Repository navigation
perf(ci): Benchmark browser integration tests on ARM - #26434
nicohrubec wants to merge 3 commits into
Conversation
Co-Authored-By: OpenAI Codex <codex@openai.com>
size-limit report 📦Compared: f944acd (PR test merge using develop at c7b765f) Against: c7b765f (develop) · base run #66318 ✅ Baseline matches: this diff contains only this PR's changes.
|
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit e42696a. Configure here.
|
|
||
| env: | ||
| HEAD_COMMIT: ${{ github.event.inputs.commit || github.sha }} | ||
| HEAD_COMMIT: ${{ github.event.inputs.commit || github.event.pull_request.head.sha || github.sha }} |
There was a problem hiding this comment.
PR CI checks out unmerged head
High Severity
HEAD_COMMIT now prefers github.event.pull_request.head.sha over github.sha, so PR jobs check out the branch tip instead of GitHub's merge commit. job_get_metadata still receives github.sha, and Nx affected compares that head against pull_request.base.sha from a pair of depth-1 checkouts. Those two commits usually do not share fetched ancestry, so git merge-base can fail and take down job_build for multi-commit or out-of-date PRs. Even when it works, CI no longer tests the merged result.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit e42696a. Configure here.
| uses: actions/checkout@v7 | ||
| with: | ||
| ref: ${{ env.HEAD_COMMIT }} | ||
| ref: ${{ github.sha }} |
There was a problem hiding this comment.
Bundler tests checkout mismatches build
Medium Severity
job_bundler_plugin_integration_tests now checks out github.sha while job_build and the rest of the workflow use HEAD_COMMIT (pull_request.head.sha on PRs). This job still restores job_build's node_modules cache and build-output artifacts, so on a PR it runs merge-commit sources against binaries built from the PR tip. Those trees diverge whenever the PR is behind develop.
Reviewed by Cursor Bugbot for commit e42696a. Configure here.
Optional E2E testsSome optional checks were not successful. They do not block the merge. 2 not successful, 61 successful. See the run.
Successful checks
|
Co-Authored-By: OpenAI Codex <codex@openai.com>
Co-Authored-By: OpenAI Codex <codex@openai.com>


What
Benchmark browser integration tests on ARM, preserving browsers, shards and six workers. Current x86 runners have eight cores; ARM has four. Architecture-separated browser caches enable a warm comparison.
Why
Keep the current x86 runners for now. Warm ARM was slower at the median and failed 10/35 browser jobs.
*The same 24 Chromium jobs that passed in all three runs. Complete-job time includes setup/cleanup. Warm ARM browser-cache restoration was verified.
ARM consistently failed eight Chromium Replay jobs on click-coordinate snapshots. Both WebKit jobs could not launch: missing
libgtk-4.so.1,libgraphene-1.0.so.0,libsecret-1.so.0, andlibx264.so. All 9,926 case identities were selected; assertions were unchanged. This draft retains the benchmark experiment for inspection.