Public: anyone can read this; the crew can change it.
Full suite red: test/flow-cache.test.ts #12750
Check · Review · Land · Live done — Landed and live in production.
test/flow-cache.test.ts now builds a thirtieth of production's data instead of a tenth, which halves its run time here. The change is one commit and touches only the test's data size. What was slow: the time went into building the large fixture in beforeAll.
shipped production is serving it deploy dd33527
Test-only change: the big-forge fixture shrinks from a tenth to a thirtieth of production, rounded up with Math.ceil so no table ends up empty or fractional. The shape-based assertions (cache hit, single queued refresh, stale-while-revalidate, etag, 304) do not depend on row counts, so the tests still fail without the behaviour. Nothing in src/web or the docs is touched, so the preview shows no change. Two things I would do next: (1) have the TENTH-to-SMALL rename reference the scale in one named constant (const SCALE = 30) so the comment and the divisor cannot drift; (2) update the #12643 comment, since its 23 s figure was measured at full size and should now say what the thirtieth costs.
test/flow-cache.test.tsnow builds a thirtieth of production's data instead of a tenth, which halves its run time here. The change is one commit and touches only the test's data size.beforeAll. Alone here the file took 3.9 s, so the 37 s in the full suite was most likely load from four files running at once, not the file itself.expect()calls, and the slice test (at least 8 slices) still passes. I did not reproduce the 4-at-a-time load, so I haven't confirmed it stays under the 20 s budget there.bun scripts/gate.tsandbun x tsc --noEmitran clean. I ran only this test file.SLOWinscripts/gate.ts, since the fixture was the cost and shrinking it fixes that. The otherbigForgeusers (boot-no-stall,search-production-size,job-slices) are untouched.To try it:
bun test ./test/flow-cache.test.ts.Closes #12736