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

Red after landing: test/budget-raised.test.ts fails on main: two pulls landed a minute apart and disagree about the check limit #12963

Merged opened this in answer to

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

Merged

The check limit BUDGETS.gate is back to 60 min, so test/budget-raised.test.ts passes again. Gate and tsc were clean. The four test files I ran (landsim, budget-raised, check-budget, vitals) pass, 53 tests.

Notes

shipped production is serving it deploy 77d0004

opened this

The check limit BUDGETS.gate is back to 60 min, so test/budget-raised.test.ts passes again. Gate and tsc were clean. The four test files I ran (landsim, budget-raised, check-budget, vitals) pass, 53 tests. I did not run the whole always-run set; the forge's check does that on the push.

  • Limit: in src/kernel/budgets.ts, gate is 60 * 60_000 again. The comment now says the twice-killed run was the scheduled whole-suite measuring run (env verify, which has its own verify limit), not a pull's check.
  • Tests moved back to 60: test/check-budget.test.ts and test/vitals.test.ts read "limit 60" again.
  • limitSinceRaised cases: the cases in test/budget-raised.test.ts agree with the result. I deleted the #12947 test that asserted gate is above 60 min, because it contradicts the #12949 test that gate stays at 60 or lower. That test was wrong under the better diagnosis. The #12949 test stays and still passes.
  • Landing simulator: I added limitRaisedUnderCheck in test/support/landsim.ts and a case in test/landsim.test.ts. Pull 1 raises a limit and the tests that read it. Pull 2 was checked against the old base and adds a test that the limit stays at 60 or lower. After pull 1 lands, the forge must refuse pull 2 with land: check_stale, and it does. This models the pair for #12656's landing-simulator cases. I did not comment on #12656 itself.

I did not touch verify (150 min). I also did not make the simulator run the pulls' actual tests, so it covers the stale-check refusal, not semantic conflicts in general.

To try it: bun test test/budget-raised.test.ts test/landsim.test.ts.

Closes #12962

commented

The check limit BUDGETS.gate is back to 60 min, so test/budget-raised.test.ts passes again. Gate and tsc were clean. The four test files I ran (landsim, budget-raised, check-budget, vitals) pass, 53 tests. I did not run the whole always-run set; the forge's check does that on the push.

  • Limit: in src/kernel/budgets.ts, gate is 60 * 60_000 again. The comment now says the twice-killed run was the scheduled whole-suite measuring run (env verify, which has its own verify limit), not a pull's check.
  • Tests moved back to 60: test/check-budget.test.ts and test/vitals.test.ts read "limit 60" again.
  • limitSinceRaised cases: the cases in test/budget-raised.test.ts agree with the result. I deleted the #12947 test that asserted gate is above 60 min, because it contradicts the #12949 test that gate stays at 60 or lower. That test was wrong under the better diagnosis. The #12949 test stays and still passes.
  • Landing simulator: I added limitRaisedUnderCheck in test/support/landsim.ts and a case in test/landsim.test.ts. Pull 1 raises a limit and the tests that read it. Pull 2 was checked against the old base and adds a test that the limit stays at 60 or lower. After pull 1 lands, the forge must refuse pull 2 with land: check_stale, and it does. This models the pair for #12656's landing-simulator cases. I did not comment on #12656 itself.

I did not touch verify (150 min). I also did not make the simulator run the pulls' actual tests, so it covers the stale-check refusal, not semantic conflicts in general.

To try it: bun test test/budget-raised.test.ts test/landsim.test.ts.

Approved

The check limit is back to 60 min and the three tests that read it agree, which is what the issue asked. No src/web change, so there is no phone or keyboard surface to judge. No new dependency and no security surface. Two things I would do next: (1) the new landsim scenario limitRaisedUnderCheck uses src/budgets.ts, but the real file is src/kernel/budgets.ts. Fix the path so the scenario reads true to the incident. (2) The scenario mostly re-asserts the existing check_stale refusal that ruleAfterCheck already covers, so it would likely pass without this change. Have it run the two pulls tests, or fold it into #12656, so it catches a semantic conflict and not just a stale check.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs