docs: republish the Linux benchmarks from a quiet run at 5f5e4957 #456

Merged

natsukium merged 2 commits from docs/linux-benchmarks-5f5e4957 into main 2026-09-25 18:52:03 +09:00

Conversation 0 Commits 2 Files changed 11 +75 -67

natsukium commented 2026-09-25 18:17:40 +09:00

Owner

Copy link

Part of #405 (the "re-run the full cross-terminal suite on a quiet host" item).

Republishes the Linux section of docs/reference/benchmarks.md and its charts from a fresh full run at 5f5e4957 (nix build .#felis), replacing the 9d1d842f figures. The macOS section is untouched.

Run. crossterm.py run over all eight suites and the full field, ROUNDS=3 with rotated order, the published page's pins (DejaVu Sans Mono 14 pt, 106x35 cells). The CI runner was stopped for the whole run (16:17-18:10); load started at 1.61. The Windows CI guest stayed up. Every leg completed in every round: no refused, unfinished, unpinned or retried legs.

What moved.

• vtebench scrolling rows: felis's round-to-round spread falls from 25-50% to 3-15%, and the order is now resolved. felis trails foot by 25-33 ms on the five non-fullscreen scrolling rows (137-149 ms against 111-116 ms), with no overlap. felis leads scrolling_fullscreen (6.2 against foot's 6.8 ms, ranges apart). #400 stays open (see the comment there).
• termbench long_lines 667 MB/s (was 552). kittenbench ASCII 446 and CSI 142 MB/s now lead ghostty-tip. DOOM-fire 2,954 fps (was 2,632). cat unicode/csi 0.80/0.74 s.
• Startup: cold 156 ms / warm 149 ms (was 236 / 174). Warm is below cold in every round.
• Render area: felis is 1167x808 px against the peers' 1166x805, so the pixel caveat shrinks to 0.5-5%.
• Idle PSS of every GPU terminal is 44-64 MB higher than in the previous run, and foot's is unchanged. The page already says to compare within one run. Its example now names only the terminals whose builds did not change.

Kept on the page. foot's vtebench legs follow ghostty-tip's in every round and start while the one-minute average is 10-14. ghostty-tip's ~1 s latency is still a redraw delay and not seen in the 1.3.1 release.

Verified: every figure came from the harness's report.md (per-round artifacts checked for the prose claims); just prose-check clean; treefmt applied. pi-review (sol): PASS, with one non-blocking wording fix applied.

Part of #405 (the "re-run the full cross-terminal suite on a quiet host" item). Republishes the Linux section of `docs/reference/benchmarks.md` and its charts from a fresh full run at `5f5e4957` (`nix build .#felis`), replacing the `9d1d842f` figures. The macOS section is untouched. **Run.** `crossterm.py run` over all eight suites and the full field, `ROUNDS=3` with rotated order, the published page's pins (DejaVu Sans Mono 14 pt, 106x35 cells). The CI runner was stopped for the whole run (16:17-18:10); load started at 1.61. The Windows CI guest stayed up. Every leg completed in every round: no refused, unfinished, unpinned or retried legs. **What moved.** - vtebench scrolling rows: felis's round-to-round spread falls from 25-50% to 3-15%, and the order is now resolved. felis trails foot by 25-33 ms on the five non-fullscreen scrolling rows (137-149 ms against 111-116 ms), with no overlap. felis leads `scrolling_fullscreen` (6.2 against foot's 6.8 ms, ranges apart). #400 stays open (see the comment there). - termbench `long_lines` 667 MB/s (was 552). kittenbench ASCII 446 and CSI 142 MB/s now lead ghostty-tip. DOOM-fire 2,954 fps (was 2,632). `cat` unicode/csi 0.80/0.74 s. - Startup: cold 156 ms / warm 149 ms (was 236 / 174). Warm is below cold in every round. - Render area: felis is 1167x808 px against the peers' 1166x805, so the pixel caveat shrinks to 0.5-5%. - Idle PSS of every GPU terminal is 44-64 MB higher than in the previous run, and foot's is unchanged. The page already says to compare within one run. Its example now names only the terminals whose builds did not change. **Kept on the page.** foot's vtebench legs follow ghostty-tip's in every round and start while the one-minute average is 10-14. ghostty-tip's ~1 s latency is still a redraw delay and not seen in the 1.3.1 release. Verified: every figure came from the harness's `report.md` (per-round artifacts checked for the prose claims); `just prose-check` clean; treefmt applied. pi-review (sol): PASS, with one non-blocking wording fix applied.

natsukium added 1 commit 2026-09-25 18:17:40 +09:00

docs: republish the Linux benchmarks from a quiet run at 5f5e4957 ... e91f784b19

The Linux figures came from 9d1d842f, before the parse, pen, repaint,
cell-size and scroll-directive work since then, and from a run the CI
runner loaded to 3.6 at the start. They no longer described what felis
ships, so the page stated gaps and round-to-round spread that the
current build does not have.

This run used the same field, workloads, font and 106x35 grid, with
three rotated rounds, and every leg completed in every round. The CI
runner was stopped for the run, so the load started at 1.6; the Windows
CI guest stayed up, and ghostty-tip's vtebench legs still lift the
one-minute average to 10-14 just before foot's legs.

What the new figures change on the page:

- vtebench scrolling rows: felis now varies by 3-15% between rounds
  instead of 25-50%, and the order is resolved: felis trails foot by
  25-33 ms on the five non-fullscreen scrolling rows, with foot's legs
  starting under the higher load average. felis leads
  scrolling_fullscreen with the ranges apart.
- termbench long_lines 667 MB/s, kittenbench ASCII and CSI now lead
  ghostty-tip, DOOM-fire 2,954 fps, cat unicode/csi 0.80/0.74 s.
- startup: warm 149 ms is below cold 156 ms in every round.
- The render area is 1167x808 px against the peers' 1166x805, so the
  pixel-count caveat shrinks to 0.5-5%.
- Idle PSS of every GPU terminal is 44-64 MB higher than in the
  9d1d842f run while foot's barely moved; the page already says to
  compare within one run, and the example now names only the terminals
  whose builds did not change.

Assisted-by: Claude Code

natsukium referenced this pull request 2026-09-25 18:17:50 +09:00

daemon: Linux vtebench scrolling rows trail the peers and vary 25-50% between rounds #400

natsukium added 1 commit 2026-09-25 18:44:43 +09:00

ci: rerun the checks after a job failure that does not reproduce locally ... 8dfbe898ef

The cargo build/clippy/test/deny job failed after 2m14s on a docs-only change; the full nextest run passes locally on the same head (3571 passed).

Assisted-by: Claude Code

natsukium merged commit af33db6145 into main 2026-09-25 18:52:03 +09:00

natsukium deleted branch docs/linux-benchmarks-5f5e4957 2026-09-25 18:52:03 +09:00

natsukium referenced this pull request from a commit 2026-09-25 18:52:04 +09:00

docs: republish the Linux benchmarks from a quiet run at 5f5e4957 (#456)

natsukium referenced this pull request 2026-09-25 19:00:17 +09:00

bench: follow up the 9d1d842f cross-terminal run #405

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!456

No description provided.