bench: measure ghostty's release and tip together, and harden wl-latency #379

Merged

natsukium merged 3 commits from bench/wl-latency-wouldblock into main 2026-09-24 18:37:13 +09:00

Conversation 0 Commits 3 Files changed 12 +238 -24

natsukium commented 2026-09-24 16:22:17 +09:00

Owner

Copy link

The published Linux benchmarks show ghostty-tip at a 965 ms mean keystroke latency. Bisecting shows this is a tip-only regression: 1.3.1 and the commit before upstream's GtkGLArea removal (ghostty-org/ghostty#14052) stay under 30 ms, while every later build sometimes draws a typed glyph only on the renderer's 600 ms cursor-blink tick. The published run had no ghostty release column to show that, and the run that reproduced it also found an instrument crash.

• wl-latency waits out a full Wayland send buffer. blocking_dispatch and roundtrip flush before reading, and the flush returned WouldBlock on a loaded host, which panicked the leg and recorded it as the terminal's failure. Both calls now poll for writability and retry.
• A full-field run refuses to start without both ghostty builds. The release had silently dropped out of the published run (its store path was most likely garbage-collected). --terminals still narrows a run.
• The Linux bench shell pins ghostty's tip through a dev-partition input (FELIS_BENCH_GHOSTTY_TIP), so it no longer needs a hand fetch and is identified by the commit in dev/flake.lock, which lock maintenance advances. macOS keeps --tool.

Not included: republishing docs/reference/benchmarks.md with the release column; that waits for the next full run.

Verified: crossterm_test.py (240 tests), nix flake check --no-build, darwin bench-shell evaluation, and a latency run of 1.3.1 and tip through the rebuilt wl-latency.

The published Linux benchmarks show ghostty-tip at a 965 ms mean keystroke latency. Bisecting shows this is a tip-only regression: 1.3.1 and the commit before upstream's GtkGLArea removal (ghostty-org/ghostty#14052) stay under 30 ms, while every later build sometimes draws a typed glyph only on the renderer's 600 ms cursor-blink tick. The published run had no ghostty release column to show that, and the run that reproduced it also found an instrument crash. - **wl-latency waits out a full Wayland send buffer.** `blocking_dispatch` and `roundtrip` flush before reading, and the flush returned `WouldBlock` on a loaded host, which panicked the leg and recorded it as the terminal's failure. Both calls now poll for writability and retry. - **A full-field run refuses to start without both ghostty builds.** The release had silently dropped out of the published run (its store path was most likely garbage-collected). `--terminals` still narrows a run. - **The Linux bench shell pins ghostty's tip** through a dev-partition input (`FELIS_BENCH_GHOSTTY_TIP`), so it no longer needs a hand fetch and is identified by the commit in `dev/flake.lock`, which lock maintenance advances. macOS keeps `--tool`. Not included: republishing `docs/reference/benchmarks.md` with the release column; that waits for the next full run. Verified: `crossterm_test.py` (240 tests), `nix flake check --no-build`, darwin bench-shell evaluation, and a latency run of 1.3.1 and tip through the rebuilt wl-latency.

natsukium added 3 commits 2026-09-24 16:22:17 +09:00

bench: let wl-latency wait out a full Wayland send buffer ... b07a900e82

wayland-client's blocking_dispatch and roundtrip flush pending requests
before they read, and the flush returns WouldBlock instead of waiting
when the compositor has not drained the socket. The instrument treated
that as fatal, so on a loaded host a latency leg panicked mid-run
("dispatch: Backend(Io(... WouldBlock ...))") and was recorded as a
failure of the terminal under test. Both calls now poll the connection
for writability and retry; any other error still aborts.

Assisted-by: Claude Code:Opus 5.5

bench: refuse a full-field run that lacks either ghostty build ... 11e9aafd0c

ghostty's release and its tip can disagree by two orders of magnitude:
on Linux, keystroke latency is about 10 ms on 1.3.1 and up to seconds
on tip, a regression that arrived with tip's GTK renderer rework. The
published Linux run carried only tip, because the release's pinned
store path had gone missing and the field silently dropped it, so the
page could not tell a tip-only regression from ghostty's own behavior.
A run over the full field now stops before measuring unless both
builds resolve; --terminals still narrows a run for iteration.

Assisted-by: Claude Code:Opus 5.5

bench: pin ghostty's tip in the Linux bench shell ... 2162d8a432

A full-field run now needs both ghostty builds, and on Linux the tip was
the one a person had to fetch by hand before every run. Upstream's flake
builds it, so the dev partition takes it as an input and the bench shell
exports it as FELIS_BENCH_GHOSTTY_TIP. The tip is then a commit recorded
in dev/flake.lock rather than whatever was fetched that day, and lock
maintenance moves it forward.

This reverses the earlier stance that a lockfile-pinned nightly is not a
nightly: what the field needs is the unreleased code next to the release,
and a recorded commit identifies it better than a build that replaces
itself. macOS keeps --tool, since upstream's flake does not build there.
The input does not follow nixpkgs-dev because the tip compiles only with
the Zig upstream pins.

Assisted-by: Claude Code:Opus 5.5

natsukium merged commit c2b995a5a4 into main 2026-09-24 18:37:13 +09:00

natsukium deleted branch bench/wl-latency-wouldblock 2026-09-24 18:37:13 +09:00

natsukium referenced this pull request from a commit 2026-09-24 18:37:15 +09:00

bench: measure ghostty's release and tip together, and harden wl-latency (#379)

natsukium referenced this pull request 2026-09-24 21:20:53 +09:00

docs: republish the macOS benchmarks with ghostty's release and resource use #382

natsukium referenced this pull request from a commit 2026-09-24 23:04:05 +09:00

docs: republish the macOS benchmarks with ghostty's release and resource use (#382)

Commenting is not possible because the repository is archived.

Reviewers

No reviewers

Labels

Clear labels   

priority/P0

Must be resolved before the v0.1.0 tag

  

priority/P1

Must be settled before the first public contract freezes

  

priority/P2

Explicitly deferred until after v0.1.0

  

release/v0.1.0

Work tracked for the first tagged release

  

status/blocked

Cannot start now: waits on an upstream release, another issue, or a recorded revisit trigger

  

status/planned

Triaged with a source-grounded plan and a passed pi review; ready to implement

  

type/bug

Shipped behavior that contradicts the docs or the code's own intent; needs a product code change

  

type/design

An open scope or trade-off question; needs a recorded decision in the owning explanation doc before code

  

type/test-gap

Documented or reviewed behavior that no test pins; usually split from a review finding, needs no product code change

  

type/tracker

Parent issue that tracks child work

No labels

priority/P0

priority/P1

priority/P2

release/v0.1.0

status/blocked

status/planned

type/bug

type/design

type/test-gap

type/tracker

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

natsukium/felis!379

No description provided.