameCall(focus, from) or for (tree.path[0..depth]) |ancestor| { - if (sameCall(get(ancestor, "from").?, from)) break true; + const cycle = sameItem(focus, from) or for (tree.path[0..depth]) |ancestor| { + if (sameItem(get(ancestor, "from").?, from)) break true; } else false; tree.path[depth] = entry; - tree.visited += 1; count += 1; + tree.collected += 1; + if (nowMs() - tree.started_ms >= 1000) { + tree.narrated = true; + post(tree.si, .chatty, "Callstack: {d} callers\u{2026}", .{tree.collected}); + } var children: usize = 0; if (!cycle) { - if (depth + 1 == tree.depth_limit or tree.visited >= 64) { + if (depth + 1 == tree.depth_limit) { tree.limited = true; } else { - const ancestors = try hierarchyResult(c, arena, from, "callHierarchy/incomingCalls", deadline); - children = try collectCallers(c, arena, focus, ancestors, deadline, tree, depth + 1); + const mark = if (parse) |fixed| fixed.end_index else 0; + defer if (parse) |fixed| { + fixed.end_index = mark; + }; + const ancestors = hierarchyResult(c, io, if (parse) |fixed| fixed.allocator() else arena, from, "callHierarchy/incomingCalls", nowMs() + req_ms) catch |e| switch (e) { + error.Timeout => blk: { + tree.out_of_time = true; + break :blk null; + }, + // A response bigger than the parse region: this caller + // stops here, the page with it. + error.OutOfMemory => blk: { + tree.out_of_memory = true; + break :blk null; + }, + else => return e, + }; + const deeper = resuming and index == tree.start[depth] and depth + 1 < tree.start_len; + if (!tree.out_of_time and !tree.out_of_memory) children = try collectCallers(c, io, arena, parse, focus, ancestors, tree, depth + 1, deeper); } } if (children == 0) try tree.addPath(arena, tree.path[0 .. depth + 1], cycle); @@ -999,51 +1158,61 @@ fn collectCallers(c: *Conn, arena: std.mem.Allocator, focus: std.json.Value, inc return count; } +/// Test callers sink below the rest; each group keeps the usual order. +// ponytail: substring heuristic, "latest.zig" or "contest" count as tests too +fn testCaller(caller: CallTree.Call) bool { + return std.ascii.findIgnoreCase(caller.name, "test") != null or + std.ascii.findIgnoreCase(std.fs.path.basenamePosix(caller.uri), "test") != null; +} + fn callerLess(tree: *const CallTree, a: usize, b: usize) bool { - const left = tree.nodes.items[a].item; - const right = tree.nodes.items[b].item; - const uri_order = std.mem.order(u8, str(get(left, "uri")).?, str(get(right, "uri")).?); + const left = tree.nodes.items[a].call; + const right = tree.nodes.items[b].call; + if (testCaller(left) != testCaller(right)) return testCaller(right); + const uri_order = std.mem.order(u8, left.uri, right.uri); if (uri_order != .eq) return uri_order == .lt; - return callRangeLess({}, get(left, "selectionRange") orelse get(left, "range").?, get(right, "selectionRange") orelse get(right, "range").?); + return rangeLess(left.at, right.at); } -fn callRangeLess(_: void, a: std.json.Value, b: std.json.Value) bool { - const left = rangeOf(a).?; - const right = rangeOf(b).?; +fn rangeLess(left: Range, right: Range) bool { return left.sl < right.sl or (left.sl == right.sl and left.sc < right.sc); } fn renderCallers(cx: *Cx, enc: Enc, tree: *CallTree, children: []usize, depth: usize) Err!void { std.mem.sort(usize, children, @as(*const CallTree, tree), callerLess); for (children, 0..) |id, index| { - if (tree.rendered >= 64) { + // emitRange's row ceiling and the writer's room, met here before a + // prefix is half written + if (cx.rows >= max_rows or (cx.fixed_out and cx.out.buffer.len - cx.out.end < row_room)) { tree.limited = true; return; } const node = &tree.nodes.items[id]; - const item = node.item; - const uri = str(get(item, "uri")).?; const more = index + 1 < children.len; try callPrefix(cx.out, tree, depth); const has_children = node.children.items.len > 0 or node.ranges.items.len > 0; const rounded = cx.callstack_corners == .rounded; try cx.out.writeAll(if (more) "├" else if (rounded) "╰" else "└"); try cx.out.writeAll(if (has_children) "──┬ " else "─── "); - const name = flat(cx.arena, str(get(item, "name")) orelse "(unnamed)"); + const name = flat(cx.arena, node.call.name); const label = if (node.cycle) try std.fmt.allocPrint(cx.arena, "{s} ↻ recursion", .{name}) else name; - try emitRange(cx, enc, uri, get(item, "selectionRange") orelse get(item, "range"), label); + try emitAt(cx, enc, node.call.uri, node.call.at, label); tree.rendered += 1; tree.siblings[depth] = more; std.mem.sort(CallTree.Site, node.ranges.items, {}, struct { fn less(_: void, a: CallTree.Site, b: CallTree.Site) bool { - return callRangeLess({}, a.range, b.range); + return rangeLess(a.range, b.range); } }.less); for (node.ranges.items, 0..) |site, site_index| { + if (cx.rows >= max_rows or (cx.fixed_out and cx.out.buffer.len - cx.out.end < row_room)) { + tree.limited = true; + return; + } if (site.focus) try cx.out.writeAll(lsp.callstack_match); try callPrefix(cx.out, tree, depth + 1); try cx.out.writeAll(if (node.children.items.len > 0 or site_index + 1 < node.ranges.items.len) "├─ " else if (rounded) "╰─ " else "└─ "); - try emitRange(cx, enc, uri, site.range, ""); + try emitAt(cx, enc, node.call.uri, site.range, ""); } try renderCallers(cx, enc, tree, node.children.items, depth + 1); } @@ -1054,7 +1223,7 @@ const Hier = enum { incoming, outgoing, supers, subs }; /// The two-step hierarchy kinds: prepare at the cursor, then follow every /// item the server returned (usually one). Gated on the server capability so /// an old server costs zero round trips. -fn hierarchy(c: *Conn, si: usize, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, pos: Pos, deadline: i64, h: Hier, tr: *Trace) Err!void { +fn hierarchy(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, cx: *Cx, uri: []const u8, pos: Pos, deadline: i64, h: Hier, tr: *Trace) Err!void { const call_side = h == .incoming or h == .outgoing; if (call_side and !c.caps.call_hier) { tr.note("STOP: {s} does not advertise callHierarchyProvider", .{specs[si].name}); @@ -1072,9 +1241,9 @@ fn hierarchy(c: *Conn, si: usize, arena: std.mem.Allocator, cx: *Cx, uri: []cons .subs => "typeHierarchy/subtypes", }; const b = try atPos(arena, uri, pos); - const prepared = (try call(c, arena, prepare, b.items, deadline)) orelse return; + const prepared = (try call(c, io, arena, prepare, b.items, deadline)) orelse return; for (items(prepared)) |item| { - const result = (try hierarchyResult(c, arena, item, follow, deadline)) orelse return; + const result = (try hierarchyResult(c, io, arena, item, follow, deadline)) orelse return; for (items(result)) |entry| switch (h) { .incoming => { // one row per CALL SITE, under the caller's name — that is @@ -1134,6 +1303,10 @@ const Cx = struct { cur_path: []const u8, cur_src: []const u8, out: *std.Io.Writer, + /// `out` is the query's fixed answer buffer, whose room is checked. + fixed_out: bool = false, + /// Rows without their source text (Callstack: the core reads it). + no_text: bool = false, rows: usize = 0, relative_parents: bool = false, callstack_corners: lsp.CallstackCorners = .rounded, @@ -1169,11 +1342,16 @@ const Cx = struct { /// columns, and a single-line range becomes the `path:LINE:COL-ENDCOL` form a /// look SELECTS. `note` overrides the source line as the row's text. fn emitRange(cx: *Cx, enc: Enc, uri: []const u8, range: ?std.json.Value, note: []const u8) Err!void { + return emitAt(cx, enc, uri, rangeOf(range) orelse return, note); +} + +fn emitAt(cx: *Cx, enc: Enc, uri: []const u8, r: Range, note: []const u8) Err!void { if (cx.rows >= max_rows) return; const path = pathOf(cx.arena, uri) orelse return; if (path.len == 0 or path[0] != '/') return; // rows promise absolute-or-rel-from-base - const r = rangeOf(range) orelse return; - const src = cx.text(path); + // Without sources (Callstack) the asking file is still known: its rows' + // columns convert from utf-16 as ever. + const src = if (cx.no_text and !std.mem.eql(u8, path, cx.cur_path)) null else cx.text(path); var bol: usize = 0; var eol: usize = 0; @@ -1323,7 +1501,8 @@ fn transportPair(sv: *[2]libc.fd_t) bool { /// Spawn-or-return, and tell an existing server about a new project root. /// Returns the generation the caller's whole query is bound to. -fn ensure(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trace) Err!u32 { +fn ensure(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trace) Err!u32 { + c.si = si; if (c.state == .disabled) { tr.note("STOP: {s} is disabled for this session (no binary; set {s})", .{ specs[si].name, specs[si].env }); return error.NoServer; @@ -1341,7 +1520,7 @@ fn ensure(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trac try app(&b, arena, "{\"jsonrpc\":\"2.0\",\"method\":\"workspace/didChangeWorkspaceFolders\",\"params\":{\"event\":{\"added\":[{\"name\":\"root\",\"uri\":\""); try uriOf(&b, arena, root); try app(&b, arena, "\"}],\"removed\":[]}}}"); - try frame(c.sock, b.items, nowMs() + reply_ms); + try send(c, io, b.items, nowMs() + wedge_ms); const owned = sa.dupe(u8, root) catch return error.OutOfMemory; c.extra_roots.append(sa, owned) catch sa.free(owned); } @@ -1430,12 +1609,13 @@ fn ensure(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trac if (c.resp) |r| sa.free(r); c.resp = null; c.progress = 0; + c.loading = false; c.timeouts = 0; forgetDocs(c); if (c.root.len > 0) sa.free(c.root); c.root = owned_root; - const th = std.Thread.spawn(.{}, reader, .{ si, g, sv[0] }) catch { + const th = std.Thread.spawn(.{}, reader, .{ io, si, g, sv[0] }) catch { // no reader means nobody would ever close the fd our way; do it here, // before anything else can see the conn _ = libc.close(sv[0]); @@ -1450,8 +1630,14 @@ fn ensure(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trac post(si, .always, "{s} starting — {s}", .{ specs[si].name, root }); tr.note("spawned {s} (pid {d}) at {s}", .{ specs[si].name, pid, root }); - handshake(c, arena, root) catch |e| { - shutdownIf(c, g); + // A start runs to the end however soon the query that began it is + // canceled: killing a half-started server would only make the next + // query start it again. The cancel lands at the next wait instead. + const protection = io.swapCancelProtection(.blocked); + const started = handshake(c, io, arena, root); + _ = io.swapCancelProtection(protection); + started catch |e| { + shutdownIf(c, io, g); // NOT a session disable. A handshake that misses the deadline is // routinely environmental — a rustup shim deciding to download the // project's pinned toolchain before launching the real server was @@ -1477,7 +1663,7 @@ fn ensure(c: *Conn, si: usize, arena: std.mem.Allocator, req: lsp.Req, tr: *Trac return g; } -fn handshake(c: *Conn, arena: std.mem.Allocator, root: []const u8) Err!void { +fn handshake(c: *Conn, io: std.Io, arena: std.mem.Allocator, root: []const u8) Err!void { var b: std.ArrayList(u8) = .empty; try b.print(arena, "\"processId\":{d},\"clientInfo\":{{\"name\":\"pardes\"}},\"rootUri\":\"", .{@as(u32, @bitCast(libc.getpid()))}); try uriOf(&b, arena, root); @@ -1509,7 +1695,7 @@ fn handshake(c: *Conn, arena: std.mem.Allocator, root: []const u8) Err!void { for (b.items) |ch| if (ch != '\n') try body.append(arena, ch); const deadline = nowMs() + init_ms; - const reply = (try call(c, arena, "initialize", body.items, deadline)) orelse return error.Protocol; + const reply = (try call(c, io, arena, "initialize", body.items, deadline)) orelse return error.Protocol; const caps = get(reply, "capabilities"); c.caps.enc = if (std.mem.eql(u8, str(get(caps, "positionEncoding")) orelse "utf-16", "utf-8")) .utf8 else .utf16; @@ -1521,7 +1707,7 @@ fn handshake(c: *Conn, arena: std.mem.Allocator, root: []const u8) Err!void { c.caps.type_hier = provider(get(caps, "typeHierarchyProvider")); c.caps.folders = if (get(get(get(caps, "workspace"), "workspaceFolders"), "supported")) |s| s == .bool and s.bool else false; - try frame(c.sock, "{\"jsonrpc\":\"2.0\",\"method\":\"initialized\",\"params\":{}}", deadline); + try send(c, io, "{\"jsonrpc\":\"2.0\",\"method\":\"initialized\",\"params\":{}}", deadline); } /// A server capability that may be `true`, an options object, or absent. @@ -1536,7 +1722,7 @@ fn provider(v: ?std.json.Value) bool { /// didOpen the first time a file is seen, didChange when its bytes moved, /// nothing when they did not — the common case between two presses of gd. -fn syncDoc(c: *Conn, si: usize, arena: std.mem.Allocator, uri: []const u8, src: []const u8) Err!void { +fn syncDoc(c: *Conn, io: std.Io, si: usize, arena: std.mem.Allocator, uri: []const u8, src: []const u8) Err!void { const hash = std.hash.Wyhash.hash(0, src); var doc: ?