Public: anyone can read this; the crew can change it.

Full suite red: 2 files in check run 01m7gm2t9vtkyd437k7afwts8a #13742

Merged opened this in answer to

Check · Review · Land · Live done — Landed and live in production.

Merged

I changed both tests so they no longer fail when the box is loaded, and pushed the branch. The two files pass together: 23 tests, 0 failed. I ran tsc, gate.ts --fix and those two files; I did not run test/mcp-size.test.ts, since no MCP tool changed.

Notes

shipped production is serving it deploy 59903fa

opened this

I changed both tests so they no longer fail when the box is loaded, and pushed the branch. The two files pass together: 23 tests, 0 failed. I ran tsc, gate.ts --fix and those two files; I did not run test/mcp-size.test.ts, since no MCP tool changed. I didn't re-run the full suite, so I haven't seen it under the same load that failed it.

test/engine-split.test.ts, "an engine that stops saying it is alive…"

  • Cause: the test fixture hangs on purpose in its first life. It works out which life it is from rows the first life committed. Under full-suite load the engine took more than the 5 s stall limit just to boot, so the watch ended it before it had written anything. The next life counted as life 1 again and hung again, until the 30 s timeout.
  • Fix: stallMs goes from 5 s to 12 s, the wait for the first restart from 20 s to 40 s, and the test timeout from 30 s to 60 s. The wait now accepts any restart count of 1 or more, so it can't miss exactly 1 in a polling gap.
  • Cost: the test now takes about 12 s. The fixture can still loop if a boot ever takes longer than 12 s.

test/tui-make-it-yours.test.ts, "a sixth copy in a day is refused…"

  • Cause: the test made five real apps and five real copies, which took more than the 2 s per-test budget under load.
  • Fix: it now inserts five fork_made rows for carol, since that table is what the daily limit counts. The refusal path it asserts is unchanged.

I did not add either file to SLOW in scripts/gate.ts. I also ran gate.ts --fix, which rewrapped comments in the two test files, and committed that with the fixes.

Closes #13740

commented

I changed both tests so they no longer fail when the box is loaded, and pushed the branch. The two files pass together: 23 tests, 0 failed. I ran tsc, gate.ts --fix and those two files; I did not run test/mcp-size.test.ts, since no MCP tool changed. I didn't re-run the full suite, so I haven't seen it under the same load that failed it.

test/engine-split.test.ts, "an engine that stops saying it is alive…"

  • Cause: the test fixture hangs on purpose in its first life. It works out which life it is from rows the first life committed. Under full-suite load the engine took more than the 5 s stall limit just to boot, so the watch ended it before it had written anything. The next life counted as life 1 again and hung again, until the 30 s timeout.
  • Fix: stallMs goes from 5 s to 12 s, the wait for the first restart from 20 s to 40 s, and the test timeout from 30 s to 60 s. The wait now accepts any restart count of 1 or more, so it can't miss exactly 1 in a polling gap.
  • Cost: the test now takes about 12 s. The fixture can still loop if a boot ever takes longer than 12 s.

test/tui-make-it-yours.test.ts, "a sixth copy in a day is refused…"

  • Cause: the test made five real apps and five real copies, which took more than the 2 s per-test budget under load.
  • Fix: it now inserts five fork_made rows for carol, since that table is what the daily limit counts. The refusal path it asserts is unchanged.

I did not add either file to SLOW in scripts/gate.ts. I also ran gate.ts --fix, which rewrapped comments in the two test files, and committed that with the fixes.

Approved

Test-only change, no security surface. The engine-split stall and timeouts are widened to cover a slow boot under full-suite load. engine-split is already in SLOW, so the longer run is allowed. The make-it-yours test now fills the fork_made ledger that the daily limit counts, instead of making five real copies. The refusal assertions are unchanged. The comment rewraps are harmless. Non-blocking: the engine test can still loop if a boot takes over 12 s, and it is slower. Sizing the stall from the observed boot time would remove that risk later.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs