path: root/src/syntax.zig

//! Tree-sitter syntax highlighting: one style byte per content byte, filled by
//! running each grammar's highlights.scm query (slurped at build time into the
//! ts_queries options module). Grammar set is tiered: `zig`; `minimal` (c, cpp,
//! zig); and `full`, which adds ~24 languages lazily on first use.
const std = @import("std");
const config = @import("pardes_config");
const tracy = @import("tracy.zig");
const grammar_manifest = @import("grammar_manifest.zig");

pub const enabled = config.syntax_highlighting;
const zig_grammar = config.syntax_zig_grammar;
const minimal_grammars = config.syntax_minimal_grammars;
const full_grammars = config.syntax_full_grammars;

const ts = if (enabled) @import("tree-sitter") else struct {
    pub const Language = opaque {};
    pub const Query = opaque {};
};
const ts_queries = if (enabled) @import("ts_queries") else struct {};

const LanguageFn = *const fn () callconv(.c) *const ts.Language;

pub const Syn = enum(u8) { none, keyword, string, number, comment };

const Spec = struct {
    name: []const u8,
    exts: []const []const u8,
    language: *const fn () callconv(.c) *const ts.Language,
    query_src: []const u8,
    compiled_query: ?*ts.Query = null,
};

fn grammarSelected(comptime g: grammar_manifest.Grammar) bool {
    return switch (g.tier) {
        .zig => zig_grammar,
        .minimal => minimal_grammars,
        .full => full_grammars,
    };
}

fn specCount() comptime_int {
    var count = 0;
    for (grammar_manifest.all) |g| {
        if (grammarSelected(g)) count += 1;
    }
    return count;
}

// Upstream's typst highlights.scm names its markup honestly —
// @markup.heading.*, @markup.bold, @markup.italic, @markup.raw.block — and
// `synFor` now understands that vocabulary, so headings, bold, italics and raw
// blocks paint on their own. What is left here is exactly the two captures we
// refuse to map globally:
//
//   - upstream tags call callees @function/@function.method; mapping "function"
//     in `synFor` would recolour every function call in every language.
//   - upstream tags the code sigil "#" @operator, and mapping "operator" would
//     likewise light up every +, -, == in the codebase.
//
// Both are worth colouring *in typst specifically*: the sigil in front of every
// #let/#if/#import/#call is the visual anchor of the code/markup split, and
// upstream leaves it uncoloured even though the keyword behind it is not.
const typst_supplement =
    \\
    \\(call item: (ident) @keyword)
    \\(call item: (field field: (ident) @keyword))
    \\"#" @keyword
    \\
;

fn querySrc(comptime g: grammar_manifest.Grammar) []const u8 {
    const base = @field(ts_queries, g.name ++ "_highlights");
    if (comptime std.mem.eql(u8, g.name, "typst")) return base ++ typst_supplement;
    return base;
}

fn initSpecs() [specCount()]Spec {
    var out: [specCount()]Spec = undefined;
    var i = 0;
    inline for (grammar_manifest.all) |g| {
        if (grammarSelected(g)) {
            out[i] = .{
                .name = g.name,
                .exts = g.exts,
                .language = @extern(LanguageFn, .{ .name = "tree_sitter_" ++ g.name }),
                .query_src = querySrc(g),
            };
            i += 1;
        }
    }
    return out;
}

var specs = initSpecs();

const allocation_header_size = 16;
var syntax_allocator: std.mem.Allocator = undefined;
var syntax_started = false;

extern fn ts_set_allocator(
    new_malloc: ?*const fn (size: usize) callconv(.c) ?*anyopaque,
    new_calloc: ?*const fn (nmemb: usize, size: usize) callconv(.c) ?*anyopaque,
    new_realloc: ?*const fn (ptr: ?*anyopaque, size: usize) callconv(.c) ?*anyopaque,
    new_free: ?*const fn (ptr: ?*anyopaque) callconv(.c) void,
) void;

fn syntaxAlloc(size_arg: usize) callconv(.c) ?*anyopaque {
    const size = @max(size_arg, 1);
    const total = std.math.add(usize, allocation_header_size, size) catch return null;
    const bytes = syntax_allocator.alignedAlloc(u8, .@"16", total) catch return null;
    const header: *align(16) usize = @ptrCast(bytes.ptr);
    header.* = total;
    return @ptrCast(bytes.ptr + allocation_header_size);
}

fn syntaxCalloc(count: usize, size: usize) callconv(.c) ?*anyopaque {
    const len = std.math.mul(usize, count, size) catch return null;
    const pointer = syntaxAlloc(len) orelse return null;
    const bytes: [*]u8 = @ptrCast(pointer);
    @memset(bytes[0..len], 0);
    return pointer;
}

fn syntaxRealloc(ptr: ?*anyopaque, new_size: usize) callconv(.c) ?*anyopaque {
    const pointer = ptr orelse return syntaxAlloc(new_size);
    if (new_size == 0) {
        syntaxFree(pointer);
        return null;
    }
    const user: [*]u8 = @ptrCast(pointer);
    const header: *align(16) usize = @ptrCast(@alignCast(user - allocation_header_size));
    const old_total = header.*;
    const old_bytes: []align(16) u8 = @as([*]align(16) u8, @ptrCast(header))[0..old_total];
    const new_total = std.math.add(usize, allocation_header_size, new_size) catch return null;
    const new_bytes = syntax_allocator.realloc(old_bytes, new_total) catch return null;
    const new_header: *align(16) usize = @ptrCast(new_bytes.ptr);
    new_header.* = new_total;
    return @ptrCast(new_bytes.ptr + allocation_header_size);
}

fn syntaxFree(ptr: ?*anyopaque) callconv(.c) void {
    const pointer = ptr orelse return;
    const user: [*]u8 = @ptrCast(pointer);
    const header: *align(16) usize = @ptrCast(@alignCast(user - allocation_header_size));
    const total = header.*;
    const bytes: []align(16) u8 = @as([*]align(16) u8, @ptrCast(header))[0..total];
    syntax_allocator.free(bytes);
}

pub fn start(gpa: std.mem.Allocator) void {
    if (comptime enabled) {
        std.debug.assert(!syntax_started);
        syntax_allocator = gpa;
        syntax_started = true;
        ts_set_allocator(syntaxAlloc, syntaxCalloc, syntaxRealloc, syntaxFree);
    }
}

pub fn stop() void {
    if (comptime enabled) {
        std.debug.assert(syntax_started);
        for (&specs) |*spec| {
            if (spec.compiled_query) |query| query.destroy();
            spec.compiled_query = null;
        }
        ts_set_allocator(null, null, null, null);
        syntax_started = false;
        syntax_allocator = undefined;
    }
}

const Selected = struct { name: []const u8, lang: *const ts.Language, query: *ts.Query };

// NOTE: don't lang.destroy() — the tree_sitter_*() languages are static
// singletons reused on every open; destroying one use-after-frees the next.
fn ensure(spec: *Spec) !Selected {
    const lang = spec.language();
    if (spec.compiled_query) |query| return .{ .name = spec.name, .lang = lang, .query = query };
    var error_offset: u32 = 0;
    const query = try ts.Query.create(lang, spec.query_src, &error_offset);
    spec.compiled_query = query;
    return .{ .name = spec.name, .lang = lang, .query = query };
}

fn forExt(ext: []const u8) !?Selected {
    for (&specs) |*spec| {
        for (spec.exts) |choice| {
            if (std.ascii.eqlIgnoreCase(ext, choice)) return try ensure(spec);
        }
    }
    return null;
}

const LangAlias = struct { []const u8, []const u8 };
const lang_aliases = [_]LangAlias{
    .{ "js", "javascript" },
    .{ "jsx", "javascript" },
    .{ "mjs", "javascript" },
    .{ "py", "python" },
    .{ "sh", "bash" },
    .{ "shell", "bash" },
    .{ "zsh", "bash" },
    .{ "bash", "bash" },
    .{ "rs", "rust" },
    .{ "c++", "cpp" },
    .{ "cxx", "cpp" },
    .{ "cc", "cpp" },
    .{ "cs", "c_sharp" },
    .{ "kt", "kotlin" },
    .{ "rb", "ruby" },
    .{ "ml", "ocaml" },
    .{ "hs", "haskell" },
};

fn forLang(name: []const u8) !?Selected {
    var canonical = name;
    for (lang_aliases) |a| {
        if (std.ascii.eqlIgnoreCase(name, a[0])) {
            canonical = a[1];
            break;
        }
    }
    for (&specs) |*spec| {
        if (std.ascii.eqlIgnoreCase(canonical, spec.name)) return try ensure(spec);
    }
    return null;
}

fn runQuery(styles: []u8, sel: Selected, tree: *ts.Tree, base: usize) void {
    const cursor = ts.QueryCursor.create();
    defer cursor.destroy();
    cursor.exec(sel.query, tree.rootNode());
    while (cursor.nextMatch()) |match| {
        for (match.captures) |cap| {
            const syn = synFor(sel.query.captureNameForId(cap.index) orelse "");
            if (syn == .none) continue;
            var b: usize = base + cap.node.startByte();
            const end = @min(base + @as(usize, cap.node.endByte()), styles.len);
            while (b < end) : (b += 1) styles[b] = @intFromEnum(syn);
        }
    }
}

// Queries compile on first use (`ensure`), never at startup. Pre-compiling the
// compact tier in Pardes.init cost EVERY boot ~120ms of ts_query__perform_analysis
// (55% of a Debug startup) to save ~40ms on the first .zig/.c/.cpp open — a pane
// of prose or a shell paid for a language it never opened. Grammar availability
// is unchanged; only the timing moved.

fn synFor(name: []const u8) Syn {
    for ([_]struct { []const u8, Syn }{
        .{ "comment", .comment },
        .{ "string", .string },
        .{ "character", .string },
        .{ "number", .number },
        .{ "numeric", .number },
        .{ "float", .number },
        .{ "boolean", .number },
        .{ "keyword", .keyword },
        .{ "include", .keyword },
        .{ "conditional", .keyword },
        .{ "repeat", .keyword },
        .{ "title", .keyword },
        .{ "uri", .string },
        .{ "reference", .number },

        // Markup grammars (markdown, typst) name prose constructs in their own
        // vocabulary rather than the code vocabulary above, so none of the
        // needles so far reach them. These are appended, and first-match-wins
        // makes that strictly additive; the needles below were audited across
        // all 27 shipped queries and occur only in the markdown and typst ones,
        // so no other language is recoloured.
        //
        // The slot assignment is forced by the palette being four wide and by
        // `synStyle` attaching the real BOLD attribute to exactly two of them,
        // `keyword` and `comment`: headings take `keyword`, so bold spans have
        // to land on `comment` to render actually bold. Nothing is left that
        // renders italic, so emphasis can only get a colour shift (`number`).
        .{ "heading", .keyword },
        .{ "strong", .comment },
        .{ "bold", .comment },
        .{ "emphasis", .number },
        .{ "italic", .number },
        .{ "literal", .string },
        .{ "raw", .string },
    }) |m| {
        if (std.mem.indexOf(u8, name, m[0]) != null) return m[1];
    }
    return .none;
}

/// One Syn byte per content byte in [start, end). Caller frees.
pub fn highlightFileRange(gpa: std.mem.Allocator, path: []const u8, content: []const u8, start_byte_raw: usize, end_byte_raw: usize) ![]u8 {
    const tz = tracy.zone(@src(), "highlightFileRange");
    defer tz.end();
    if (!enabled) return &.{};
    const ext = std.fs.path.extension(path);
    const selected = (forExt(ext) catch return &.{}) orelse return &.{};

    const start_byte = @min(start_byte_raw, content.len);
    const end_byte = @max(start_byte, @min(end_byte_raw, content.len));
    const source = content[start_byte..end_byte];
    const styles = try gpa.alloc(u8, source.len);
    errdefer gpa.free(styles);
    @memset(styles, 0);

    const parser = ts.Parser.create();
    defer parser.destroy();
    parser.setLanguage(selected.lang) catch return styles;
    const tree = parser.parseString(source, null) orelse return styles;
    defer tree.destroy();

    runQuery(styles, selected, tree, 0);

    if (InjectSite.forGrammar(selected.name)) |site| injectCodeBlocks(styles, source, tree.rootNode(), site);
    // Disjoint from the fenced-block pass above: `code_fence_content` is never
    // an `inline` node, so the two never write the same byte.
    if (std.mem.eql(u8, selected.name, "markdown")) injectMarkdownInline(styles, source, tree.rootNode());
    return styles;
}

// Markdown fenced blocks and Typst raw blocks are the same construct — a
// language tag plus a literal payload — under different node shapes, so one
// walker drives both and only the (lang, content) extraction differs.
const InjectSite = enum {
    markdown_fence,
    typst_raw,

    fn forGrammar(name: []const u8) ?InjectSite {
        if (std.mem.eql(u8, name, "markdown")) return .markdown_fence;
        if (std.mem.eql(u8, name, "typst")) return .typst_raw;
        return null;
    }

    fn blockKind(self: InjectSite) []const u8 {
        return switch (self) {
            .markdown_fence => "fenced_code_block",
            .typst_raw => "raw_blck",
        };
    }

    /// null when the block carries no language tag (an untagged fence, or a
    /// Typst raw block written without one) — nothing to inject, leave it alone.
    fn parts(self: InjectSite, block: ts.Node) ?struct { lang: ts.Node, content: ts.Node } {
        switch (self) {
            .markdown_fence => {
                const info = childOfKind(block, "info_string") orelse return null;
                return .{
                    .lang = childOfKind(info, "language") orelse return null,
                    .content = childOfKind(block, "code_fence_content") orelse return null,
                };
            },
            .typst_raw => return .{
                .lang = block.childByFieldName("lang") orelse return null,
                .content = childOfKind(block, "blob") orelse return null,
            },
        }
    }
};

fn injectCodeBlocks(styles: []u8, source: []const u8, node: ts.Node, site: InjectSite) void {
    if (std.mem.eql(u8, node.kind(), site.blockKind())) {
        highlightCodeBlock(styles, source, node, site);
        return;
    }
    var i: u32 = 0;
    const count = node.childCount();
    while (i < count) : (i += 1) {
        if (node.child(i)) |c| injectCodeBlocks(styles, source, c, site);
    }
}

fn childOfKind(node: ts.Node, kind: []const u8) ?ts.Node {
    var i: u32 = 0;
    const count = node.childCount();
    while (i < count) : (i += 1) {
        if (node.child(i)) |c| {
            if (std.mem.eql(u8, c.kind(), kind)) return c;
        }
    }
    return null;
}

fn highlightCodeBlock(styles: []u8, source: []const u8, block: ts.Node, site: InjectSite) void {
    const p = site.parts(block) orelse return;
    const langtext = source[p.lang.startByte()..p.lang.endByte()];
    const sub_sel = (forLang(langtext) catch return) orelse return;
    const cs: usize = p.content.startByte();
    const ce: usize = p.content.endByte();
    if (ce > source.len or cs > ce) return;

    const parser = ts.Parser.create();
    defer parser.destroy();
    parser.setLanguage(sub_sel.lang) catch return;
    const tree = parser.parseString(source[cs..ce], null) orelse return;
    defer tree.destroy();
    // The outer grammar already painted these bytes (Typst blankets the whole
    // raw block `string`), and runQuery only writes bytes it captures, so the
    // outer colour would survive as a wash behind the injected code. The sub
    // grammar owns the payload outright: clear it first.
    @memset(styles[cs..ce], @intFromEnum(Syn.none));
    runQuery(styles, sub_sel, tree, cs);
}

// tree-sitter-markdown is a split grammar: the block parser bottoms out at named
// `inline` nodes whose bytes it never looks inside, and emphasis /
// strong_emphasis / code_span exist only in the companion inline parser. So
// every `inline` node is re-parsed with `markdown_inline`, which is what
// upstream's injections.scm, helix and nvim all do (per node, uncombined — the
// stray block_continuation markers inside a multi-line paragraph's range are
// just plain text to the inline parser).
//
// The grammar is resolved and the parser built once per file, not once per node:
// only the parse is inherently per node.
fn injectMarkdownInline(styles: []u8, source: []const u8, root: ts.Node) void {
    // Absent in builds below the `full` tier — nothing to inject, leave the
    // block grammar's colours alone.
    const sub_sel = (forLang("markdown_inline") catch return) orelse return;
    const parser = ts.Parser.create();
    defer parser.destroy();
    parser.setLanguage(sub_sel.lang) catch return;
    inlineNodes(styles, source, root, sub_sel, parser);
}

fn inlineNodes(styles: []u8, source: []const u8, node: ts.Node, sel: Selected, parser: *ts.Parser) void {
    if (std.mem.eql(u8, node.kind(), "inline")) {
        highlightInline(styles, source, node, sel, parser);
        return; // `inline` nodes never nest
    }
    var i: u32 = 0;
    const count = node.childCount();
    while (i < count) : (i += 1) {
        if (node.child(i)) |c| inlineNodes(styles, source, c, sel, parser);
    }
}

fn highlightInline(styles: []u8, source: []const u8, node: ts.Node, sel: Selected, parser: *ts.Parser) void {
    const s: usize = node.startByte();
    const e: usize = node.endByte();
    if (s >= e or e > source.len) return; // an empty atx heading has a zero-length inline
    const tree = parser.parseString(source[s..e], null) orelse return;
    defer tree.destroy();
    // Deliberately NOT clearing the range first, unlike highlightCodeBlock: the
    // block query already painted a heading's inline text `keyword`, and that is
    // the colour the heading must keep wherever the inline pass captures
    // nothing. Painting over instead of resetting is what makes *italic* inside
    // a heading recolour while the rest of the heading stays heading-coloured.
    runQuery(styles, sel, tree, s);
}

/// One Syn byte per byte of content[start, end) for unified diffs/patches.
/// Pure byte scan; independent of tree-sitter and the `enabled` flag. Caller frees.
pub fn highlightDiff(gpa: std.mem.Allocator, content: []const u8, start_byte_raw: usize, end_byte_raw: usize) ![]u8 {
    const start_byte = @min(start_byte_raw, content.len);
    const end_byte = @max(start_byte, @min(end_byte_raw, content.len));
    const source = content[start_byte..end_byte];
    const styles = try gpa.alloc(u8, source.len);
    errdefer gpa.free(styles);
    @memset(styles, 0);

    var offset: usize = 0;
    var lines = std.mem.splitScalar(u8, source, '\n');
    while (lines.next()) |line| {
        const syn = diffLineSyn(line);
        if (syn != .none) @memset(styles[offset .. offset + line.len], @intFromEnum(syn));
        offset += line.len + 1;
    }
    return styles;
}

fn diffLineSyn(line: []const u8) Syn {
    if (std.mem.startsWith(u8, line, "@@")) return .keyword;
    if (std.mem.startsWith(u8, line, "+++") or
        std.mem.startsWith(u8, line, "---") or
        std.mem.startsWith(u8, line, "diff ") or
        std.mem.startsWith(u8, line, "index ") or
        std.mem.startsWith(u8, line, "\\ No newline")) return .comment;
    if (line.len == 0) return .none;
    if (line[0] == '+') return .string;
    if (line[0] == '-') return .number;
    return .none;
}

test "tree-sitter allocator callbacks preserve and free exact allocations" {
    syntax_allocator = std.testing.allocator;
    defer syntax_allocator = undefined;

    var live: ?*anyopaque = syntaxCalloc(4, 1) orelse return error.OutOfMemory;
    defer if (live) |pointer| syntaxFree(pointer);
    const original: [*]u8 = @ptrCast(live.?);
    try std.testing.expectEqualSlices(u8, &.{ 0, 0, 0, 0 }, original[0..4]);
    @memcpy(original[0..4], "data");

    live = syntaxRealloc(live, 32) orelse return error.OutOfMemory;
    const grown: [*]u8 = @ptrCast(live.?);
    try std.testing.expectEqualSlices(u8, "data", grown[0..4]);

    try std.testing.expect(syntaxCalloc(std.math.maxInt(usize), 2) == null);
    try std.testing.expect(syntaxRealloc(live, 0) == null);
    live = null;
}

test "default full grammar set highlights Typst source" {
    if (!enabled or !full_grammars) return;
    start(std.testing.allocator);
    defer stop();

    const source = "// note\n#let answer = 42\n#let text = \"hello\"\n";
    const styles = try highlightFileRange(std.testing.allocator, "paper.typst", source, 0, source.len);
    defer std.testing.allocator.free(styles);

    const comment_at = std.mem.indexOf(u8, source, "// note").?;
    const keyword_at = std.mem.indexOf(u8, source, "let").?;
    const number_at = std.mem.indexOf(u8, source, "42").?;
    const string_at = std.mem.indexOf(u8, source, "\"hello\"").?;
    try std.testing.expectEqual(Syn.comment, @as(Syn, @enumFromInt(styles[comment_at])));
    try std.testing.expectEqual(Syn.keyword, @as(Syn, @enumFromInt(styles[keyword_at])));
    try std.testing.expectEqual(Syn.number, @as(Syn, @enumFromInt(styles[number_at])));
    try std.testing.expectEqual(Syn.string, @as(Syn, @enumFromInt(styles[string_at])));

    const short_ext = try highlightFileRange(std.testing.allocator, "paper.typ", source, 0, source.len);
    defer std.testing.allocator.free(short_ext);
    try std.testing.expectEqual(Syn.keyword, @as(Syn, @enumFromInt(short_ext[keyword_at])));
}

test "Typst markup constructs paint and raw blocks inject their language" {
    if (!enabled or !full_grammars) return;
    start(std.testing.allocator);
    defer stop();

    const source =
        \\// note
        \\= Heading Title
        \\Some *bold* text with `raw` inline.
        \\#import "mod.typ": helper
        \\#let answer = 42
        \\#emit(7, "arg")
        \\```zig
        \\fn widget() u8 { return 99; }
        \\```
        \\
    ;
    const styles = try highlightFileRange(std.testing.allocator, "paper.typ", source, 0, source.len);
    defer std.testing.allocator.free(styles);

    const synAt = struct {
        fn f(s: []const u8, src: []const u8, needle: []const u8, offset: usize) Syn {
            return @enumFromInt(s[std.mem.indexOf(u8, src, needle).? + offset]);
        }
    }.f;

    // marker and text of the heading both take the bold accent
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "= Heading", 0));
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "Heading", 0));
    // *bold* is @markup.bold, which lands on the one slot that still renders
    // with the real bold attribute now that headings own `keyword`.
    try std.testing.expectEqual(Syn.comment, synAt(styles, source, "bold", 0));
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "raw` inline", 0));
    // the callee and the code sigil in front of it
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "emit", 0));
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "#emit", 0));

    // fence and lang tag keep the literal colour of the raw block...
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "```zig", 0));
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "```zig", 3));
    // ...while the blob is reset and re-painted by the injected zig grammar. Its
    // `fn` and `99` prove the injection ran; `widget` proves the reset, since the
    // zig query names it @function (Syn.none here) and without clearing the blob
    // first the raw block's `string` would still be washing over it.
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "fn widget", 0));
    try std.testing.expectEqual(Syn.number, synAt(styles, source, "99", 0));
    try std.testing.expectEqual(Syn.none, synAt(styles, source, "widget", 0));

    try std.testing.expectEqual(Syn.comment, synAt(styles, source, "// note", 0));
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "import", 0));
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "let", 0));
    try std.testing.expectEqual(Syn.number, synAt(styles, source, "42", 0));
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "\"arg\"", 0));
}

test "markdown highlights markup, injects inline spans and fenced code blocks" {
    if (!enabled or !full_grammars) return;
    start(std.testing.allocator);
    defer stop();

    const source =
        "# Heading *slant* Title\n" ++
        "\n" ++
        "Prose with **stout** and *lean* plus `snippet` inline.\n" ++
        "\n" ++
        "```zig\n" ++
        "fn gadget() u8 { return 77; }\n" ++
        "```\n";
    const styles = try highlightFileRange(std.testing.allocator, "doc.md", source, 0, source.len);
    defer std.testing.allocator.free(styles);

    const synAt = struct {
        fn f(s: []const u8, src: []const u8, needle: []const u8, offset: usize) Syn {
            return @enumFromInt(s[std.mem.indexOf(u8, src, needle).? + offset]);
        }
    }.f;

    // The block grammar washes the whole heading `keyword`; the inline pass then
    // paints the emphasis over it without resetting, so the heading keeps its
    // colour everywhere the emphasis is not.
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "Title", 0));
    try std.testing.expectEqual(Syn.number, synAt(styles, source, "slant", 0));

    // bold/italic/code-span live only in the inline grammar, so all three prove
    // the inline injection ran.
    try std.testing.expectEqual(Syn.comment, synAt(styles, source, "stout", 0));
    try std.testing.expectEqual(Syn.number, synAt(styles, source, "lean", 0));
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "snippet", 0));

    // the fence is inside the block query's @text.literal wash...
    try std.testing.expectEqual(Syn.string, synAt(styles, source, "```zig", 0));
    // ...while the payload is cleared and re-painted by the injected zig
    // grammar: `fn` and `77` prove the injection ran, and `gadget` proves the
    // clear, since the zig query names it @function (Syn.none here) and without
    // clearing first the fence's `string` would still be washing over it.
    try std.testing.expectEqual(Syn.keyword, synAt(styles, source, "fn gadget", 0));
    try std.testing.expectEqual(Syn.number, synAt(styles, source, "77", 0));
    try std.testing.expectEqual(Syn.none, synAt(styles, source, "gadget", 0));
}

test "highlightDiff colors unified diff lines by prefix" {
    const diff =
        "diff --git a/x b/x\n" ++
        "--- a/x\n" ++
        "+++ b/x\n" ++
        "@@ -1,3 +1,3 @@\n" ++
        " context\n" ++
        "-old line\n" ++
        "+new line\n";
    const styles = try highlightDiff(std.testing.allocator, diff, 0, diff.len);
    defer std.testing.allocator.free(styles);

    const byteSyn = struct {
        fn at(s: []const u8, src: []const u8, needle: []const u8) Syn {
            const i = std.mem.indexOf(u8, src, needle).?;
            return @enumFromInt(s[i]);
        }
    }.at;

    try std.testing.expectEqual(Syn.comment, byteSyn(styles, diff, "diff --git"));
    try std.testing.expectEqual(Syn.comment, byteSyn(styles, diff, "--- a/x"));
    try std.testing.expectEqual(Syn.comment, byteSyn(styles, diff, "+++ b/x"));
    try std.testing.expectEqual(Syn.keyword, byteSyn(styles, diff, "@@ -1,3"));
    try std.testing.expectEqual(Syn.none, byteSyn(styles, diff, " context"));
    try std.testing.expectEqual(Syn.number, byteSyn(styles, diff, "-old line"));
    try std.testing.expectEqual(Syn.string, byteSyn(styles, diff, "+new line"));
}