*Doc = null; for (c.docs.items) |*d| if (std.mem.eql(u8, d.uri, uri)) { @@ -1568,13 +1754,16 @@ fn syncDoc(c: *Conn, si: usize, arena: std.mem.Allocator, uri: []const u8, src: try jstr(&b, arena, src); try app(&b, arena, "}}}"); } - try frame(c.sock, b.items, nowMs() + req_ms); + // A server busy with what it was sent last reads no more meanwhile + // (clangd parses its compile database at a first open): this waits as + // an answer does, for a cancel or a server that has stopped. + try send(c, io, b.items, nowMs() + wedge_ms); } /// Send a request and wait on the mailbox. Returns the parsed `result`, or /// null for an error response / null result — both are legal "no answer". /// Runs with the conn mutex held; the mutex is released inside the wait. -fn call(c: *Conn, arena: std.mem.Allocator, method: []const u8, params: []const u8, deadline: i64) Err!?std.json.Value { +fn call(c: *Conn, io: std.Io, arena: std.mem.Allocator, method: []const u8, params: []const u8, deadline: i64) Err!?std.json.Value { const g = c.gen; const id = c.next_id; c.next_id +%= 1; @@ -1589,17 +1778,36 @@ fn call(c: *Conn, arena: std.mem.Allocator, method: []const u8, params: []const c.want_id = id; defer c.want_id = 0; - try frame(c.sock, b.items, deadline); + try send(c, io, b.items, if (c.state == .ready) nowMs() + wedge_ms else deadline); + // The handshake (still `.starting`) has its deadline. A ready server is + // waited for until it answers or the shell cancels this query for a + // newer one, so the first query after a start is answered, not lost: + // past `deadline` the message row says it is still working, and only + // silent and idle for wedge_ms is it not answering. + var said = false; + var busy_at = nowMs(); while (c.resp == null and c.gen == g and c.alive()) { - if (!timedWait(c, deadline)) break; + const ready = c.state == .ready; + const waiting = timedWait(c, io, if (ready) nowMs() + 500 else deadline) catch |e| { + cancelRequest(c, io, arena, id); + return e; + }; + if (!ready) { + if (waiting) continue; + break; + } + if (c.progress > 0) busy_at = nowMs(); + if (!said and nowMs() >= deadline) { + said = true; + post(c.si, .always, "{s} is still working on it\u{2026}", .{specs[c.si].name}); + } + if (nowMs() - busy_at >= wedge_ms) break; } + if (said and c.resp != null) post(c.si, .always, "{s} answered", .{specs[c.si].name}); if (c.gen != g or !c.alive()) return error.Dead; const raw = c.resp orelse { - // give the server leave to abandon the work we stopped waiting for - var cb: std.ArrayList(u8) = .empty; - cb.print(arena, "{{\"jsonrpc\":\"2.0\",\"method\":\"$/cancelRequest\",\"params\":{{\"id\":{d}}}}}", .{id}) catch return error.Timeout; - frame(c.sock, cb.items, nowMs() + reply_ms) catch {}; + cancelRequest(c, io, arena, id); return error.Timeout; }; c.resp = null; @@ -1609,49 +1817,23 @@ fn call(c: *Conn, arena: std.mem.Allocator, method: []const u8, params: []const if (result == .null) return null; return result; } -/// One bounded cond wait. False when the deadline has passed. -/// -/// The deadline arithmetic everywhere here is CLOCK_MONOTONIC, so the wait -/// must be too: a REALTIME abstime plus an NTP step backwards would park a -/// worker far past `req_ms`, violating the seam's bounded-wait promise. On -/// linux the condvar is lazily initialized with a monotonic condattr; darwin -/// has no setclock but has a RELATIVE wait, which never consults the wall -/// clock at all. -fn timedWait(c: *Conn, deadline: i64) bool { +/// Give the server leave to abandon the work we stopped waiting for. +fn cancelRequest(c: *Conn, io: std.Io, arena: std.mem.Allocator, id: u32) void { + var cb: std.ArrayList(u8) = .empty; + cb.print(arena, "{{\"jsonrpc\":\"2.0\",\"method\":\"$/cancelRequest\",\"params\":{{\"id\":{d}}}}}", .{id}) catch return; + send(c, io, cb.items, nowMs() + reply_ms) catch {}; +} + +/// One bounded cond wait, and a cancelation point: a worker the shell +/// cancels for a newer query stops here. False when the deadline has passed. +fn timedWait(c: *Conn, io: std.Io, deadline: i64) error{Canceled}!bool { const left = deadline - nowMs(); if (left <= 0) return false; const wait_ms = @min(left, 500); // tick so gen/state changes are noticed - if (comptime builtin.os.tag.isDarwin()) { - const rel: libc.timespec = .{ - .sec = @divTrunc(wait_ms, 1000), - .nsec = @rem(wait_ms, 1000) * 1_000_000, - }; - _ = pthread_cond_timedwait_relative_np(&c.cond, &c.mu, &rel); - return true; - } - if (!c.cond_monotonic) { - // lazily re-initialize the zeroed (REALTIME) condvar with a - // monotonic clock; under the mutex, and only before the first wait, - // so no thread can be parked on the old one - var attr: pthread_condattr_t = undefined; - if (pthread_condattr_init(&attr) == 0) { - _ = pthread_condattr_setclock(&attr, CLOCK_MONOTONIC); - _ = pthread_cond_init(&c.cond, &attr); - _ = pthread_condattr_destroy(&attr); - } - c.cond_monotonic = true; - } - var now: libc.timespec = undefined; - _ = libc.clock_gettime(.MONOTONIC, &now); - var abs: libc.timespec = .{ - .sec = now.sec + @divTrunc(wait_ms, 1000), - .nsec = now.nsec + @rem(wait_ms, 1000) * 1_000_000, + c.cond.waitTimeout(io, &c.mu, .{ .duration = .{ .raw = .fromMilliseconds(wait_ms), .clock = .awake } }) catch |e| switch (e) { + error.Timeout => {}, + error.Canceled => return error.Canceled, }; - if (abs.nsec >= 1_000_000_000) { - abs.sec += 1; - abs.nsec -= 1_000_000_000; - } - _ = libc.pthread_cond_timedwait(&c.cond, &c.mu, &abs); return true; } @@ -1659,7 +1841,7 @@ fn timedWait(c: *Conn, deadline: i64) bool { /// respawn that happened during the caller's wait must not be killed for its /// predecessor's crimes. Mutex held. The reader closes the fd; this only /// shuts the transport down and reaps the child. -fn shutdownIf(c: *Conn, g: u32) void { +fn shutdownIf(c: *Conn, io: std.Io, g: u32) void { if (c.gen != g or c.sock < 0) return; _ = libc.shutdown(c.sock, 2); // SHUT_RDWR: the reader sees EOF and closes reap(c.pid); @@ -1669,7 +1851,7 @@ fn shutdownIf(c: *Conn, g: u32) void { forgetDocs(c); if (c.resp) |r| sa.free(r); c.resp = null; - _ = libc.pthread_cond_broadcast(&c.cond); + c.cond.broadcast(io); } /// TERM, a bounded grace, then KILL, to the server's whole process group @@ -1715,6 +1897,7 @@ fn forgetDocs(c: *Conn) void { sa.free(d.body); } c.diags.clearRetainingCapacity(); + c.diag_bytes = 0; } // ------------------------------------------------------------ reader thread @@ -1722,20 +1905,22 @@ fn forgetDocs(c: *Conn) void { /// Owns the read side of one socket for one generation, and is the only /// closer of that fd. Routes responses to the mailbox, answers server /// requests, feeds the diagnostics store, narrates progress. -fn reader(si: usize, g: u32, sock: c_int) void { +fn reader(io: std.Io, si: usize, g: u32, sock: c_int) void { const c = &conns[si]; var buf = sa.alloc(u8, 256 * 1024) catch { - readerExit(c, si, g, sock); + readerExit(c, io, si, g, sock); return; }; defer sa.free(buf); var len: usize = 0; + var parse_arena: std.heap.ArenaAllocator = .init(sa); + defer parse_arena.deinit(); outer: while (true) { // drain every complete frame already buffered drain: while (true) switch (frameNext(buf[0..len])) { .frame => |m| { - dispatch(c, si, g, sock, buf[m.body_start..m.body_end]); + dispatch(c, io, si, g, sock, buf[m.body_start..m.body_end], &parse_arena); std.mem.copyForwards(u8, buf[0 .. len - m.consumed], buf[m.consumed..len]); len -= m.consumed; }, @@ -1761,9 +1946,9 @@ fn reader(si: usize, g: u32, sock: c_int) void { if (pr == 0) { // tick: a respawn while the server was silent leaves this thread // reading a corpse; notice and go - c.lock(); + c.mu.lockUncancelable(io); const stale = c.gen != g; - c.unlock(); + c.mu.unlock(io); if (stale) break :outer; continue; } @@ -1776,11 +1961,11 @@ fn reader(si: usize, g: u32, sock: c_int) void { if (got == 0) break :outer; // the server exited or shutdownIf spoke len += @intCast(got); } - readerExit(c, si, g, sock); + readerExit(c, io, si, g, sock); } -fn readerExit(c: *Conn, si: usize, g: u32, sock: c_int) void { - c.lock(); +fn readerExit(c: *Conn, io: std.Io, si: usize, g: u32, sock: c_int) void { + c.mu.lockUncancelable(io); if (c.gen == g and c.alive()) { // the server died on its own — shutdownIf never saw it, so the corpse // is ours to reap and the conn's transport fields are ours to clear @@ -1800,10 +1985,14 @@ fn readerExit(c: *Conn, si: usize, g: u32, sock: c_int) void { } else { post(si, .always, "{s} exited — restarts on the next query", .{specs[si].name}); } - _ = libc.pthread_cond_broadcast(&c.cond); + c.cond.broadcast(io); } - c.unlock(); + c.mu.unlock(io); + // A query writes with the mutex let go (send): the fd stays its until + // that write is done, and shutdown(2) has already made it fail fast. + c.write_mu.lockUncancelable(io); _ = libc.close(sock); + c.write_mu.unlock(io); } /// 10s doubling to 2min: the fork-per-keystroke guard's schedule. @@ -1845,32 +2034,33 @@ fn frameNext(data: []const u8) FrameStep { /// One message from the server. Parses into a throwaway arena; touches conn /// state only under the mutex; replies never hold it (bounded writes must not /// stall the mailbox). -fn dispatch(c: *Conn, si: usize, g: u32, sock: c_int, raw: []const u8) void { - var arena_state: std.heap.ArenaAllocator = .init(sa); - defer arena_state.deinit(); +fn dispatch(c: *Conn, io: std.Io, si: usize, g: u32, sock: c_int, raw: []const u8, arena_state: *std.heap.ArenaAllocator) void { + // The reader's one parse arena, reset per message: no mapping per + // `$/progress` report, and no more than parse_kept kept between them. + defer _ = arena_state.reset(.{ .retain_with_limit = parse_kept }); const arena = arena_state.allocator(); const v = std.json.parseFromSliceLeaky(std.json.Value, arena, raw, .{}) catch return; if (v != .object) return; if (str(get(v, "method"))) |method| { - if (get(v, "id")) |id| return serverRequest(arena, sock, method, id, get(v, "params")); - return notification(c, si, g, raw, method, get(v, "params")); + if (get(v, "id")) |id| return serverRequest(c, io, arena, sock, method, id, get(v, "params")); + return notification(c, io, si, g, raw, method, get(v, "params")); } // a response: deliver to the one waiting query, if it is still waiting const id = num(get(v, "id")) orelse return; - c.lock(); - defer c.unlock(); + c.mu.lockUncancelable(io); + defer c.mu.unlock(io); if (c.gen == g and c.want_id != 0 and id == c.want_id and c.resp == null) { c.resp = sa.dupe(u8, raw) catch null; - _ = libc.pthread_cond_broadcast(&c.cond); + c.cond.broadcast(io); } } /// The server asked US something. Everything optional was declined in the /// handshake, so a null-ish answer is always legal — but it must ARRIVE, or /// the server blocks forever on its own question. -fn serverRequest(arena: std.mem.Allocator, sock: c_int, method: []const u8, id: std.json.Value, params: ?std.json.Value) void { +fn serverRequest(c: *Conn, io: std.Io, arena: std.mem.Allocator, sock: c_int, method: []const u8, id: std.json.Value, params: ?std.json.Value) void { var b: std.ArrayList(u8) = .empty; app(&b, arena, "{\"jsonrpc\":\"2.0\",\"id\":") catch return; switch (id) { @@ -1890,17 +2080,17 @@ fn serverRequest(arena: std.mem.Allocator, sock: c_int, method: []const u8, id: } else { app(&b, arena, ",\"result\":null}") catch return; } - frame(sock, b.items, nowMs() + reply_ms) catch {}; + answerServer(c, io, sock, b.items); } -fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8, params: ?std.json.Value) void { +fn notification(c: *Conn, io: std.Io, si: usize, g: u32, raw: []const u8, method: []const u8, params: ?std.json.Value) void { if (std.mem.eql(u8, method, "textDocument/publishDiagnostics")) { const uri = str(get(params, "uri")) orelse return; // keep the raw MESSAGE (params included); re-parsing it per query // beats owning a parsed tree past this arena's life const body = sa.dupe(u8, raw) catch return; - c.lock(); - defer c.unlock(); + c.mu.lockUncancelable(io); + defer c.mu.unlock(io); if (c.gen != g) { sa.free(body); return; @@ -1914,10 +2104,25 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 slot = d; break; }; + // Bounded in files and bytes: a new set that does not fit pushes + // out the oldest files' (the server sends a file's again when it + // changes); one bigger than the whole store is not kept. + if (body.len > max_diag_bytes) { + sa.free(body); + return; + } if (slot) |d| { + c.diag_bytes -= d.body.len; sa.free(d.body); d.body = body; - } else if (c.diags.items.len < 256) { + c.diag_bytes += body.len; + } else { + while (c.diags.items.len > 0 and (c.diags.items.len >= max_diag_files or c.diag_bytes + body.len > max_diag_bytes)) { + const old = c.diags.orderedRemove(0); + c.diag_bytes -= old.body.len; + sa.free(old.uri); + sa.free(old.body); + } const owned = sa.dupe(u8, uri) catch { sa.free(body); return; @@ -1927,17 +2132,18 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 sa.free(body); return; }; - } else sa.free(body); - _ = libc.pthread_cond_broadcast(&c.cond); // waitFresh watches this + c.diag_bytes += body.len; + } + c.cond.broadcast(io); // waitFresh watches this return; } if (std.mem.eql(u8, method, "$/progress")) { const value = get(params, "value") orelse return; const pkind = str(get(value, "kind")) orelse return; const msg = str(get(value, "message")) orelse ""; - c.lock(); + c.mu.lockUncancelable(io); if (c.gen != g) { - c.unlock(); + c.mu.unlock(io); return; } if (std.mem.eql(u8, pkind, "begin")) { @@ -1948,7 +2154,7 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 var tbuf: [48]u8 = undefined; const t = tbuf[0..c.title_len]; @memcpy(t, c.title[0..c.title_len]); - c.unlock(); + c.mu.unlock(io); post(si, .chatty, "{s}: {s}\u{2026} {s}", .{ specs[si].name, t, msg }); return; } @@ -1957,7 +2163,7 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 const t = tbuf[0..c.title_len]; @memcpy(t, c.title[0..c.title_len]); const pct = num(get(value, "percentage")); - c.unlock(); + c.mu.unlock(io); if (pct) |p| post(si, .chatty, "{s}: {s} {d}% {s}", .{ specs[si].name, t, p, msg }) else @@ -1968,7 +2174,7 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 c.progress -= 1; const idle = c.progress <= 0; if (idle) c.progress = 0; - c.unlock(); + c.mu.unlock(io); // the setpoint every burst of progress ends on; `.always` + the // dedupe above means it lands exactly once however the reports raced if (idle) post(si, .always, "{s}: ready", .{specs[si].name}); @@ -1985,6 +2191,12 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 if (std.mem.eql(u8, method, "experimental/serverStatus")) { const health = str(get(params, "health")) orelse "ok"; const quiescent = if (get(params, "quiescent")) |q| q == .bool and q.bool else false; + c.mu.lockUncancelable(io); + if (c.gen == g) { + c.loading = !quiescent; + c.cond.broadcast(io); + } + c.mu.unlock(io); if (!std.mem.eql(u8, health, "ok")) post(si, .always, "{s}: {s} — {s}", .{ specs[si].name, health, str(get(params, "message")) orelse "" }) else if (quiescent) @@ -1995,12 +2207,12 @@ fn notification(c: *Conn, si: usize, g: u32, raw: []const u8, method: []const u8 // ------------------------------------------------------------- introspection -fn status(req: lsp.Req, out: *std.Io.Writer) !void { +fn status(io: std.Io, req: lsp.Req, out: *std.Io.Writer) !void { _ = req; try out.print("\nprotocol servers (lsp-client):\n", .{}); for (&specs, 0..) |*s, i| { const c = &conns[i]; - c.lock(); + c.mu.lockUncancelable(io); const state = c.state; var rootbuf: [256]u8 = undefined; const rootlen = @min(c.root.len, rootbuf.len); @@ -2012,7 +2224,7 @@ fn status(req: lsp.Req, out: *std.Io.Writer) !void { const pull = c.caps.pull; const chier = c.caps.call_hier; const thier = c.caps.type_hier; - c.unlock(); + c.mu.unlock(io); const env_v = if (getenv(s.env)) |v| std.mem.span(v) else null; // Never spawned is not off: the first query of its files starts it. try out.print(" {s:<28} {s}", .{ s.name, if (state == .off) "not started yet (its first query starts it)" else @tagName(state) }); @@ -2069,21 +2281,80 @@ fn atPos(arena: std.mem.Allocator, uri: []const u8, pos: Pos) Err!std.ArrayList( return b; } -fn frame(sock: c_int, body: []const u8, deadline: i64) Err!void { +/// A message to the server from a query, the conn's mutex let go while it is +/// written: the reader goes on taking the server's output meanwhile, which a +/// server may be waiting to write before it reads more of ours. +fn send(c: *Conn, io: std.Io, body: []const u8, deadline: i64) Err!void { + const sock = c.sock; + c.mu.unlock(io); + defer c.mu.lockUncancelable(io); + return frame(c, io, sock, body, deadline); +} + +fn frame(c: *Conn, io: std.Io, sock: c_int, body: []const u8, deadline: i64) Err!void { if (sock < 0) return error.Dead; + c.write_mu.lockUncancelable(io); + defer releaseWrite(c, io, sock); var hdr: [64]u8 = undefined; const h = std.fmt.bufPrint(&hdr, "Content-Length: {d}\r\n\r\n", .{body.len}) catch return error.Protocol; - try sockWrite(sock, h, deadline); - try sockWrite(sock, body, deadline); + var started = false; + try sockWrite(io, sock, h, deadline, &started); + try sockWrite(io, sock, body, deadline, &started); +} + +/// The reader's answer to a server request: written now if the socket is +/// free, else queued for the query writing (bounded; one that does not fit is +/// dropped, and the server's request goes unanswered). +fn answerServer(c: *Conn, io: std.Io, sock: c_int, body: []const u8) void { + var hdr: [64]u8 = undefined; + const h = std.fmt.bufPrint(&hdr, "Content-Length: {d}\r\n\r\n", .{body.len}) catch return; + c.replies_mu.lockUncancelable(io); + const half = &c.replies[c.filling]; + const len = &c.replies_len[c.filling]; + if (len.* + h.len + body.len <= half.len) { + @memcpy(half[len.*..][0..h.len], h); + @memcpy(half[len.* + h.len ..][0..body.len], body); + len.* += h.len + body.len; + } + c.replies_mu.unlock(io); + if (c.write_mu.tryLock()) releaseWrite(c, io, sock); +} + +/// Lets the socket go, first sending what the reader queued meanwhile, and +/// again for what it queues while that goes out. +fn releaseWrite(c: *Conn, io: std.Io, sock: c_int) void { + while (true) { + c.replies_mu.lockUncancelable(io); + const sent = c.filling; + const n = c.replies_len[sent]; + if (n > 0) c.filling ^= 1; + c.replies_mu.unlock(io); + if (n > 0) { + var started = false; + sockWrite(null, sock, c.replies[sent][0..n], nowMs() + reply_ms, &started) catch {}; + c.replies_len[sent] = 0; + } + c.write_mu.unlock(io); + c.replies_mu.lockUncancelable(io); + const more = c.replies_len[c.filling] > 0; + c.replies_mu.unlock(io); + if (!more or !c.write_mu.tryLock()) return; + } } -fn sockWrite(sock: c_int, bytes: []const u8, deadline: i64) Err!void { +/// Writes `bytes`, `started` once any byte of their message is out. Before +/// then, the deadline and a cancel stop it; after, the message is finished +/// however long the server takes to read it (bounded by wedge_ms), since +/// half of one ends the server's reading of every message after it. +fn sockWrite(io: ?std.Io, sock: c_int, bytes: []const u8, deadline: i64, started: *bool) Err!void { var off: usize = 0; while (off < bytes.len) { - const left = deadline - nowMs(); - if (left <= 0) return error.Timeout; + if (!started.*) { + if (deadline - nowMs() <= 0) return error.Timeout; + if (io) |cancel_io| cancel_io.checkCancel() catch return error.Canceled; + } else if (nowMs() - deadline > wedge_ms) return error.Dead; var pfd = [_]libc.pollfd{.{ .fd = sock, .events = POLLOUT, .revents = 0 }}; - const pr = libc.poll(&pfd, 1, @intCast(@min(left, 1000))); + const pr = libc.poll(&pfd, 1, 500); if (pr < 0) { if (libc.errno(pr) == .INTR) continue; return error.Dead; @@ -2096,6 +2367,7 @@ fn sockWrite(sock: c_int, bytes: []const u8, deadline: i64) Err!void { return error.Dead; } if (n == 0) return error.Dead; + started.* = true; off += @intCast(n); } } @@ -2318,7 +2590,7 @@ test "caller trees reverse incoming paths and merge shared callers and sites" { try std.testing.expectEqual(@as(usize, 1), tree.roots.items.len); try std.testing.expectEqual(@as(usize, 4), tree.nodes.items.len); const root = tree.nodes.items[tree.roots.items[0]]; - try std.testing.expectEqualStrings("main", str(get(root.item, "name")).?); + try std.testing.expectEqualStrings("main", root.call.name); try std.testing.expectEqual(@as(usize, 1), root.ranges.items.len); const page = tree.nodes.items[root.children.items[0]]; try std.testing.expectEqual(@as(usize, 2), page.children.items.len); @@ -2355,6 +2627,32 @@ test "caller trees reverse incoming paths and merge shared callers and sites" { try std.testing.expect(recursive.nodes.items[1].cycle); } +test "test callers sort after the rest, each group by place" { + var arena_state = std.heap.ArenaAllocator.init(std.testing.allocator); + defer arena_state.deinit(); + const arena = arena_state.allocator(); + const fixture = try std.json.parseFromSliceLeaky(std.json.Value, arena, "[{\"name\":\"test b\",\"uri\":\"file:///a.zig\",\"selectionRange\":{\"start\":{\"line\":0,\"character\":0},\"end\":{\"line\":0,\"character\":1}}},{\"name\":\"z\",\"uri\":\"file:///b.zig\",\"selectionRange\":{\"start\":{\"line\":0,\"character\":0},\"end\":{\"line\":0,\"character\":1}}},{\"name\":\"x\",\"uri\":\"file:///a_test.zig\",\"selectionRange\":{\"start\":{\"line\":0,\"character\":0},\"end\":{\"line\":0,\"character\":1}}},{\"name\":\"y\",\"uri\":\"file:///a.zig\",\"selectionRange\":{\"start\":{\"line\":5,\"character\":0},\"end\":{\"line\":5,\"character\":1}}}]", .{}); + var tree: CallTree = .{}; + for (items(fixture)) |item| try tree.nodes.append(arena, .{ .call = try callOf(arena, item) }); + var order = [_]usize{ 0, 1, 2, 3 }; + std.mem.sort(usize, &order, @as(*const CallTree, &tree), callerLess); + try std.testing.expectEqualSlices(usize, &.{ 3, 1, 0, 2 }, &order); +} + +test "a mailbox wait is a cancelation point: a canceled worker stops waiting" { + const io = std.testing.io; + var c: Conn = .{}; + const Waiter = struct { + fn run(conn: *Conn, wio: std.Io) error{Canceled}!void { + conn.mu.lockUncancelable(wio); + defer conn.mu.unlock(wio); + while (true) _ = try timedWait(conn, wio, nowMs() + 60_000); + } + }; + var future = try io.concurrent(Waiter.run, .{ &c, io }); + try std.testing.expectError(error.Canceled, future.cancel(io)); +} + test "callstack cross-crate paths remain resolvable without absolute path clutter" { var arena_state = std.heap.ArenaAllocator.init(std.testing.allocator); defer arena_state.deinit(); @@ -2584,37 +2882,38 @@ test "a language with no server to start says so, not that it did not answer" { defer arena.deinit(); var buf: [256]u8 = undefined; var out: std.Io.Writer = .fixed(&buf); - try query(std.testing.allocator, arena.allocator(), .{ .kind = .hover, .path = "/x/a.py", .source = "", .offset = 0 }, &out); + try query(std.testing.io, std.testing.allocator, arena.allocator(), .{ .kind = .hover, .path = "/x/a.py", .source = "", .offset = 0 }, &out); try std.testing.expectEqualStrings("@none language server for .py not found\n", out.buffered()); } test "Lspinfo before any query says a server is not started yet, not off" { var out: std.Io.Writer.Allocating = .init(std.testing.allocator); defer out.deinit(); - try status(.{ .kind = .status, .path = "", .source = "", .offset = 0 }, &out.writer); + try status(std.testing.io, .{ .kind = .status, .path = "", .source = "", .offset = 0 }, &out.writer); try std.testing.expect(std.mem.indexOf(u8, out.written(), "not started yet (its first query starts it)") != null); } test "a server that failed a moment ago is said to have failed, and when it is tried again, not to be missing" { if (getenv("PARDES_LSP_ZIG") != null) return error.SkipZigTest; + const io = std.testing.io; const si = specFor("/x/a.zig").?; const c = &conns[si]; - c.lock(); + c.mu.lockUncancelable(io); const was = .{ c.state, c.retry_after_ms }; c.state = .dead; c.retry_after_ms = nowMs() + 4500; - c.unlock(); + c.mu.unlock(io); defer { - c.lock(); + c.mu.lockUncancelable(io); c.state = was[0]; c.retry_after_ms = was[1]; - c.unlock(); + c.mu.unlock(io); } var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); var out: std.Io.Writer.Allocating = .init(std.testing.allocator); defer out.deinit(); - try query(std.testing.allocator, arena.allocator(), .{ .kind = .hover, .path = "/x/a.zig", .source = "", .offset = 0 }, &out.writer); + try query(std.testing.io, std.testing.allocator, arena.allocator(), .{ .kind = .hover, .path = "/x/a.zig", .source = "", .offset = 0 }, &out.writer); try std.testing.expect(std.mem.indexOf(u8, out.written(), "zls failed recently; retry in 5s") != null); } diff --git a/src/macos.zig b/src/macos.zig index 50fa04fa..5d3dfec4 100644 --- a/src/macos.zig +++ b/src/macos.zig @@ -44,7 +44,11 @@ fn hostWatchFile(pane: u8, generation: u32, path: ?[*]const u8, path_len: usize) const TIOCSWINSZ: c_int = @bitCast(@as(u32, if (@hasDecl(posix.T, "IOCSWINSZ")) posix.T.IOCSWINSZ else 0x80087467)); -pub const std_options: std.Options = .{ .logFn = logFn }; +// Every thread's alternate signal stack is a threadlocal that libc zeroes as +// the thread starts, so std's 256 KiB made each 9P connection's thread cost +// that much resident. A stack-overflow segfault on a thread still printed its +// whole symbolized trace from 32 KiB up (16 KiB double-faulted): 64 KiB. +pub const std_options: std.Options = .{ .logFn = logFn, .signal_stack_size = 64 * 1024 }; fn logFn( comptime level: std.log.Level, @@ -2419,7 +2423,7 @@ fn pipeWorker(st: *State, job: *selection_pipe.Job) anyerror!void { } fn lspWorker(st: *State, job: *host_io.Lsp.Job) anyerror!void { - host_io.Lsp.work(st.gpa, job, st, deliverLspRows); + host_io.Lsp.work(st.io, st.gpa, job, st, deliverLspRows); } fn deliverLspRows(ctx: ?*anyopaque, id: u32, rows: ?[]u8) void { @@ -2582,12 +2586,11 @@ fn reap(st: *State, pane: u8) void { } fn readPty(st: *State, io: std.Io, pty: std.Io.File, id: u8, gen: u32) anyerror!void { - var read_buf: [0x10000]u8 = undefined; - var reader = pty.readerStreaming(io, &read_buf); + // One buffer, read into directly: a Reader around it only added a second + // (and in a safe build, filled both as `undefined` on every start). + var buf: [0x10000]u8 = undefined; while (true) { - var buf: [0x10000]u8 = undefined; - var vec = [_][]u8{&buf}; - const n = reader.interface.readVec(&vec) catch break; + const n = pty.readStreaming(io, &.{&buf}) catch break; if (n == 0) break; const bytes = st.gpa.dupe(u8, buf[0..n]) catch break; st.inbox.pushOutput(st.gpa, io, .{ .pane = id, .gen = gen, .bytes = bytes }) catch break; diff --git a/src/main.zig b/src/main.zig index 64614daa..ea2db917 100644 --- a/src/main.zig +++ b/src/main.zig @@ -16,7 +16,11 @@ pub const std_options_debug_io: std.Io = if (is_emscripten) else std_options_debug_threaded_io.?.io(); -pub const std_options: std.Options = .{ .logFn = logFn }; +// Every thread's alternate signal stack is a threadlocal that libc zeroes as +// the thread starts, so std's 256 KiB made each 9P connection's thread cost +// that much resident. A stack-overflow segfault on a thread still printed its +// whole symbolized trace from 32 KiB up (16 KiB double-faulted): 64 KiB. +pub const std_options: std.Options = .{ .logFn = logFn, .signal_stack_size = 64 * 1024 }; fn logFn( comptime level: std.log.Level, @@ -328,7 +332,11 @@ fn webMain() !void { // malloc otherwise. Kept on purpose: glibc's malloc was measured at least as // fast as std.heap.smp_allocator on every scenario and faster on MuPDF's // page-sized rasters. macos.zig makes the same choice for itself. -fn nativeMain(init: std.process.Init) !void { +fn nativeMain(std_init: std.process.Init) !void { + var init = std_init; + // Zig 0.17's std.start gives a safe build a SafeAllocator over the page + // allocator, not libc's malloc: hand the shells malloc, as above. + if (builtin.mode != .debug) init.gpa = std.heap.c_allocator; var stderr_probe: [64]u8 = undefined; _ = std.debug.lockStderr(&stderr_probe); std.debug.unlockStderr(); diff --git a/src/memory.zig b/src/memory.zig index 48403c55..853d641a 100644 --- a/src/memory.zig +++ b/src/memory.zig @@ -16,6 +16,10 @@ pub const limits = struct { pub const pending_write_cap: usize = if (board) 0 else 4 * MiB; pub const wrap_rows = if (board) 128 else 256; pub const undo_max = if (board) 16 else 256; + /// The text a file's undo and redo steps hold together, kept by dropping + /// the oldest; the newest step always stays, so the bound per file is + /// this plus one snapshot (at most max_file_bytes). + pub const undo_bytes: usize = if (board) 64 * KiB else 64 * MiB; pub const message_log = if (board) 16 else 128; /// ranges one selection holds; helix has no limit, and a `%s` over a /// file's words wants hundreds @@ -23,6 +27,19 @@ pub const limits = struct { /// /log's ring, which records whether or not anyone reads it. pub const log_bytes: usize = if (board) 4 * KiB else 64 * KiB; pub const max_tag_tail: usize = if (board) 512 else 4096; + /// Pixels one image may decode to (RGBA, 4 bytes each): checked from its + /// header before stb_image allocates them. + pub const image_decode_pixels: usize = if (board) 0 else 64 * 1024 * 1024; + /// What the core's per-step scratch arena and the frame arena keep + /// between steps: a step that needed more (a huge Callstack's sources) + /// gives the rest back instead of holding its high-water mark. + pub const scratch_kept: usize = if (board) 4 * KiB else 16 * MiB; + pub const frame_kept: usize = if (board) 4 * KiB else 16 * MiB; + /// The other open files one language query copies for its server, in + /// bytes: past it the rest are left out (the asked file always goes). + pub const lsp_snapshot_bytes: usize = if (board) 0 else 64 * MiB; + /// One paste, from any shell: past it, the rest is left out and said. + pub const max_paste_bytes: usize = if (board) 16 * KiB else 4 * MiB; pub const host_path_cap: usize = if (board) 0 else 4095; /// -Dembed-sources: the editor serves its own sources under /src. pub const embedded_sources = config.embed_sources; @@ -34,6 +51,10 @@ pub const limits = struct { pub const tree_sitter = if (board) 0 else if (reduced_target) 4 * MiB else 16 * MiB; pub const image = if (board) 0 else if (reduced_target) 64 * KiB else 32 * MiB; pub const pdf = if (board) 0 else if (reduced_target or !config.mupdf) 64 * KiB else 64 * MiB; + /// One language query's working memory (its parsed answers, the rows + /// it writes): a hard bound, nothing spills. Callstack walks until it + /// would leave less than its rendering reserve. + pub const lsp_query = if (board or reduced_target) 0 else 256 * MiB; }; }; @@ -96,6 +117,82 @@ var frame_buffer: [limits.arena.frame]u8 = @splat(0); var tree_sitter_buffer: [limits.arena.tree_sitter]u8 = @splat(0); var image_buffer: [limits.arena.image]u8 = @splat(0); var pdf_buffer: [limits.arena.pdf]u8 = @splat(0); +var lsp_query_buffer: [limits.arena.lsp_query]u8 align(std.heap.page_size_max) = @splat(0); +var lsp_query_busy: std.atomic.Value(bool) = .init(false); +/// What a query may leave resident of the region (giveLspQuery). +const lsp_query_kept = 8 * MiB; + +/// The language query region, for the one query running; null while another +/// holds it (the shells cancel and join a query before starting the next). +pub fn takeLspQuery() ?[]u8 { + if (lsp_query_busy.swap(true, .acquire)) return null; + return &lsp_query_buffer; +} + +/// Gives the region back after a query. A big one's touched pages go back to +/// the system (zeroed again on their next touch, as .bss is), so one huge +/// answer does not stay resident; a small one's stay, for the next query to +/// reuse without faulting them in again. +pub fn giveLspQuery(used: usize) void { + if (comptime builtin.os.tag == .linux) if (used > lsp_query_kept) { + const page = std.heap.pageSize(); + const len = std.mem.alignForward(usize, @min(used, lsp_query_buffer.len), page); + std.posix.madvise(@as([*]align(std.heap.page_size_min) u8, @ptrCast(&lsp_query_buffer)), len, std.posix.MADV.DONTNEED) catch {}; + }; + lsp_query_busy.store(false, .release); +} + +/// `n` bytes as they come: `Allocator.alloc` fills a safe build's new memory +/// as `undefined` (0xAA), which writes, and makes resident, all of a buffer +/// sized for the worst case: every language query wrote its 51 MB answer +/// room, every search its 16 MiB. For buffers written before they are read. +pub fn unfilled(allocator: Allocator, n: usize) Allocator.Error![]u8 { + if (n == 0) return allocator.alloc(u8, 0); + const ptr = allocator.rawAlloc(n, .of(u8), @returnAddress()) orelse return error.OutOfMemory; + return ptr[0..n]; +} + +/// `bytes` copied into memory `unfilled` takes: a text-sized copy (an undo +/// snapshot) is written once, not filled first. +pub fn dupeUnfilled(allocator: Allocator, bytes: []const u8) Allocator.Error![]u8 { + const out = try unfilled(allocator, bytes.len); + @memcpy(out, bytes); + return out; +} + +/// Frees without the safe-build overwrite `free` does, for text-sized buffers +/// freed on every edit. +pub fn freeUnfilled(allocator: Allocator, bytes: []const u8) void { + if (bytes.len > 0) allocator.rawFree(@constCast(bytes), .of(u8), @returnAddress()); +} + +/// Gives libc's free memory back to the system, inside its heap too: after a +/// big result is formatted or a big pane torn down, glibc keeps what was freed +/// (a V8 Callstack page left 21 MiB free in the heap; this returned 20). +/// Walks the heap, so for rare, large events only. +pub fn trim() void { + if (comptime builtin.os.tag != .linux or builtin.abi.isMusl() or !builtin.link_libc) return; + const glibc = struct { + extern "c" fn malloc_trim(pad: usize) c_int; + }; + _ = glibc.malloc_trim(0); +} + +/// Marks a big region that is touched sparsely as not for huge pages. With +/// transparent huge pages on `always` (a common default), touching one byte +/// of a 2 MiB-aligned stretch makes all 2 MiB resident: the 9P runner's +/// slots and these arenas' first pages held 45 MiB of an idle session's +/// 71 MiB that way. Only the region given, never the whole process (a prctl +/// would reach every shell and server pardes starts). +pub fn sparse(region: []u8) void { + if (comptime builtin.os.tag != .linux) return; + const page = std.heap.pageSize(); + const start = std.mem.alignForward(usize, @intFromPtr(region.ptr), page); + const end = std.mem.alignBackward(usize, @intFromPtr(region.ptr) + region.len, page); + if (end <= start) return; + std.posix.madvise(@ptrFromInt(start), end - start, std.posix.MADV.NOHUGEPAGE) catch {}; +} + var pardes_fallback: SharedStackFallback = .{ .buffer = &pardes_buffer }; var frame_fallback: SharedStackFallback = .{ .buffer = &frame_buffer }; var tree_sitter_fallback: SharedStackFallback = .{ .buffer = &tree_sitter_buffer }; @@ -118,6 +215,7 @@ pub fn init(fallback: Allocator) Allocators { // a Debug build `fallback` is std.process.Init's gpa, a DebugAllocator // already, and a second one checked every block twice. const backing = if (debug_enabled) std.heap.page_allocator else fallback; + inline for (.{ &pardes_buffer, &frame_buffer, &tree_sitter_buffer, &image_buffer, &pdf_buffer, &lsp_query_buffer }) |buffer| sparse(buffer); pardes_fallback.fallback_allocator = backing; frame_fallback.fallback_allocator = backing; tree_sitter_fallback.fallback_allocator = backing; diff --git a/src/mini.zig b/src/mini.zig index ce291c99..8d8ea0b2 100644 --- a/src/mini.zig +++ b/src/mini.zig @@ -313,7 +313,21 @@ test "Mini publishes only complete snapshots and content replacement drops metad const dir = dir_buf[0..try tmp.dir.realPath(std.testing.io, &dir_buf)]; var path_buf: [4096]u8 = undefined; const path = try std.fmt.bufPrint(&path_buf, "{s}/mini.txt", .{dir}); - try std.testing.checkAllAllocationFailures(std.testing.allocator, Case.run, .{path}); + // Growth that never resizes in place: whether a resize fits where it is + // depends on the heap's layout, and with it how many allocations a run + // makes, which checkAllAllocationFailures needs to be the same each run. + const NoResize = struct { + fn resize(_: *anyopaque, _: []u8, _: std.mem.Alignment, _: usize, _: usize) bool { + return false; + } + fn remap(_: *anyopaque, _: []u8, _: std.mem.Alignment, _: usize, _: usize) ?[*]u8 { + return null; + } + }; + const testing_vtable = std.testing.allocator.vtable; + const no_resize_vtable: std.mem.Allocator.VTable = .{ .alloc = testing_vtable.alloc, .resize = NoResize.resize, .remap = NoResize.remap, .free = testing_vtable.free }; + const backing: std.mem.Allocator = .{ .ptr = std.testing.allocator.ptr, .vtable = &no_resize_vtable }; + try std.testing.checkAllAllocationFailures(backing, Case.run, .{path}); } test "Mini refuses a file that is no UTF-8, before its decoding could panic" { diff --git a/src/modal.zig b/src/modal.zig index 82f1a3a0..ad06833f 100644 --- a/src/modal.zig +++ b/src/modal.zig @@ -989,13 +989,77 @@ pub fn lineCount(content: []const u8) usize { // byte offset of the start of line `row` (0-based). row may == lineCount() // (== content.len, the end). pub fn lineStartOffset(content: []const u8, row: usize) usize { + // Newlines counted 32 bytes at a time, the last block byte by byte: one + // indexOfScalarPos per line was a call per line, 7000 of them to find + // line 7000 of a source, for every cursor move, selection and language + // request there. + const block = 32; + const Block = @Vector(block, u8); var off: usize = 0; - var r: usize = 0; - while (r < row) : (r += 1) { - const nl = std.mem.indexOfScalarPos(u8, content, off, '\n') orelse return content.len; - off = nl + 1; + var left = row; + if (left == 0) return 0; + while (off + block <= content.len) : (off += block) { + const chunk: Block = content[off..][0..block].*; + const found: usize = @popCount(@as(u32, @bitCast(chunk == @as(Block, @splat('\n'))))); + if (found >= left) break; + left -= found; + } + while (off < content.len) : (off += 1) if (content[off] == '\n') { + left -= 1; + if (left == 0) return off + 1; + }; + return content.len; +} + +test "lineStartOffset by blocks and offsetAt agree with a line at a time" { + var prng: std.Random.DefaultPrng = .init(0x9e3779b9); + const random = prng.random(); + var buf: [300]u8 = undefined; + for (0..2000) |_| { + const len = random.uintLessThan(usize, buf.len); + for (buf[0..len]) |*b| b.* = if (random.uintLessThan(u8, 4) == 0) '\n' else 'a'; + const text = buf[0..len]; + for (0..len + 3) |row| { + var off: usize = 0; + var r: usize = 0; + const want = while (r < row) : (r += 1) { + off = (std.mem.indexOfScalarPos(u8, text, off, '\n') orelse break text.len) + 1; + } else off; + try std.testing.expectEqual(want, lineStartOffset(text, row)); + const cursor: Cursor = .{ .row = row, .col = random.uintLessThan(usize, 6) }; + const clamped = lineStartOffset(text, @min(row, cursorLineCount(text) - 1)); + const end = std.mem.indexOfScalarPos(u8, text, clamped, '\n') orelse text.len; + try std.testing.expectEqual(clamped + graphemeStart(text[clamped..end], @min(cursor.col, end - clamped)), offsetAt(text, cursor)); + // lineSpan, as it was: a line at a time, null past the last. + const span: ?LineSpan = span: { + if (text.len == 0) break :span null; + var start: usize = 0; + for (0..row) |_| start = (std.mem.indexOfScalarPos(u8, text, start, '\n') orelse break :span null) + 1; + if (start == text.len and !(row == 0 or text[text.len - 1] == '\n')) break :span null; + break :span .{ .start = start, .end = std.mem.indexOfScalarPos(u8, text, start, '\n') orelse text.len }; + }; + try std.testing.expectEqual(span, lineSpan(text, row)); + // rangeBytes, as it was: each line looked for from the top, twice. + const other: Cursor = .{ .row = random.uintLessThan(usize, len + 3), .col = random.uintLessThan(usize, 6) }; + var lo = cursor; + var hi = other; + if (hi.row < lo.row or (hi.row == lo.row and hi.col < lo.col)) std.mem.swap(Cursor, &lo, &hi); + const lo_line = lineSlice(text, lo.row); + const hi_line = lineSlice(text, hi.row); + const s0 = lineStartOffset(text, lo.row) + graphemeStart(lo_line, @min(lo.col, lo_line.len)); + var e0 = lineStartOffset(text, hi.row) + graphemeStart(hi_line, @min(hi.col, hi_line.len)); + if (e0 < text.len) e0 = nextGrapheme(text, e0); + const got = rangeBytes(text, cursor, other); + try std.testing.expectEqual(s0, got.s); + try std.testing.expectEqual(@max(s0, e0), got.e); + } } - return off; +} + +/// The line that starts at `start` (lineSlice's, once its start is known). +fn lineFrom(content: []const u8, start: usize) []const u8 { + if (start >= content.len) return ""; + return content[start .. std.mem.indexOfScalarPos(u8, content, start, '\n') orelse content.len]; } // the text of line `row` (no terminator), a slice into `content`. @@ -1011,23 +1075,29 @@ pub fn lineSlice(content: []const u8, row: usize) []const u8 { /// A bounded scan of one line; null if the row does not exist. pub const LineSpan = struct { start: usize, end: usize }; +/// Just past the `row`-th newline; null if the text has fewer. +pub fn lineStartExact(content: []const u8, row: usize) ?usize { + const at = lineStartOffset(content, row); + // At the end: either the row-th newline was the last byte, or there were fewer. + if (at == content.len and std.mem.count(u8, content, "\n") < row) return null; + return at; +} + pub fn lineSpan(content: []const u8, row: usize) ?LineSpan { // Match lineCount: empty content has no lines; a trailing newline adds one. if (content.len == 0) return null; - var start: usize = 0; - var r: usize = 0; - while (r < row) : (r += 1) { - const nl = std.mem.indexOfScalarPos(u8, content, start, '\n') orelse return null; - start = nl + 1; - } - if (start > content.len) return null; + const start = lineStartExact(content, row) orelse return null; if (start == content.len and !(row == 0 or content.len == 0 or content[content.len - 1] == '\n')) return null; const end = std.mem.indexOfScalarPos(u8, content, start, '\n') orelse content.len; return .{ .start = start, .end = end }; } -fn spliceAlloc(alloc: std.mem.Allocator, content: []const u8, start: usize, end: usize, replacement: []const u8) ![]u8 { - const out = try alloc.alloc(u8, content.len - (end - start) + replacement.len); +pub fn spliceAlloc(alloc: std.mem.Allocator, content: []const u8, start: usize, end: usize, replacement: []const u8) ![]u8 { + // Raw: every byte is written below, where `alloc` fills a safe build's + // new memory first, a whole file's worth per keystroke. + const len = content.len - (end - start) + replacement.len; + if (len == 0) return alloc.alloc(u8, 0); + const out = (alloc.rawAlloc(len, .of(u8), @returnAddress()) orelse return error.OutOfMemory)[0..len]; @memcpy(out[0..start], content[0..start]); @memcpy(out[start..][0..replacement.len], replacement); @memcpy(out[start + replacement.len ..], content[end..]); @@ -1036,13 +1106,18 @@ fn spliceAlloc(alloc: std.mem.Allocator, content: []const u8, start: usize, end: /// insert `text` at (row, col). col is clamped to the line length. pub fn insertAt(alloc: std.mem.Allocator, content: []const u8, c: Cursor, text: []const u8) ![]u8 { + const off = insertOffset(content, c); + return spliceAlloc(alloc, content, off, off, text); +} + +/// Where insertAt puts its text: col clamped to the line, no further. +pub fn insertOffset(content: []const u8, c: Cursor) usize { // Only an out-of-range row requires counting the whole document. const span = lineSpan(content, c.row) orelse blk: { const last = lineCount(content) -| 1; break :blk lineSpan(content, last) orelse LineSpan{ .start = content.len, .end = content.len }; }; - const off = span.start + @min(c.col, span.end - span.start); - return spliceAlloc(alloc, content, off, off, text); + return span.start + @min(c.col, span.end - span.start); } // delete the character at (row, col). no-op if col is past the line end. @@ -1105,12 +1180,15 @@ fn rangeBytes(content: []const u8, a: Cursor, b: Cursor) struct { s: usize, e: u lo = b; hi = a; } - const lo_line = lineSlice(content, lo.row); - const hi_line = lineSlice(content, hi.row); + // Each line found once, the second from the first. + const lo_start = lineStartOffset(content, lo.row); + const hi_start = if (hi.row == lo.row) lo_start else lo_start + lineStartOffset(content[lo_start..], hi.row - lo.row); + const lo_line = lineFrom(content, lo_start); + const hi_line = lineFrom(content, hi_start); const lo_col = graphemeStart(lo_line, @min(lo.col, lo_line.len)); const hi_col = graphemeStart(hi_line, @min(hi.col, hi_line.len)); - const s = lineStartOffset(content, lo.row) + lo_col; - var e = lineStartOffset(content, hi.row) + hi_col; + const s = lo_start + lo_col; + var e = hi_start + hi_col; if (e < content.len) e = nextGrapheme(content, e); // include the grapheme under the head return .{ .s = s, .e = @max(s, e) }; } @@ -1138,12 +1216,12 @@ pub fn clearLine(alloc: std.mem.Allocator, content: []const u8, row: usize) ![]u // paste `text` as a new line BELOW `row`. Multiline `text` becomes several lines. pub fn pasteLineBelow(alloc: std.mem.Allocator, content: []const u8, row: usize, text: []const u8) ![]u8 { - const n = lineCount(content); - const off = if (row + 1 < n) lineStartOffset(content, row + 1) else content.len; + // The next line's start, or the end: no whole-text line count to know which. + const off = lineStartExact(content, row + 1) orelse content.len; const suffix_newline = off < content.len or (content.len > 0 and content[content.len - 1] == '\n'); const prefix_newline = !suffix_newline and off > 0; const extra: usize = @intFromBool(suffix_newline or prefix_newline); - const out = try alloc.alloc(u8, content.len + text.len + extra); + const out = try @import("memory.zig").unfilled(alloc, content.len + text.len + extra); var write: usize = 0; @memcpy(out[write..][0..off], content[0..off]); write += off; @@ -1393,8 +1471,10 @@ pub fn positionAt(text: []const u8, off: usize) Cursor { /// (row, col) -> clamped gap offset; col == line length lands ON the '\n' pub fn offsetAt(text: []const u8, c: Cursor) usize { - const row = @min(c.row, cursorLineCount(text) - 1); - const s = lineStartOffset(text, row); + // A row past the last is the last: found back from the end, where + // counting every line of the text to clamp it cost a pass over all of it. + var s = lineStartOffset(text, c.row); + if (s == text.len) s = if (std.mem.lastIndexOfScalar(u8, text, '\n')) |nl| nl + 1 else 0; // lineEndOffset(text, row) inlined: it starts by walking to `row` again, // and we are already standing there const e = std.mem.indexOfScalarPos(u8, text, s, '\n') orelse text.len; @@ -1783,15 +1863,6 @@ test "selection textobjects distinguish words and paragraph boundaries" { try std.testing.expectEqual(@as(usize, 7), r.head); } -/// leading-whitespace visual width (tab -> next multiple of INDENT_W) -pub fn indentWidth(line: []const u8) usize { - var w: usize = 0; - for (line) |ch| { - if (ch == ' ') w += 1 else if (ch == '\t') w = (w / INDENT_W + 1) * INDENT_W else break; - } - return w; -} - /// A line's indent as it is, tabs and spaces both, as acme's autoindent /// copies it: what o/O/insert-newline give the new line. pub fn indentText(line: []const u8) []const u8 { diff --git a/src/mouse.zig b/src/mouse.zig index 76c65a16..eeb987b2 100644 --- a/src/mouse.zig +++ b/src/mouse.zig @@ -1114,7 +1114,9 @@ fn mirrorTtySelection(p: *Pardes, pane: *Pane) void { /// second selects what edit.doubleClickRange says, from the caret the click /// left. A plain click still puts the caret on its cell. fn doubleClick(p: *Pardes, id: usize, t: *panes.Text, on_tag: bool, row: i32, col: i32) void { - const now = clockMs(); + // The host's clock, which every host keeps (a libc clock is no clock on + // the web or the board) and a test's virtual one replaces. + const now: i64 = @intCast(p.now_ns / std.time.ns_per_ms); const prev = p.last_click; p.last_click = .{ .id = id, .on_tag = on_tag, .row = row, .col = col, .ms = now }; const c = prev orelse return; @@ -1160,13 +1162,6 @@ fn sweepPoint(p: *Pardes, pane: *panes.Pane, r: i32, c: i32) ?modal.Cursor { return .{ .row = @intCast(w.line), .col = @intCast(@max(0, col)) }; } -fn clockMs() i64 { - if (comptime !pardes.hosted) return 0; - var ts: std.c.timespec = undefined; - if (std.c.clock_gettime(.MONOTONIC, &ts) != 0) return 0; - return @as(i64, ts.sec) * 1000 + @divTrunc(@as(i64, ts.nsec), 1_000_000); -} - fn dragRelease(p: *Pardes, button: Mouse.Button) void { // A different button's release belongs to a chord or another host // stream; only the button which began this gesture may end it. diff --git a/src/ninep/addr.zig b/src/ninep/addr.zig index 86d2d49e..7753c94d 100644 --- a/src/ninep/addr.zig +++ b/src/ninep/addr.zig @@ -239,21 +239,15 @@ pub const Addr = struct { k = @intFromBool(a.text[p] == '\n'); p += 1; } - while (k < n) { - if (p >= nc) { + if (k < n) { + p = nthLineAfter(a.text, p, n - k) orelse { a.err = e_range; return null; - } - if (a.text[p] == '\n') k += 1; - p += 1; + }; } p1 = p; } - while (p < nc) { - const ch = a.text[p]; - p += 1; - if (ch == '\n') break; - } + p = if (std.mem.indexOfScalarPos(u8, a.text, p, '\n')) |nl| nl + 1 else nc; return .{ .q0 = clip(p1), .q1 = clip(p) }; } @@ -678,3 +672,31 @@ test "a line just past the last newline is an empty one there, as sam's lineaddr try testing.expectEqual(Status.ok, th.wr(p, Node.of(th.serialOf(p), .ctl), "Edit 1i/header/\n").reply.status); try testing.expectEqualStrings("header", p.panes[0].?.file.?.content); } + +/// Just past the `count`-th newline at or after `from`; null if the text has +/// fewer. Counted a block at a time (modal.lineStartOffset), where a byte at +/// a time made `echo 7000 > addr` walk every byte above line 7000. +fn nthLineAfter(text: []const u8, from: usize, count: usize) ?usize { + return from + (modal.lineStartExact(text[from..], count) orelse return null); +} + +test "nthLineAfter finds what a byte at a time finds" { + var prng: std.Random.DefaultPrng = .init(0x51ab); + const random = prng.random(); + var buf: [200]u8 = undefined; + for (0..3000) |_| { + const len = random.uintLessThan(usize, buf.len); + for (buf[0..len]) |*b| b.* = if (random.uintLessThan(u8, 3) == 0) '\n' else 'x'; + const text = buf[0..len]; + const from = random.uintLessThan(usize, len + 1); + const count = 1 + random.uintLessThan(usize, 40); + var p = from; + var k: usize = 0; + const want: ?usize = while (k < count) { + if (p >= text.len) break null; + if (text[p] == '\n') k += 1; + p += 1; + } else p; + try std.testing.expectEqual(want, nthLineAfter(text, from, count)); + } +} diff --git a/src/ninep/ctl.zig b/src/ninep/ctl.zig index 733d9c8e..480c1dc7 100644 --- a/src/ninep/ctl.zig +++ b/src/ninep/ctl.zig @@ -4684,6 +4684,10 @@ test "what a /pager open writes after its directory is the text: plain in the bo // An edit drops the colours (pager.zig's ponytail). try testing.expectEqual(Status.ok, call(p, .{ .tag = 6, .op = .write, .node = Node.of(pn.serial, .body), .data = "more\n" }).reply.status); try testing.expectEqual(@as(usize, 0), f.pager_spans.len); + // Undone back to the paged text: its colours come back with it. + panes.File.undo(p, pn); + try testing.expectEqualStrings("red /a\nNext\n", f.content); + try testing.expectEqual(@as(usize, 2), f.pager_spans.len); // PagerColor off, and an open closed with no read: the same text, plain, // in at the editor's next step (a close through a mount never waits). diff --git a/src/panes.zig b/src/panes.zig index 7ce85419..b535993e 100644 --- a/src/panes.zig +++ b/src/panes.zig @@ -133,6 +133,9 @@ pub const Pane = struct { /// Its pty is still open: a job the command left in the background may /// be printing to it, and a reuse would hang it up. command_pty: bool = false, + /// Its child pages as a terminal's does (`pardes -`, a +Pager), not + /// through cat: Look's `git show` of a commit (look.lookCommit). + command_pages: bool = false, /// The shell a terminal runs when it is not the configured one (`Tty /// fish`), owned. shell: ?[]u8 = null, diff --git a/src/pardes.zig b/src/pardes.zig index aaa9d177..c2a3ba21 100644 --- a/src/pardes.zig +++ b/src/pardes.zig @@ -478,6 +478,7 @@ pub const Loc = struct { }; test { + _ = @import("modal.zig"); _ = @import("tagline.zig"); _ = @import("hunks.zig"); _ = @import("builtins_json.zig"); @@ -4645,6 +4646,10 @@ pub const Pardes = struct { /// The last query asked (Lspwhy explains it): its kind, argument, and /// the pane and place it was asked at. lsp_last: ?struct { kind: lsp.Kind, arg: Effect.Buf(128), serial: u32, offset: u32 } = null, + /// The Callstack page `+Callstack` shows stopped short: its file's pane, + /// the place it was asked at, and the direct caller the next page starts + /// at. Callstack run in `+Callstack` asks for it. + callstack_more: ?struct { serial: u32, offset: u32, next: Effect.Buf(128) } = null, lsp_wait: ?struct { id: u32, kind: lsp.Kind, @@ -5168,10 +5173,13 @@ pub const Pardes = struct { } fn reapPanes(p: *Pardes) void { + var reaped = false; for (&p.retired_panes.previous) |*slot| if (slot.*) |pane| { p.teardownPane(pane); slot.* = null; + reaped = true; }; + if (reaped) memory.trim(); p.retired_panes.previous = p.retired_panes.current; p.retired_panes.current = @splat(null); } @@ -5544,7 +5552,7 @@ pub const Pardes = struct { p.invalidate(); look.invalidateLookHover(p, id); const applied = p.applyWatchedFileChanged(id, bytes); - _ = p.scratch.reset(.retain_capacity); + _ = p.scratch.reset(.{ .retain_with_limit = memory.limits.scratch_kept }); p.sync(); return applied; } @@ -6000,7 +6008,7 @@ pub const Pardes = struct { // A pty streaming into an unwatched session spent most of its time // drawing what nobody looked at. if (v.watched) |f| if (!f(h.ctx)) return; - _ = p.frame_arena.reset(.retain_capacity); + _ = p.frame_arena.reset(.{ .retain_with_limit = memory.limits.frame_kept }); const surface = try p.render(p.frame_arena.allocator()); if (v.present) |f| f(h.ctx, surface); // Settled before post_present: a shell ticks there, and what that @@ -6202,7 +6210,7 @@ pub const Pardes = struct { look.advanceLookHover(p); }, } - _ = p.scratch.reset(.retain_capacity); + _ = p.scratch.reset(.{ .retain_with_limit = memory.limits.scratch_kept }); p.sync(); p.fsReport(); ctlfs.events.announce(p); @@ -6777,6 +6785,19 @@ pub const Pardes = struct { pub const SavePath = Effect.Buf(effect_path_cap); pub fn lspRequest(p: *Pardes, asked: usize, kind: lsp.Kind, arg_in: []const u8) void { + p.lspRequestAt(asked, kind, arg_in, null); + } + + /// The next page of the Callstack `+Callstack` shows, asked where the + /// first was. + pub fn callstackMore(p: *Pardes, from: usize) void { + const more = p.callstack_more orelse return p.reportFailure(from, "Callstack: this is the last page; run it in a file to start again"); + const id = p.paneBySerial(more.serial) orelse return p.reportFailure(from, "Callstack: the file it was asked in is closed"); + p.lspRequestAt(id, .callstack, more.next.slice(), more.offset); + } + + /// `lspRequest` at byte `at` of the pane's file rather than its cursor. + fn lspRequestAt(p: *Pardes, asked: usize, kind: lsp.Kind, arg_in: []const u8, at_byte: ?u32) void { var arg = arg_in; if (!normal.multiOnce(p)) return; // one question per keystroke, from the primary // Lspwhy narrates the last query where it was asked -- that pane's @@ -6800,7 +6821,7 @@ pub const Pardes = struct { (!zig_buf and std.mem.indexOfAny(u8, arg, " \t\r\n\"\\") != null)) return p.reportFailure(id, std.fmt.bufPrint(&said, "Rename: {s} is no name", .{arg}) catch "Rename: no name"); } - var off = if (pane.file) |f| modal.offsetAt(f.content, .{ + var off = if (at_byte) |at| @min(at, if (pane.file) |f| f.content.len else 0) else if (pane.file) |f| modal.offsetAt(f.content, .{ .row = @intCast(@max(0, pane.body.cur_row)), .col = @intCast(@max(0, pane.body.cur_col)), }) else 0; @@ -7089,13 +7110,24 @@ pub const Pardes = struct { } const dir = if (pane.file) |f| panes.Dir.ofName(f.path) else pane.cwdSlice(); - const content = p.gpa.dupe(u8, rows) catch return; + // A Callstack page that stops short ends in its marker: said as a + // row naming the word that shows the next page, and kept. + if (w.kind == .callstack) p.callstack_more = null; + const marker = if (w.kind == .callstack) std.mem.lastIndexOf(u8, rows, lsp.callstack_more) else null; + const content = if (marker) |at| more: { + var fields = std.mem.tokenizeAny(u8, rows[at + lsp.callstack_more.len ..], " \n"); + const next = fields.next() orelse ""; + const shown = fields.next() orelse "?"; + if (p.lsp_last) |last| p.callstack_more = .{ .serial = w.serial, .offset = last.offset, .next = .from(next) }; + break :more std.fmt.allocPrint(p.gpa, "{s}… {s} callers on this page, more after them: Callstack here shows the next page\n", .{ rows[0..at], shown }) catch return; + } else p.gpa.dupe(u8, rows) catch return; const serial_before = p.next_serial; p.fs.refilled = null; panes.Output.fillResults(p, w.pane, dir, from, w.arg.slice(), content, null) catch |err| { p.reportError(w.pane, "language response", err); return; }; + if (rows.len >= 1 << 20) memory.trim(); // What a 9P write that asked it answers: the pane the rows are in // (a multi-file Rename's preview, Lspinfo's +Lsp), made or filled // again, not the pane it was asked from. diff --git a/src/regexp.zig b/src/regexp.zig index 4cbca57e..6d281859 100644 --- a/src/regexp.zig +++ b/src/regexp.zig @@ -283,7 +283,10 @@ pub const Regex = struct { /// its top level and every branch starts with `^`; null when it is not /// one, or no branch does. fn anchoredAlternation(pat: []const u8, out: *[5 * max_ops + 2]u8) error{ Bad, Mixed }!?[]const u8 { - var bars: [16]usize = undefined; + // No rewrite is longer than `out`, so neither is any pattern it + // takes, nor its count of bars. + if (pat.len > out.len) return error.Bad; + var bars: [5 * max_ops + 2]u16 = undefined; var n: usize = 0; var depth: usize = 0; var in_class = false; @@ -301,8 +304,7 @@ pub const Regex = struct { } else if (c == ')') { depth -|= 1; } else if (c == '|' and depth == 0) { - if (n == bars.len) return error.Bad; - bars[n] = i; + bars[n] = @intCast(i); n += 1; } } @@ -518,6 +520,9 @@ test "a ^ after | anchors that branch: ^def|^ finds a line that starts either w try std.testing.expectError(error.Mixed, Regex.compile("^def|x")); var plain = try Regex.compile("a|b"); try std.testing.expectEqual(@as(usize, 7), (try plain.find(text, 0, text.len, text.len)).?.start); + // Seventeen branches and more are a pattern like any other. + var many = try Regex.compile("q|r|s|t|u|v|w|x|y|z|k|l|m|n|o|p|j|b"); + try std.testing.expectEqual(@as(usize, 0), (try many.find(text, 0, text.len, text.len)).?.start); } test "a quoted backslash before n is no newline, and a class ends where mvzr ends it" { diff --git a/src/sam_edit.zig b/src/sam_edit.zig index f6253c89..94b76816 100644 --- a/src/sam_edit.zig +++ b/src/sam_edit.zig @@ -259,6 +259,10 @@ fn fail(why: *Why, comptime fmt: []const u8, args: anytype) Failure { // ---- parsing: edit.c ---- +/// Commands one Edit may nest, loops (`x`, `y`, `g`, `v`, `X`, `Y`) and +/// braces together: each level is a stack frame holding a compiled regex. +pub const max_nest = 32; + const Parser = struct { arena: std.mem.Allocator, src: []const u8, @@ -271,6 +275,9 @@ const Parser = struct { open_text: bool = false, /// The address being read had a `,` after a `"re"`. file_comma: bool = false, + /// Commands open around the one being read, braces and loops alike: + /// each is a frame of `parse` and later of `execIn`, so it is bounded. + depth: u8 = 0, why: *Why, fn getch(ps: *Parser) ?u8 { @@ -460,6 +467,9 @@ const Parser = struct { /// edit.c:467, parsecmd. fn parse(ps: *Parser, nest: u32) Failure!?*Cmd { + if (ps.depth >= max_nest) return fail(ps.why, "commands nested more than {d} deep", .{max_nest}); + ps.depth += 1; + defer ps.depth -= 1; var file_re: ?[]const u8 = null; ps.file_comma = false; const a = try ps.address(&file_re); @@ -1163,6 +1173,11 @@ test "an Edit that fails halfway changes nothing, and says why in acme's words" .{ "e /tmp/x", "can't open /tmp/x: no files here" }, .{ "\"zz\" p", "no file matches \"zz\"" }, .{ ",s/(a)/\\1/", "no \\1: mvzr keeps no submatches" }, + .{ comptime deep: { + var loops: []const u8 = ""; + for (0..40) |_| loops = loops ++ ",x/a/"; + break :deep loops ++ "d"; + }, "commands nested more than 32 deep" }, .{ "1 m 1,2", "move overlaps itself" }, .{ ",x/(^|\\n)foo/d", "bad regular expression: in a pattern with \\n, ^ can only come first and $ only just before a \\n" }, }) |c| { diff --git a/src/selection_pipe.zig b/src/selection_pipe.zig index c3e5e3a6..229b3d67 100644 --- a/src/selection_pipe.zig +++ b/src/selection_pipe.zig @@ -401,7 +401,7 @@ pub fn runIn( }, }; child.stdin = null; // writer_context owns and closes this endpoint - var writer: ?std.Thread = std.Thread.spawn(.{}, writeInput, .{writer_context}) catch { + var writer: ?std.Io.Future(void) = io.concurrent(writeInput, .{writer_context}) catch { writer_context.file.close(io); gpa.free(writer_context.input); gpa.destroy(writer_context); @@ -410,8 +410,9 @@ pub fn runIn( }; // On every early return (a timeout, a ceiling) the answer goes at once: // the group is killed and the shell left to be reaped elsewhere - // (abandon), and the writer, which owns its input, let go. - defer if (writer) |thread| thread.detach(); + // (abandon), and the writer's write canceled (it closes and frees what + // it owns as it ends). + defer if (writer) |*w| w.cancel(io); defer if (child.id != null) abandonNow(io, &child, pid); defer if (token != 0) Running.remove(pid); if (token != 0 and !Running.add(token, pid)) return fail.k(.signal); @@ -443,7 +444,9 @@ pub fn runIn( multi_reader.checkAnyError() catch return fail.k(.io); if (token != 0) Running.remove(pid); const term = child.wait(io) catch return fail.k(.io); - writer.?.join(); + // A cancel of this job reaches the write too: a grandchild that keeps + // stdin open and never reads holds no worker. + writer.?.await(io); writer = null; const stdout = multi_reader.toOwnedSlice(0) catch return fail.k(.io); diff --git a/src/surface.zig b/src/surface.zig index e67cf21c..d6ec016d 100644 --- a/src/surface.zig +++ b/src/surface.zig @@ -533,11 +533,6 @@ pub const Surface = struct { return s.cell_diffs[@as(usize, y) * s.cols + x].changed(); } - pub fn panelCellDiff(s: *const Surface, x: u16, y: u16) PanelCellDiff { - if (!s.hasPanelDiff() or x >= s.cols or y >= s.rows) return .unchanged; - return s.cell_diffs[@as(usize, y) * s.cols + x]; - } - pub fn at(s: *Surface, x: u16, y: u16) *Cell { std.debug.assert(x < s.cols and y < s.rows); return &s.cells[@as(usize, y) * s.cols + x]; diff --git a/src/terminal.zig b/src/terminal.zig index 33ad22e6..05054521 100644 --- a/src/terminal.zig +++ b/src/terminal.zig @@ -1,4 +1,4 @@ -//! A terminal pane: its VT emulator and replay, prompt rows, the edit overlay +//! A terminal pane: its VT emulator, prompt rows, the edit overlay //! typed over shell rows, and the Filter that recolours output to the theme. const panes = @import("panes.zig"); const std = @import("std"); @@ -141,16 +141,9 @@ pub const MarkedHandler = if (enabled) struct { } } else void; -pub const Replay = struct { - bytes: [1024 * 1024]u8 = undefined, - head: usize = 0, - len: usize = 0, -}; - pub const State = struct { vt: VtSlot, stream: StreamSlot, - replay: Replay, reply: [256]u8 = undefined, reply_len: u16 = 0, /// The grid cell the program last heard the mouse at, so motion is @@ -613,8 +606,6 @@ pub fn create(gpa: std.mem.Allocator, cols: u16, rows: u16) !