<feed xmlns='http://www.w3.org/2005/Atom'>
<title>pardes.git/test/perf.zig, branch macos</title>
<subtitle>Pardes</subtitle>
<id>https://git.0x4200.cafe/pardes.git/atom?h=macos</id>
<link rel='self' href='https://git.0x4200.cafe/pardes.git/atom?h=macos'/>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/'/>
<updated>2026-10-30T02:59:23Z</updated>
<entry>
<title>perf --live: the sessions a run leaves (a killed stub's, a wrapper's detached one) ended by their private runtime directory, after each cell and in the cleanup every path takes</title>
<updated>2026-10-30T02:59:23Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-30T02:14:10Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=c02ab55482c055c57146b9ea9a5a78c20bcdf122'/>
<id>urn:sha1:c02ab55482c055c57146b9ea9a5a78c20bcdf122</id>
<content type='text'>
Since phase 3 a tty stub killed detaches its session rather than ending it, and an exe that wraps a session of its own (`--detach`, as the remote agent's client-live.sh starts one) leaves it when the harness kills the wrapper. Every process a live run starts inherits its private XDG_RUNTIME_DIR, which nothing else has: `liveSweep` sends TERM, then KILL, to each whose /proc environ names it, after every cell and in liveCleanup, which every path takes. With a wrapper that leaves a daemon-kept detached session, the old harness left three (one a cell) and their daemon; this one none.

perf.zig is the harness's identity, so the gate's baselines are recorded again from base (main and this perf.zig), each built on the build cores and run alone on the perf cores; their build identities are the old ones. .scratch/daemon/base/pardes-perf-live is rebuilt from it (ReleaseFast).

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>perf --live: the flood through the real pty is `yes | head -c 32M`, not 1 GiB</title>
<updated>2026-10-29T21:17:34Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-29T20:36:58Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=efcb4c35e76ef33bbd62123a5c5574f88bc8bcc7'/>
<id>urn:sha1:efcb4c35e76ef33bbd62123a5c5574f88bc8bcc7</id>
<content type='text'>
The gigabyte took three minutes a run (seven runs to an interleaved
A/B), most of a measurement's time, and 32 MiB floods the pty as well:
the reader is starved or not in its first megabyte. The case is named
for its size (`live-tty-yes-32m`). perf.zig is the harness's identity,
so the gate's baselines are recorded again, from the base the line
started from, as before.

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>perf --live: keystroke-to-present through the real tty and GUI, `yes | head -c 1G` through a real pty, and a big file's open</title>
<updated>2026-10-29T17:30:01Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-29T13:46:16Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=97f83717cf8be3e4eedd3f7b932766c1cf72fd37'/>
<id>urn:sha1:97f83717cf8be3e4eedd3f7b932766c1cf72fd37</id>
<content type='text'>
The baseline the daemon work is measured against (docs/next-daemon.md). `pardes-perf --live &lt;pardes&gt;` runs the real tty binary in a pty (test/e2e_harness.zig, as the snapshot harness does): FrameStats.keys's gestures (key-down, key-right, edit-char, page-down, wheel) on the 50k-line fixture, a phase each, each frame's latency from the shell taking the input off its queue to the end of its present, read back from the session's PARDES_FRAMES dump; `yes | head -c 1G` typed into a bash pane in tty mode, wall time to the mark after it and the frames drawn; and a 1,000,000-line file opened on the command line, wall time to its first line on screen. `--live-gui &lt;pardes-gui&gt;` runs the same gestures as the GUI's own timedemo (PARDES_FRAMES_DEMO=keys) in a hidden test window. p50 and p99 per phase, pooled over --reps runs; text or --json.

Each run has a private home (bash with a `$ ` prompt), XDG_RUNTIME_DIR and fixtures under $TMPDIR, all removed after. The frames dump now says the monotonic time of its first frame (`# t0_ns`), so a driver lines its own phase times up with the frames'.

The perf gate's baselines (test/perf-baseline-{tty-debug,tty-fast,gui-debug}.json) are recorded again: the harness id is perf.zig's hash, and the medium cells came out within about 10% of the old ones (edit-char and edit-twin faster).

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>Callstack lists every caller and pages a tree too big for one query; memory bounds and resident-size fixes</title>
<updated>2026-10-23T20:10:58Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-23T20:10:58Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=0073d05329335dbb0eb85f5c3044e42805d144e5'/>
<id>urn:sha1:0073d05329335dbb0eb85f5c3044e42805d144e5</id>
<content type='text'>
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 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>zig fmt for Zig 0.17</title>
<updated>2026-10-22T13:03:15Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-21T20:25:52Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=23518c9d7c061f8a06f6632da8b64b37801b514f'/>
<id>urn:sha1:23518c9d7c061f8a06f6632da8b64b37801b514f</id>
<content type='text'>
</content>
</entry>
<entry>
<title>Build with Zig 0.17</title>
<updated>2026-10-22T13:03:15Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-21T17:35:15Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=bc3d18d27f9fc1155db7fac3946c4d38774f9e65'/>
<id>urn:sha1:bc3d18d27f9fc1155db7fac3946c4d38774f9e65</id>
<content type='text'>
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 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>Zerox opens a second pane on a file pane's buffer: one text, undo and unsaved state, a scroll and a cursor each; an edit in either is spliced into both at File.setContentSpan, O(the edit) more, and what a Save, a get, the disk watch or a rename changes on one is copied to the rest after each step</title>
<updated>2026-10-01T22:57:27Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-10-01T22:50:32Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=dce94f4453735a6f71385cc020ee79c7604609e7'/>
<id>urn:sha1:dce94f4453735a6f71385cc020ee79c7604609e7</id>
<content type='text'>
Twins share the content buffer, the History and, while it is one, the line
index; the other twin's cursor and scroll move past the splice as acme
moves a clone's q0. Del of one twin asks nothing while another holds the
text, Exit lists the text once, and a closed twin keeps a copy of its own
until it is torn down. Both are in the index, each its serial, one name.
An event reader on one hears no edit made through another (first cut).
A dump keeps the twins as two panes on the file, not linked again.
perf: edit-twin, typing beside a twin of the 50k-line file, gated within
3x of edit-char in the same run; the three baselines re-recorded for it.

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>A selection deep in a long line costs a render what one at its start does: 5 MB in, /screen goes from 14.7 s to 5 ms</title>
<updated>2026-10-01T03:12:17Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-09-30T21:40:41Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=9a256d34ebca317e3007adf71c2b94016059d4a3'/>
<id>urn:sha1:9a256d34ebca317e3007adf71c2b94016059d4a3</id>
<content type='text'>
Every render measured a selection's and the cursor's columns from the
start of their line (lineDisplayOffset walked the whole prefix twice),
so a selection near the end of a 5 MB line stalled each frame, 3 s at
1 MB in and 14.7 s at 5 MB. The cost grew with the offset, with Wrap on
or off. Three changes make it independent of where the selection is:
- display width adds (a tab is tab_width wherever it stands), so the
  offset between two columns is the width of the text between them,
  never the prefix;
- the selection's rows are clamped to what is shown: a row wholly
  before the selection is skipped, a start before the row's first
  column starts there, and an end past its last column ends there;
- a head or cursor whose column lies past its visible row (the last
  visible row of a line owns every column after its start) is not
  measured, since it is not drawn.
Measured on a Debug build, 5 MB line: offset 10, 1 MB and 5 MB now take
11, 6 and 5 ms per /screen. A new perf gate case, deep-sel, renders with
the selection 2 MB into a line: 83 us on Debug. The baselines were
re-recorded for the new harness.

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>An append to a body through 9P costs its own bytes, not five passes over the whole body: 60 KB appends go from 35.7 to 2.1 ms each</title>
<updated>2026-10-01T03:12:17Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-09-30T13:50:18Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=048e81e3471c73b9ac690901819976f689f6a4ce'/>
<id>urn:sha1:048e81e3471c73b9ac690901819976f689f6a4ce</id>
<content type='text'>
Round 25 measured bulk body writes at about 8 ms a write. There is no frame wait in it: a profile of open, write, clunk in a loop found the flush of each close walking the whole body five times. dotOf, setDot and showOffset turned the cursor between offsets and rows by counting every newline from the top; setContent found the line index's changed span by comparing old and new byte for byte, and hashed the new text to see whether it was back to the saved one. The rows now come from the file's line index by binary search (checked against the counting at every offset), a splice tells setContent the span it changed, and the text is hashed only when its length is the saved text's. Measured over 9P on a Debug build: 60 KB appends 35.7 to 2.1 ms, 8 KB 5.3 to 0.8 ms, 1 KB 1.3 to 0.9 ms. What is left is two copies of the text an edit (the splice's and the undo snapshot's). The perf gate gains body-appends, 64 appends of 8 KB on the 50k-line file, and its three baselines are recorded again.

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
<entry>
<title>Terminal.zig is terminal.zig: a file of functions and no fields takes a namespace's lowercase name</title>
<updated>2026-10-01T03:12:17Z</updated>
<author>
<name>Gabriel Schneider</name>
<email>gbrls@0x4200.cafe</email>
</author>
<published>2026-09-29T23:56:14Z</published>
<link rel='alternate' type='text/html' href='https://git.0x4200.cafe/pardes.git/commit/?id=04b047fb366ea2c595708fb62467f690fce4b9ec'/>
<id>urn:sha1:04b047fb366ea2c595708fb62467f690fce4b9ec</id>
<content type='text'>
The last of the deferred renames, now that the tty and theme agents have landed: panes.terminal at its importers, the alias lines in Text.zig and File.zig and panes.zig's own uses following. dump.zig's Terminal struct, a dump record, is not this. The served sources list and docs/design.typ name the new file. test/perf.zig's references change, so the three perf baselines take its new harness id with their numbers as recorded. No behaviour changes.

Co-Authored-By: Claude Opus 5.5 &lt;noreply@anthropic.com&gt;
</content>
</entry>
</feed>
