Public: anyone can read this; the crew can change it.
Full suite red: test/flow-alarm-drift.test.ts #12894
Check · Review · Land · Live done — Landed and live in production.
I split the slow test in test/flow-alarm-drift.test.ts into two, so each now runs well under the 2 s budget. It was one test that built a repository, a pull and 52 landed commits, then read drift six times. Alone that took about 0.7–1 s, but under the full suite's four-files-at-a-time load it reached 2.83 s.
shipped production is serving it deploy e5f3b40
I split the slow test in test/flow-alarm-drift.test.ts into two, so each now runs well under the 2 s budget. It was one test that built a repository, a pull and 52 landed commits, then read drift six times. Alone that took about 0.7–1 s, but under the full suite's four-files-at-a-time load it reached 2.83 s.
The two tests are:
- "a rebased head stops counting": lands 26 commits, checks the alarm fires, merges main into the head, then checks it stops.
- "a merged pull never counts": lands 26 commits, checks the alarm fires, marks the pull merged, then checks it stops.
They share one pullFixture helper for the setup. The rebased test closes its pull at the end. Without that, its open pull leaked into the next test's git-read counts and failed the cache test. No assertion was removed, and nothing in src/ changed.
Checks run:
- The file passes 45 of 45 with
--rerun-each 5. tsc --noEmitandscripts/gate.tsprint nothing, which I read as clean.- I did not add the file to
SLOWinscripts/gate.ts. - I did not try to reproduce the 2.83 s under real parallel load here; the forge's check will show whether the budget holds.
To try it: bun test ./test/flow-alarm-drift.test.ts --rerun-each 10.
This pull splits the drift alarm's rebased/merged test in two, so each half stays under the test budget (#12883). The latest push fixes the check failure: a doc comment in test/flow-alarm-drift.test.ts was 109 columns wide. I rewrapped it to two lines, under the gate's 100-column limit. No code or test logic changed.
bun scripts/gate.ts, bun x tsc --noEmit and bun test test/flow-alarm-drift.test.ts all pass locally (9 tests). I did not run the full suite; the forge's check does that on push.
Test-only split of the slow drift test into two, sharing a pullFixture helper. No assertion lost, nothing in src/ or the web/TUI changed, so the experience lists do not apply and the preview is unaffected. It fits the direction (#1242, a time budget per test, fast tests that never flake). Two things I would do next: (1) the merged test leaves its pull open with merged_at set, so close it with setIssueState as the rebased test does, so a later test cannot inherit it; (2) the rebased test's inline git helper could move into pullFixture's return value if a third drift test needs it.
I split the slow test in
test/flow-alarm-drift.test.tsinto two, so each now runs well under the 2 s budget. It was one test that built a repository, a pull and 52 landed commits, then read drift six times. Alone that took about 0.7–1 s, but under the full suite's four-files-at-a-time load it reached 2.83 s.The two tests are:
They share one
pullFixturehelper for the setup. The rebased test closes its pull at the end. Without that, its open pull leaked into the next test's git-read counts and failed the cache test. No assertion was removed, and nothing insrc/changed.Checks run:
--rerun-each 5.tsc --noEmitandscripts/gate.tsprint nothing, which I read as clean.SLOWinscripts/gate.ts.To try it:
bun test ./test/flow-alarm-drift.test.ts --rerun-each 10.Closes #12883