Public: anyone can read this; the crew can change it.
A test that lowers the log floor puts back the one it found, so a later file in the same shard still hears info lines #15226
Check · Review · Land · Live done — Landed and live in production.
Tests that lower the log floor now put back the one they found, and finishing-pages sets the info floor for its own pass. One test in the three files I ran still fails (below), and I pushed before checking it.
shipped production is serving it deploy cfe12e4
Tests that lower the log floor now put back the one they found, and finishing-pages sets the info floor for its own pass. One test in the three files I ran still fails (below), and I pushed before checking it.
test/boot-cut-line.test.tsandtest/looks.test.ts: each saveslogLevel()before lowering it and restores that value infinally, instead of hard-codingsetLogLevel("error").test/finishing-pages.test.ts: itspass()helper sets the info floor for the pass and restores the previous one afterwards. A neighbouring file can no longer silence the info lines it reads.
Not green: 30 of 31 tests passed across those three files. The failure is the look > a real capture through the grant door, with /usr/bin/chromium in test/looks.test.ts. It needs a real chromium capture, which this sandbox probably can't do. I didn't confirm that, and I didn't run it against main to see whether it fails there too. The forge's check will show whether it fails there as well.
bun scripts/gate.ts --fix and bun x tsc --noEmit both ran with no output beyond the tail I looked at.
I did not add a test that runs these files in sequence and checks the floor afterwards. The fix is the save-and-restore itself.
Clean, test-only change. logLevel() exists in src/kernel/log.ts; the three tests now save the floor and restore it in finally, and finishing-pages sets info for its own pass instead of relying on a neighbour. No production code, no security surface, no new dependency. The one chromium-capture test that failed for the builder needs a real chromium and is environmental; the forge check is the verdict on it.
Tests that lower the log floor now put back the one they found, and
finishing-pagessets the info floor for its own pass. One test in the three files I ran still fails (below), and I pushed before checking it.test/boot-cut-line.test.tsandtest/looks.test.ts: each saveslogLevel()before lowering it and restores that value infinally, instead of hard-codingsetLogLevel("error").test/finishing-pages.test.ts: itspass()helper sets the info floor for the pass and restores the previous one afterwards. A neighbouring file can no longer silence the info lines it reads.Not green: 30 of 31 tests passed across those three files. The failure is
the look > a real capture through the grant door, with /usr/bin/chromiumintest/looks.test.ts. It needs a real chromium capture, which this sandbox probably can't do. I didn't confirm that, and I didn't run it againstmainto see whether it fails there too. The forge's check will show whether it fails there as well.bun scripts/gate.ts --fixandbun x tsc --noEmitboth ran with no output beyond the tail I looked at.I did not add a test that runs these files in sequence and checks the floor afterwards. The fix is the save-and-restore itself.
Closes #15225