*Pane { .kitty_image_storage_limit = if (images_enabled) TerminalImages.storage_limit else 0, }); state.stream = .init(.{ .handler = .{ .inner = state.vt.vtHandler() }, .allocator = gpa }); - state.replay.head = 0; - state.replay.len = 0; state.reply_len = 0; state.mouse_cell = null; state.images = if (images_enabled) .{} else {}; @@ -678,43 +669,12 @@ pub fn restore(p: *Pardes, src: dump.Pane, live: bool) !*Pane { return pane; } -/// Feed the emulator and retain the bounded suffix a dump can replay. Live -/// output and restoration share this byte path, then apply different views. +/// Feed the emulator. Live output and restoration share this byte path, +/// then apply different views. fn ingest(pane: *Pane, bytes: []const u8) void { - const state = pane.terminal.?; - const replay = &state.replay; - if (bytes.len >= replay.bytes.len) { - const kept = bytes[bytes.len - replay.bytes.len ..]; - @memcpy(&replay.bytes, kept); - replay.head = 0; - replay.len = replay.bytes.len; - } else { - const overflow = bytes.len -| (replay.bytes.len - replay.len); - replay.head = (replay.head + overflow) % replay.bytes.len; - replay.len -= overflow; - const tail = (replay.head + replay.len) % replay.bytes.len; - const first = @min(bytes.len, replay.bytes.len - tail); - @memcpy(replay.bytes[tail..][0..first], bytes[0..first]); - @memcpy(replay.bytes[0 .. bytes.len - first], bytes[first..]); - replay.len += bytes.len; - } - state.stream.nextSlice(bytes); + pane.terminal.?.stream.nextSlice(bytes); } -/// Return the replay ring in chronological order. Wrapped records are copied -/// into `allocator`; contiguous records remain a borrowed slice of the pane. -fn replayBytes(pane: *const Pane, allocator: std.mem.Allocator) ![]const u8 { - const state = pane.terminal orelse return &.{}; - const replay = &state.replay; - if (replay.len == 0) return &.{}; - if (replay.head + replay.len <= replay.bytes.len) - return replay.bytes[replay.head..][0..replay.len]; - const out = try allocator.alloc(u8, replay.len); - const first = replay.bytes.len - replay.head; - @memcpy(out[0..first], replay.bytes[replay.head..]); - @memcpy(out[first..], replay.bytes[0 .. replay.len - first]); - return out; -} test "replay ownership is terminal-only and construction rolls back on allocation failure" { const Case = struct { @@ -753,7 +713,6 @@ test "document construction allocates only the pane and has inert terminal defau try std.testing.expectEqual(GridCursor{}, gridCursor(pane)); try std.testing.expectEqual(Scrollbar{}, scrollbar(pane)); try std.testing.expectEqualStrings("", try screenTextAlloc(pane, allocator)); - try std.testing.expectEqualStrings("", try replayBytes(pane, allocator)); try std.testing.expect(!failing.has_induced_failure); } @@ -779,46 +738,18 @@ test "terminal state keeps parser fragments and sends emulator replies through i } try std.testing.expectEqual(@as(usize, 1), replies); try std.testing.expectEqual(@as(u16, 0), state.reply_len); - try std.testing.expectEqualStrings("\x1b[6n", try replayBytes(pane, std.testing.allocator)); } -test "replay keeps a bounded chronological suffix across large writes and wrapping" { - if (comptime !enabled) return error.SkipZigTest; - const gpa = std.testing.allocator; - const pane = try create(gpa, 40, 12); - defer gpa.destroy(pane); - defer deinitEmulator(pane, gpa); - try std.testing.expectEqualStrings("", try replayBytes(pane, gpa)); - ingest(pane, "hello"); - const small = try replayBytes(pane, gpa); - try std.testing.expectEqualStrings("hello", small); - try std.testing.expectEqual(@intFromPtr(&pane.terminal.?.replay.bytes), @intFromPtr(small.ptr)); - - const capacity = pane.terminal.?.replay.bytes.len; - const input = try gpa.alloc(u8, capacity + 7); - defer gpa.free(input); - @memset(input, 0); - const ending = "\x1b[31mred\x1b[0m\r\n"; - @memcpy(input[input.len - ending.len ..], ending); - ingest(pane, input); - try std.testing.expectEqualSlices(u8, input[7..], try replayBytes(pane, gpa)); - ingest(pane, "next\r\n"); - const wrapped = try replayBytes(pane, gpa); - defer gpa.free(wrapped); - try std.testing.expectEqual(capacity, wrapped.len); - try std.testing.expectEqualSlices(u8, input[7 + "next\r\n".len ..], wrapped[0 .. capacity - "next\r\n".len]); - try std.testing.expectEqualStrings("next\r\n", wrapped[capacity - "next\r\n".len ..]); -} - -test "replay restores terminal cells from the retained stream" { +test "a terminal stream restores the cells it drew" { if (comptime !enabled) return error.SkipZigTest; const gpa = std.testing.allocator; const p = try Pardes.init(gpa, .{ .tty_only = true, .cols = 40, .rows = 12 }); defer p.deinit(); while (p.nextEffect()) |_| {} const pane = p.panes[0].?; - feedOutput(p, pane, "\x1b[31mred\x1b[0m\r\ncafé\r\n"); - const encoded = try dump.encodeBytes(gpa, try replayBytes(pane, gpa)); + const drawn = "\x1b[31mred\x1b[0m\r\ncafé\r\n"; + feedOutput(p, pane, drawn); + const encoded = try dump.encodeBytes(gpa, drawn); defer gpa.free(encoded); const restored = try restore(p, .{ .kind = .terminal, @@ -835,7 +766,6 @@ test "replay restores terminal cells from the retained stream" { const after = try screenTextAlloc(restored, gpa); defer gpa.free(after); try std.testing.expectEqualStrings(before, after); - try std.testing.expectEqualSlices(u8, try replayBytes(pane, gpa), try replayBytes(restored, gpa)); } test "a restored command pane comes back finished, its command not run again" { diff --git a/src/tty/tty.zig b/src/tty/tty.zig index 4cd1bc9a..afc0dbf0 100644 --- a/src/tty/tty.zig +++ b/src/tty/tty.zig @@ -34,6 +34,8 @@ var kitty_shm = std.atomic.Value(bool).init(false); var kitty_shm_probing = std.atomic.Value(bool).init(false); const kitty_shm_probe_id = 0x7fff_fff1; var kitty_shm_serial: u32 = 0; +/// Transmissions an unread shm segment outlives before it is removed. +const kitty_shm_window = 64; fn kittyShmName(buf: []u8, n: u32) ![:0]const u8 { return std.fmt.bufPrintSentinel(buf, "/pardes-{d}-{d}", .{ libc.getpid(), n }, 0); @@ -1787,6 +1789,13 @@ const Shell = struct { if (!s.kitty.contains(key) and vx.caps.kitty_graphics and kitty_shm.load(.acquire)) shm: { var buf: [64]u8 = undefined; kitty_shm_serial +%= 1; + // Bounded: a segment this many transmissions old was never + // read (a terminal reads one as its command arrives), and a + // tmpfs holds it in memory until it goes. + if (kitty_shm_serial > kitty_shm_window) { + var old_buf: [64]u8 = undefined; + if (kittyShmName(&old_buf, kitty_shm_serial - kitty_shm_window)) |old| _ = libc.shm_unlink(old) else |_| {} + } const name = kittyShmName(&buf, kitty_shm_serial) catch break :shm; if (!kittyShmWrite(name, place.rgba)) break :shm; if (vx.transmitLocalImagePath(s.gpa, s.tty.writer(), name, @intCast(place.iw), @intCast(place.ih), .shared_mem, .rgba) catch null) |handle| @@ -2010,7 +2019,7 @@ const Shell = struct { old.future.cancel(s.io) catch {}; s.lsp_task = null; } - const future = s.io.concurrent(lspWorker, .{ s.lsp_gpa, job, s.loop }) catch |err| { + const future = s.io.concurrent(lspWorker, .{ s.io, s.lsp_gpa, job, s.loop }) catch |err| { job.free(s.lsp_gpa); s.core.update(.{ .lsp_resp = .{ .id = req.id, .rows = null } }); return s.core.reportError(req.pane, "lsp", err); @@ -2144,9 +2153,9 @@ fn requestClipboard(vx: *vaxis.Vaxis, tty: *vaxis.Tty) void { vx.requestSystemClipboard(tty.writer()) catch {}; } -fn lspWorker(allocator: std.mem.Allocator, job: *host_io.Lsp.Job, loop: *Loop) anyerror!void { +fn lspWorker(io: std.Io, allocator: std.mem.Allocator, job: *host_io.Lsp.Job, loop: *Loop) anyerror!void { var sink: LspRowSink = .{ .allocator = allocator, .loop = loop }; - host_io.Lsp.work(allocator, job, &sink, LspRowSink.take); + host_io.Lsp.work(io, allocator, job, &sink, LspRowSink.take); } const LspRowSink = struct { @@ -2185,23 +2194,30 @@ fn watchFiles(io: std.Io, fd: c_int, loop: *Loop) anyerror!void { return; } const file: std.Io.File = .{ .handle = fd, .flags = .{ .nonblocking = false } }; - var read_buf: [4096]u8 = undefined; - var reader = file.readerStreaming(io, &read_buf); + // One buffer, read into directly: a Reader around it only added a second + // (and in a safe build, filled both as `undefined` on every start). + var buf: [4096]u8 = undefined; while (true) { - var buf: [4096]u8 = undefined; - var vec = [_][]u8{&buf}; - const n = reader.interface.readVec(&vec) catch break; + const n = file.readStreaming(io, &.{&buf}) catch break; if (n == 0) break; loop.postEvent(.files_changed) catch break; } } fn winchWatch(loop: *Loop, vx: *vaxis.Vaxis, tty: *vaxis.Tty) void { + winchWatchUntil(loop, vx, tty, null); +} + +/// A SIGWINCH watcher that ends: `stop` set, then a SIGWINCH sent to its +/// thread, wakes its sigwait and it returns, so whoever owns `loop` can join +/// it before the loop goes (attachSession's loop is on its stack). +fn winchWatchUntil(loop: *Loop, vx: *vaxis.Vaxis, tty: *vaxis.Tty, stop: ?*const std.atomic.Value(bool)) void { var set = posix.sigemptyset(); posix.sigaddset(&set, posix.SIG.WINCH); while (true) { var sig: c_int = 0; if (libc.sigwait(&set, &sig) != 0) continue; + if (stop) |flag| if (flag.load(.acquire)) return; if (vx.state.in_band_resize) continue; // terminal reports via CSI 48 const ws = tty.getWinsize() catch continue; loop.postEvent(.{ .winsize = ws }) catch {}; @@ -2285,12 +2301,11 @@ fn wakeFs(ctx: ?*anyopaque) void { } fn readPty(io: std.Io, gpa: std.mem.Allocator, pty: std.Io.File, id: usize, gen: u32, loop: *Loop) anyerror!void { - var read_buf: [0x10000]u8 = undefined; - var reader = pty.readerStreaming(io, &read_buf); + // One buffer, read into directly: a Reader around it only added a second + // (and in a safe build, filled both as `undefined` on every start). + var buf: [0x10000]u8 = undefined; while (true) { - var buf: [0x10000]u8 = undefined; - var vec = [_][]u8{&buf}; - const n = reader.interface.readVec(&vec) catch break; + const n = pty.readStreaming(io, &.{&buf}) catch break; if (n == 0) break; const bytes = try gpa.dupe(u8, buf[0..n]); loop.postEvent(.{ .pty_read = .{ .id = id, .gen = gen, .bytes = bytes } }) catch { @@ -2301,7 +2316,7 @@ fn readPty(io: std.Io, gpa: std.mem.Allocator, pty: std.Io.File, id: usize, gen: _ = loop.tryPostEvent(.{ .pty_eof = .{ .id = id, .gen = gen } }) catch {}; } -const max_paste_bytes: usize = 4 << 20; +const max_paste_bytes: usize = pardes.memory.limits.max_paste_bytes; fn pasteBytes(key: vaxis.Key) []const u8 { const text = key.text orelse ""; @@ -3023,7 +3038,13 @@ fn attachSession(init: std.process.Init, name: []const u8, tty: *vaxis.Tty, vx: startInput(&loop, &input_cache) catch |err| return .{ .lost = err }; defer stopInput(&loop); - (std.Thread.spawn(.{}, winchWatch, .{ &loop, vx, tty }) catch |err| return .{ .lost = err }).detach(); + var winch_stop: std.atomic.Value(bool) = .init(false); + const winch = std.Thread.spawn(.{}, winchWatchUntil, .{ &loop, vx, tty, &winch_stop }) catch |err| return .{ .lost = err }; + defer { + winch_stop.store(true, .release); + _ = libc.pthread_kill(winch.getHandle(), posix.SIG.WINCH); + winch.join(); + } vx.queryTerminalSend(tty.writer()) catch {}; return attachLoop(&a); diff --git a/test/lspbench.zig b/test/lspbench.zig index d0e25aee..08df9f4c 100644 --- a/test/lspbench.zig +++ b/test/lspbench.zig @@ -110,8 +110,13 @@ test "lspbench options reject unknown flags and multiple roots before probing" { try std.testing.expectError(error.MultipleRoots, Options.parse(&.{ "first", "--check", "second" })); } +/// The process's Io, for the queries: their waits are woken by the client's +/// reader threads, which a single-threaded Io cannot be. +var query_io: std.Io = undefined; + pub fn main(init: std.process.Init) !void { const gpa = std.heap.page_allocator; + query_io = init.io; const args = try init.minimal.args.toSlice(init.arena.allocator()); if (args.len > 1 and std.mem.eql(u8, args[1], "probe")) return probeCommand(init); @@ -171,13 +176,19 @@ fn checkResults(results: []const Result, required: usize) !void { for (results) |r| if (r.empty or !r.correct) return error.IncorrectProbe; } -fn runOnce(gpa: std.mem.Allocator, req: lsp.Req, expected: []const u8, r: *Result) !u64 { - var arena: std.heap.ArenaAllocator = .init(gpa); - defer arena.deinit(); +fn runOnce(gpa: std.mem.Allocator, req_in: lsp.Req, expected: []const u8, r: *Result) !u64 { + // The editor's own path (host_io.Lsp.work): the query in the fixed + // region reserved at startup, not a heap arena that maps its memory anew. + const memory = @import("pardes").memory; + const region = memory.takeLspQuery() orelse return error.QueryRegionBusy; + var arena: std.heap.FixedBufferAllocator = .init(region); + defer memory.giveLspQuery(arena.end_index); + var req = req_in; + req.budget = &arena; var out: std.Io.Writer.Allocating = .init(gpa); defer out.deinit(); const t0 = nowNs(); - try lsp.query(gpa, arena.allocator(), req, &out.writer); + try lsp.query(query_io, gpa, arena.allocator(), req, &out.writer); const ns = nowNs() -| t0; r.accept(out.written(), expected); return ns / 1000; @@ -341,6 +352,7 @@ extern "c" fn usleep(usec: c_uint) c_int; /// [--reps N] [--warm S]`, rows on stdout, the timing on stderr. fn probeCommand(init: std.process.Init) !void { const gpa = std.heap.page_allocator; + query_io = init.io; const arena = init.arena.allocator(); const io = std.Io.Threaded.global_single_threaded.io(); @@ -431,7 +443,7 @@ fn probeCommand(init: std.process.Init) !void { var scratch: std.heap.ArenaAllocator = .init(gpa); defer scratch.deinit(); const t0 = nowUs(); - try lsp.query(gpa, scratch.allocator(), .{ + try lsp.query(query_io, gpa, scratch.allocator(), .{ .kind = kind, .path = path, .source = source, diff --git a/test/panes.zig b/test/panes.zig index e98e3a54..1460b355 100644 --- a/test/panes.zig +++ b/test/panes.zig @@ -1913,9 +1913,6 @@ test "terminal output keeps following new rows after scrollback eviction" { try std.testing.expectEqualStrings("f", effect.write.bytes.slice()); try std.testing.expect(p.nextEffect() == null); } - const replay = &pane.terminal.?.replay; - try std.testing.expectEqual(replay.bytes.len, replay.len); - try std.testing.expect(replay.head > 0); } test "terminal edit stays on its surviving row when old scrollback is evicted" { diff --git a/test/perf-baseline-gui-debug.json b/test/perf-baseline-gui-debug.json index ffcd678f..a41c92b2 100644 --- a/test/perf-baseline-gui-debug.json +++ b/test/perf-baseline-gui-debug.json @@ -1 +1 @@ -{"harness":"15bee9f100e5ef64","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"gui","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1744,"med_us":1832,"p90_us":2468,"max_us":2524},{"op":"render","fixture":"medium","min_us":88,"med_us":97,"p90_us":114,"max_us":115},{"op":"key-down","fixture":"medium","min_us":95,"med_us":98,"p90_us":104,"max_us":105},{"op":"key-right","fixture":"medium","min_us":92,"med_us":96,"p90_us":98,"max_us":105},{"op":"page-down","fixture":"medium","min_us":91,"med_us":95,"p90_us":97,"max_us":105},{"op":"wheel","fixture":"medium","min_us":87,"med_us":93,"p90_us":99,"max_us":100},{"op":"edit-char","fixture":"medium","min_us":1241,"med_us":1459,"p90_us":1691,"max_us":1724},{"op":"edit-twin","fixture":"medium","min_us":2592,"med_us":3003,"p90_us":3481,"max_us":3503},{"op":"body-2m","fixture":"medium","min_us":4576,"med_us":8262,"p90_us":15424,"max_us":17259},{"op":"body-appends","fixture":"medium","min_us":152003,"med_us":227268,"p90_us":272872,"max_us":326510},{"op":"deep-sel","fixture":"medium","min_us":82,"med_us":83,"p90_us":90,"max_us":93}]} +{"harness":"442f274a5da9136c","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"gui","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":2253,"med_us":2626,"p90_us":3065,"max_us":3090},{"op":"render","fixture":"medium","min_us":100,"med_us":103,"p90_us":103,"max_us":110},{"op":"key-down","fixture":"medium","min_us":102,"med_us":103,"p90_us":105,"max_us":113},{"op":"key-right","fixture":"medium","min_us":98,"med_us":104,"p90_us":108,"max_us":113},{"op":"page-down","fixture":"medium","min_us":103,"med_us":104,"p90_us":111,"max_us":114},{"op":"wheel","fixture":"medium","min_us":98,"med_us":100,"p90_us":102,"max_us":108},{"op":"edit-char","fixture":"medium","min_us":1251,"med_us":1940,"p90_us":2652,"max_us":3041},{"op":"edit-twin","fixture":"medium","min_us":2779,"med_us":3177,"p90_us":3366,"max_us":3657},{"op":"body-2m","fixture":"medium","min_us":5812,"med_us":11423,"p90_us":13681,"max_us":13758},{"op":"body-appends","fixture":"medium","min_us":153898,"med_us":241075,"p90_us":284518,"max_us":305761},{"op":"deep-sel","fixture":"medium","min_us":86,"med_us":91,"p90_us":94,"max_us":122}]} diff --git a/test/perf-baseline-tty-debug.json b/test/perf-baseline-tty-debug.json index b8e534c4..dc37e033 100644 --- a/test/perf-baseline-tty-debug.json +++ b/test/perf-baseline-tty-debug.json @@ -1 +1 @@ -{"harness":"15bee9f100e5ef64","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1747,"med_us":1912,"p90_us":2150,"max_us":2247},{"op":"render","fixture":"medium","min_us":89,"med_us":95,"p90_us":102,"max_us":104},{"op":"key-down","fixture":"medium","min_us":97,"med_us":101,"p90_us":107,"max_us":133},{"op":"key-right","fixture":"medium","min_us":93,"med_us":104,"p90_us":109,"max_us":112},{"op":"page-down","fixture":"medium","min_us":92,"med_us":96,"p90_us":106,"max_us":108},{"op":"wheel","fixture":"medium","min_us":90,"med_us":92,"p90_us":96,"max_us":96},{"op":"edit-char","fixture":"medium","min_us":1167,"med_us":1345,"p90_us":1483,"max_us":1627},{"op":"edit-twin","fixture":"medium","min_us":2319,"med_us":2921,"p90_us":3425,"max_us":3539},{"op":"body-2m","fixture":"medium","min_us":4333,"med_us":7752,"p90_us":15941,"max_us":17192},{"op":"body-appends","fixture":"medium","min_us":149149,"med_us":234177,"p90_us":285302,"max_us":295516},{"op":"deep-sel","fixture":"medium","min_us":83,"med_us":88,"p90_us":99,"max_us":119}]} +{"harness":"442f274a5da9136c","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"debug","c_optimize":"fast","terminal_optimize":"safe","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":2181,"med_us":2495,"p90_us":2621,"max_us":2660},{"op":"render","fixture":"medium","min_us":96,"med_us":98,"p90_us":102,"max_us":112},{"op":"key-down","fixture":"medium","min_us":101,"med_us":103,"p90_us":107,"max_us":108},{"op":"key-right","fixture":"medium","min_us":100,"med_us":112,"p90_us":118,"max_us":119},{"op":"page-down","fixture":"medium","min_us":97,"med_us":101,"p90_us":105,"max_us":107},{"op":"wheel","fixture":"medium","min_us":98,"med_us":100,"p90_us":108,"max_us":110},{"op":"edit-char","fixture":"medium","min_us":1377,"med_us":1842,"p90_us":2336,"max_us":2418},{"op":"edit-twin","fixture":"medium","min_us":2655,"med_us":3200,"p90_us":3321,"max_us":3660},{"op":"body-2m","fixture":"medium","min_us":5420,"med_us":10753,"p90_us":12826,"max_us":15385},{"op":"body-appends","fixture":"medium","min_us":153569,"med_us":232519,"p90_us":281625,"max_us":309036},{"op":"deep-sel","fixture":"medium","min_us":91,"med_us":98,"p90_us":111,"max_us":111}]} diff --git a/test/perf-baseline-tty-fast.json b/test/perf-baseline-tty-fast.json index 46125349..bd8f32dc 100644 --- a/test/perf-baseline-tty-fast.json +++ b/test/perf-baseline-tty-fast.json @@ -1 +1 @@ -{"harness":"15bee9f100e5ef64","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"fast","c_optimize":"fast","terminal_optimize":"fast","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":1904,"med_us":2040,"p90_us":2290,"max_us":2600},{"op":"render","fixture":"medium","min_us":80,"med_us":83,"p90_us":88,"max_us":93},{"op":"key-down","fixture":"medium","min_us":81,"med_us":87,"p90_us":93,"max_us":109},{"op":"key-right","fixture":"medium","min_us":79,"med_us":82,"p90_us":86,"max_us":96},{"op":"page-down","fixture":"medium","min_us":79,"med_us":83,"p90_us":89,"max_us":93},{"op":"wheel","fixture":"medium","min_us":77,"med_us":79,"p90_us":82,"max_us":83},{"op":"edit-char","fixture":"medium","min_us":1139,"med_us":1338,"p90_us":1453,"max_us":1578},{"op":"edit-twin","fixture":"medium","min_us":2515,"med_us":2776,"p90_us":2879,"max_us":2907},{"op":"body-2m","fixture":"medium","min_us":4832,"med_us":7650,"p90_us":15076,"max_us":15371},{"op":"body-appends","fixture":"medium","min_us":165796,"med_us":240881,"p90_us":284919,"max_us":292573},{"op":"deep-sel","fixture":"medium","min_us":73,"med_us":75,"p90_us":83,"max_us":93}]} +{"harness":"442f274a5da9136c","build":{"compiler":"0.17.0","target":"x86_64-linux.5.10...7.2-gnu.2.38","cpu":"x86_64","driver_optimize":"fast","core_optimize":"fast","c_optimize":"fast","terminal_optimize":"fast","platform":"tty","grammars":"full","mupdf":true,"jpx":true,"tracy":false,"quic":false,"theme_animation":true,"terminal_simd":true,"prebuilt_shaders":false},"cols":120,"rows":40,"reps":10,"only":"medium","fixtures":[{"name":"small","lines":1000,"cols":60,"bytes":62694},{"name":"medium","lines":50000,"cols":60,"bytes":3201944},{"name":"large","lines":300000,"cols":60,"bytes":19464444},{"name":"longline","lines":400,"cols":8000,"bytes":3200400},{"name":"esp32p4","lines":12,"cols":240,"bytes":2892}],"cells":[{"op":"open","fixture":"medium","min_us":2283,"med_us":2611,"p90_us":2812,"max_us":2917},{"op":"render","fixture":"medium","min_us":89,"med_us":97,"p90_us":105,"max_us":119},{"op":"key-down","fixture":"medium","min_us":88,"med_us":95,"p90_us":100,"max_us":116},{"op":"key-right","fixture":"medium","min_us":90,"med_us":99,"p90_us":103,"max_us":106},{"op":"page-down","fixture":"medium","min_us":86,"med_us":90,"p90_us":96,"max_us":97},{"op":"wheel","fixture":"medium","min_us":85,"med_us":91,"p90_us":94,"max_us":101},{"op":"edit-char","fixture":"medium","min_us":1229,"med_us":1823,"p90_us":1874,"max_us":2101},{"op":"edit-twin","fixture":"medium","min_us":2196,"med_us":2745,"p90_us":2928,"max_us":2972},{"op":"body-2m","fixture":"medium","min_us":5606,"med_us":10369,"p90_us":14063,"max_us":14406},{"op":"body-appends","fixture":"medium","min_us":155867,"med_us":237195,"p90_us":289128,"max_us":306283},{"op":"deep-sel","fixture":"medium","min_us":82,"med_us":84,"p90_us":86,"max_us":91}]} diff --git a/test/perf.zig b/test/perf.zig index 7662d028..ab284739 100644 --- a/test/perf.zig +++ b/test/perf.zig @@ -665,8 +665,6 @@ fn measureTerm(op: TermOp, fx: TermFixture, reps: usize) !Cell { .row = pane.body.cur_row, .col = pane.body.cur_col, .edit_len = if (pane.ovl) |overlay| overlay.text.len else 0, - .replay_head = state.replay.head, - .replay_len = state.replay.len, .cols = pane.cols, .rows = pane.rows, }; @@ -689,22 +687,11 @@ fn measureTerm(op: TermOp, fx: TermFixture, reps: usize) !Cell { _ = try frame(core); const dt = nowNs() -| t0; switch (op) { - .render => if (pane.body.cur_row != before.row or pane.body.cur_col != before.col or - state.replay.head != before.replay_head or state.replay.len != before.replay_len) + .render => if (pane.body.cur_row != before.row or pane.body.cur_col != before.col) return error.RenderChangedTerminal, - .output => { - const capacity = state.replay.bytes.len; - if (term_chunk.len == 0 or term_chunk.len >= capacity) return error.InvalidOutputFixture; - const overflow = term_chunk.len -| (capacity - before.replay_len); - if (state.replay.len != @min(capacity, before.replay_len + term_chunk.len) or - state.replay.head != (before.replay_head + overflow) % capacity) - return error.OutputWasNotRecorded; - const start = (state.replay.head + state.replay.len - term_chunk.len) % capacity; - const first = @min(term_chunk.len, capacity - start); - if (!std.mem.eql(u8, term_chunk[0..first], state.replay.bytes[start..][0..first]) or - !std.mem.eql(u8, term_chunk[first..], state.replay.bytes[0 .. term_chunk.len - first])) - return error.WrongTerminalOutput; - }, + // What the output did to the screen is the snapshot suite's to + // check; here it is timed. + .output => if (term_chunk.len == 0) return error.InvalidOutputFixture, .resize_rows => if (pane.rows == before.rows or pane.cols != before.cols or state.vt.screens.active.pages.rows != pane.rows) return error.TerminalDidNotResize, .resize_cols => if (pane.cols == before.cols or pane.rows != before.rows or -- cgit v1.3