apps/bench/src/__tests__/bench-app.test.tsx → "dispatches comparator interaction scripts through measureBenchInteractionRun (B2 #5b)" now runs close to, and often past, vitest's default 5000ms timeout.
Evidence
| where |
duration |
outcome |
| CI, PR #433 first run |
13807ms |
failed — Test timed out in 5000ms |
| CI, same commit re-run |
— |
passed |
local, apps/bench at f22cf928 |
6399ms |
failed |
The local run is on a tree whose apps/bench is byte-identical to f22cf928 (git diff f22cf928 --name-only -- apps/bench is empty), so this reproduces on main, not on the PR that surfaced it — #433 adds a single JSON file under status/milestones/ and touches nothing in apps/bench.
Main's own CI run at f22cf928 passed, which is why this did not surface there. The test is marginal, not broken: whether it passes depends on runner speed.
Why now
#415 made the comparator adapters read column.wrap. This test renders ?adapter=ag-grid&scenario=S2&script=sort, so it now exercises ag-grid's wrapText + autoHeight path — per-cell measurement on mount — in jsdom, at the default dev scale, which is 750 rows × 40 columns for S2.
That is the intended behaviour change; the test simply was not re-timed against it. Its sibling "renders the requested mui competitor surface" moved to ~1-3.7s for the same reason and is heading the same way.
Suggested fix
A timeout bump is the safe change, but probably the wrong one. What the test asserts — that a comparator interaction script dispatches through measureBenchInteractionRun — is scale-independent: measureBenchInteractionRun is mocked with mockResolvedValueOnce, and the assertion is about the call, not about volume. Running it at scale=smoke (120 rows) would make it fast and stable while asserting exactly the same thing.
Worth confirming that reading before changing it, since the test belongs to the B2 comparator series and may be load-bearing in a way the name does not show.
Not fixed here: #433 was unblocked by a re-run, and picking the fix is a judgement call for whoever owns the B2 tests.
apps/bench/src/__tests__/bench-app.test.tsx→ "dispatches comparator interaction scripts through measureBenchInteractionRun (B2 #5b)" now runs close to, and often past, vitest's default 5000ms timeout.Evidence
Test timed out in 5000msapps/benchatf22cf928The local run is on a tree whose
apps/benchis byte-identical tof22cf928(git diff f22cf928 --name-only -- apps/benchis empty), so this reproduces on main, not on the PR that surfaced it — #433 adds a single JSON file understatus/milestones/and touches nothing inapps/bench.Main's own CI run at
f22cf928passed, which is why this did not surface there. The test is marginal, not broken: whether it passes depends on runner speed.Why now
#415 made the comparator adapters read
column.wrap. This test renders?adapter=ag-grid&scenario=S2&script=sort, so it now exercises ag-grid'swrapText+autoHeightpath — per-cell measurement on mount — in jsdom, at the defaultdevscale, which is 750 rows × 40 columns for S2.That is the intended behaviour change; the test simply was not re-timed against it. Its sibling "renders the requested mui competitor surface" moved to ~1-3.7s for the same reason and is heading the same way.
Suggested fix
A timeout bump is the safe change, but probably the wrong one. What the test asserts — that a comparator interaction script dispatches through
measureBenchInteractionRun— is scale-independent:measureBenchInteractionRunis mocked withmockResolvedValueOnce, and the assertion is about the call, not about volume. Running it atscale=smoke(120 rows) would make it fast and stable while asserting exactly the same thing.Worth confirming that reading before changing it, since the test belongs to the B2 comparator series and may be load-bearing in a way the name does not show.
Not fixed here: #433 was unblocked by a re-run, and picking the fix is a judgement call for whoever owns the B2 tests.