Repository navigation
fix(js-render): bump Playwright to 1.64.1 for the Ubuntu 26.04 runner - #12259
Merged
Merged
Conversation
GitHub's ubuntu-latest label now resolves to Ubuntu 26.04 (image 20261012.163.1). Playwright 1.55.1 has no host-platform entry for it, so `npx playwright install --with-deps chromium` in the setup-node composite action failed with "Playwright does not support chromium on ubuntu26.04-x64", and every impl-generate and impl-repair run for the five JavaScript libraries died in its setup step. 1.64.1 carries the ubuntu26.04-x64 download paths and apt dependency list and installs Chromium 156 instead of 140. The lockfile change is playwright + playwright-core plus the darwin-only fsevents optional entry 1.64 no longer declares; esbuild stays at 0.28.1. This matches the playwright-core@1.64.1 pin site-smoke and ci-image already use. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0175hA62juZoajMcqKxWFGza
Since the move to Python 3.14 (#12247), ruff format rewrites a parenthesised multi-exception `except` to the PEP 758 unparenthesised form. The two tokenizer fallbacks from #12244 predate that and have kept "CI: Lint and Format Check" red on main since 10:16 UTC, which fails every open PR's lint job on its merge ref. Same semantics; the mirror file stays byte-identical to core/adaptability.py's block. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0175hA62juZoajMcqKxWFGza
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0175hA62juZoajMcqKxWFGza
Contributor
There was a problem hiding this comment.
🟢 Approval recommended
The only remaining review comment is a non-blocking changelog-format nit; no approval-blocking issues were identified.
1 open issue
What changed in this PR
This pull request restores JavaScript rendering on Ubuntu 26.04 by upgrading Playwright and synchronizes the adaptability checker with Python 3.14 formatting requirements.
Changes:
- Upgrades Playwright and Playwright Core from 1.55.1 to 1.64.1.
- Updates lockfile metadata and removes the obsolete optional
fseventsentry. - Applies Python 3.14 exception syntax to the adaptability checker and generated mirror.
- Adds a changelog fragment documenting the Ubuntu 26.04 compatibility fix.
| File | Description |
|---|---|
package.json |
Pins the JavaScript rendering toolchain to Playwright 1.64.1. |
package-lock.json |
Records the updated Playwright packages, integrity metadata, and dependency changes. |
core/adaptability.py |
Applies Python 3.14-compatible exception syntax. |
changelog.d/playwright-ubuntu-26.md |
Documents the Ubuntu 26.04 rendering compatibility fix. |
automation/scripts/adaptability_check.py |
Keeps the generated standalone checker synchronized with the source implementation. |
Review effort: Lite
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0175hA62juZoajMcqKxWFGza
6 tasks done
MarkusNeusinger
added a commit
that referenced
this pull request
Oct 23, 2026
…imit (#12260) ## Summary - **Root cause of the `TestRender::test_a_launcher_that_outlives_its_kill_blocks_renders_until_it_exits` flake** (two CI failures on 2026-10-23, runs 39932413037 and 39940278486). The fake launcher execs the harness, so the process the test keeps alive *is* the harness and carries the run's `RLIMIT_CPU`, one second for a one-second job (`wire.run_cpu_seconds`). The harness start (matplotlib import, a fresh font cache per test) costs about 0.33 CPU s locally and roughly 2.4 times that on the CI runner (the plain render test takes 0.54 s here and 1.31 s there). Whenever it spent the full second, the kernel's SIGKILL ended the "stuck" launcher about 50 ms after the timeout, inside the 0.2 s grace window, so `stuck` read 0. The failing run's `wall_s=1.054` is that death: a survivor keeps `_stop` waiting the full grace, about 1.22 s. - **The executor is correct; only the test raced.** A real `sandbox do` runs on the host, outside the limit the harness sets inside the sandbox, and `stuck` counts only launchers still alive. The test now gives the run the host's full CPU cap and asserts that cap on the survivor with `resource.prlimit`, so it fails every time without the fix instead of once in a while. - Takes ruff's PEP 758 `except` form in two lines from #12244 that main's Lint job flags since the 3.14 move. These are the same hunks as #12259, so whichever PR lands first, the other merges cleanly. Ruled out: the fake binary's `delete --force` only logs and exits 0. The sleeper writes nothing, so it gets no SIGPIPE. Asyncio cancellation of the readers does not close the transport, and `PYTHONASYNCIODEBUG=1` with `-X dev` reports no "close running child" or unclosed-transport warning. Coverage does not follow subprocesses (pytest-cov 7, no `patch = subprocess`). Python 3.14 makes the harness start only about 5 % heavier than 3.13 (0.33 against 0.32 CPU s, same harness, same matplotlib), so it narrowed the margin but did not cause the flake. ## Plan N/A ## Test plan - [x] Deterministic reproduction of the CI signature (`assert (True and 0 == 1)`, `wall_s` about 1.02, launcher exit -9): the old test body with code that spends more than one CPU second before it sleeps. - [x] Adjusted test with the CPU-cap line removed: 31 of 31 failures, 16 idle and 15 under 48 busy loops on 32 threads. - [x] Adjusted test as committed: 50 of 50 passes, 25 idle and 25 under 48 busy loops. - [x] `uv run pytest tests/unit tests/integration -n 4 --dist worksteal`: 12,715 passed, 3 skipped. - [x] `tests/unit/agents/renderer` under `python -X dev` with `PYTHONASYNCIODEBUG=1`: 179 passed, no asyncio warnings. - [x] `ruff check .`, `ruff format --check .`, `mypy api core agents` and `tools.changelog check --base origin/main` are clean. --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0175hA62juZoajMcqKxWFGza
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
package.jsonandpackage-lock.json. GitHub'subuntu-latestnow resolves to Ubuntu 26.04 (image 20261012.163.1). Playwright 1.55.1 has noubuntu26.04-x64host platform, sonpx playwright install --with-deps chromiumin.github/actions/setup-nodefailed (run 39938127853). Everyimpl-generate/impl-repairrun for chartjs, d3, echarts, highcharts and muix died in setup. 1.64.1 ships the 26.04 download paths and apt dependency list. The lock diff is playwright + playwright-core plus the darwin-onlyfseventsoptional entry that 1.64 no longer declares. esbuild stays at 0.28.1. Integrity hashes match the registry. This also aligns with theplaywright-core@1.64.1pin thatscripts/site-smoke.mjsandci-image.ymlalready use.exceptlines that main's lint needed. fix(tests): keep the stuck-launcher survivor clear of the job's CPU limit #12260 landed the identical lines, and after merging main they collapsed, so the diff is Playwright only.Plan
N/A
Test plan
ubuntu:26.04(26.04.1 LTS, resolute) with Node 22.23.3 (the action's defaultnode-version: "22"). Everyrun:step of.github/actions/setup-node/action.ymlran verbatim against a minimal checkout:npm ci,npx playwright install --with-deps chromium(apt deps + Chrome for Testing 156.0.8078.4) and both smoke renders all pass. The chartjs PNG and the MUI X React-path PNG are both 3200×1800.Error: ERROR: Playwright does not support chromium on ubuntu26.04-x64.spiral-timeseries,biplot-pca) import@mui/x-charts/ChartContainer. MUI X 9.16 (deps(js): bump MUI X Charts to 9.16 with React 19 and Material 9.5 #12243) no longer exports that path, so both fail identically under BOTH Playwright versions. This is the known deps(js): bump MUI X Charts to 9.16 with React 19 and Material 9.5 #12243 blast radius, and the muix bulk regeneration covers it. The muix sample therefore uses three snippets that still bundle.ruff check,ruff format --check,mypy api core,pytest tests/unit tests/integration(12715 passed) andtools.changelog checkare green locally.Known gaps
.github/actions/setup-node. Onlyimpl-generate.ymlandimpl-repair.ymluse it, and neither runs on pull requests. A green PR therefore does not prove the install on the real runner. The 26.04 container run above is the closest local proof: same OS release, same Node major, same verbatim steps. The GitHub runner image adds preinstalled software on top of that.impl-generaterun for a JavaScript library after the merge. The muix bulk-regeneration pilot will be that run.impl-generaterun has yet passed setup on the new image. Python, uv and the pngquant apt install already pass there (run 39938127853, steps 9–13). Claude, image processing and the GCS upload have not run on 26.04 yet.