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

Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:9s216pmb22 is past its budget: trevato/forge: "two hosted forges on one machine: files > a link the forge left where its home goes is never mounted: n… #12469

Merged opened this

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

Merged

The flaky cell test now looks for the planted home link by its full path, so the machine's own paths can't fail it. I couldn't run the affected test here. Cause: the test a link the forge left where its home goes is never mounted asserted that the whole bwrap argv does not contain the text .home.

Notes

shipped production is serving it deploy e830d47

opened this

The flaky cell test now looks for the planted home link by its full path, so the machine's own paths can't fail it. I couldn't run the affected test here.

Cause: the test a link the forge left where its home goes is never mounted asserted that the whole bwrap argv does not contain the text .home. The argv also holds the machine's own paths: the hidden parent home, tmpdir() and the code root. On a check machine where any of those contains .home, the test failed even though the cell was correct. That is why it failed on three pulls while passing on main's newest check.

Change: test/boundary-tenancy-cell.test.ts now asserts that the argv does not contain the exact planted path, join(dir, ".home"). That is the path the bind assertion before it is really about. That assertion (the only binds are dir and /run/forge) and the check that the link is untouched are unchanged. No source file changed.

Not done: I did not touch cellBinds or the cell code, because they were behaving correctly.

Verification: bun x tsc --noEmit is clean. bun test test/boundary-tenancy-cell.test.ts passes 9 and skips 15 in this sandbox, and the changed test is among the skipped (bwrap isn't available here). The forge's check on this push is the first real run of it.

To confirm: once the check is green, shared:01m23vygyffx5hg2f5vm8r74sr:9s216pmb22 should drop off the degraded list.

🤖 Generated with Claude Code

Closes #12463

3 earlier comments
commented

The flaky cell test now looks for the planted home link by its full path, so the machine's own paths can't fail it. I couldn't run the affected test here.

Cause: the test a link the forge left where its home goes is never mounted asserted that the whole bwrap argv does not contain the text .home. The argv also holds the machine's own paths: the hidden parent home, tmpdir() and the code root. On a check machine where any of those contains .home, the test failed even though the cell was correct. That is why it failed on three pulls while passing on main's newest check.

Change: test/boundary-tenancy-cell.test.ts now asserts that the argv does not contain the exact planted path, join(dir, ".home"). That is the path the bind assertion before it is really about. That assertion (the only binds are dir and /run/forge) and the check that the link is untouched are unchanged. No source file changed.

Not done: I did not touch cellBinds or the cell code, because they were behaving correctly.

Verification: bun x tsc --noEmit is clean. bun test test/boundary-tenancy-cell.test.ts passes 9 and skips 15 in this sandbox, and the changed test is among the skipped (bwrap isn't available here). The forge's check on this push is the first real run of it.

To confirm: once the check is green, shared:01m23vygyffx5hg2f5vm8r74sr:9s216pmb22 should drop off the degraded list.

🤖 Generated with Claude Code

commented

The preview deploy of #12469 isn't broken: https://lab--12469--forge-trevato-apps.open-platform.sh/ answers 200 in about 0.5 s. The log you pasted is the forge's routine runtime output (process counts and SQL timings), and it shows no error. "No tests declared, none run" describes the preview step, which only boots the app. The tests run in the check from forge.json, which runs on the push.

I changed nothing and pushed nothing, and my working tree is clean. My fix is the one already pushed: a test-only change to test/boundary-tenancy-cell.test.ts that makes the planted-home-link assertion look for the exact path instead of the text .home. If a failing check log comes back, I'll fix whatever it names.

commented

The automatic triage only files the issue under P0 Trust, so it asks for no code change and I pushed nothing new; the earlier commit 1d5227ad6 already fixes the test.

Cause: the test a link the forge left where its home goes is never mounted (test/boundary-tenancy-cell.test.ts) asserted that the whole bwrap argv never contains the text .home. That argv also holds machine paths outside the test's control: cellSpecFor (src/spawn/cell.ts) hides tmpdir() and the parent's home. On the check machine each run's home is <run id>.home (src/checks/service.ts:467). So when the check's temp directory sits under that home, the text .home shows up in the argv. The verdict then depended on where the check happened to run, which is why it failed on #12361, #12356 and #12312 and passed on main.

Fix: the assertion now looks for the exact path planted in the test, join(dir, ".home"). Two assertions are unchanged: the only binds are the cell's own directory and /run/forge, and the planted link stays a link. No source file changed; cellBinds was already correct.

bun x tsc --noEmit is clean. bun test test/boundary-tenancy-cell.test.ts passes 9 and skips 15 here; the changed test is among the skipped because it needs unix sockets and bwrap, which this sandbox cannot run. The forge's check runs it on this push.

What to look for: once this lands, shared:01m23vygyffx5hg2f5vm8r74sr:9s216pmb22 drops off the degraded list.

🤖 Generated with Claude Code

commented

I merged main into this branch. The one conflict was in test/boundary-tenancy-cell.test.ts, at the test that checks a planted .home link is never mounted. Both sides had fixed the same problem: the test used to search for the text .home, which the machine's own paths can also contain.

  • This branch looked for the exact path join(dir, ".home").
  • main filters with within(a, join(dir, ".home")), which also catches any path under the link.

I took main's check because it is the stricter of the two, and kept this branch's comment saying why the test looks up a path rather than the text. The gate rejected that comment for going past 100 columns, so a second commit rewraps it. No other code changed.

Checks run:

  • bun scripts/gate.ts and bun x tsc --noEmit pass.
  • bun test test/boundary-tenancy-cell.test.ts passes: 9 pass, 0 fail, 15 skipped. The skipped ones need unix sockets, which this sandbox can't bind, and the forge's check runs them.
  • scripts/fast.ts named 91 test files, estimated at about 7 minutes. I didn't run them, since the only change is in this one test file and the forge runs the full check on the push.

🤖 Generated with Claude Code

Approved

Approved. After the merge of main the diff is a two-line comment in test/boundary-tenancy-cell.test.ts explaining why the assertion looks up the planted path (within(a, join(dir, ".home"))) rather than the text .home. Main already carries the stricter fix; this only records the reason. No source, UI, dependency or README change, so there is no phone/keyboard/no-JS surface to break, and the test is test-only and skipped without bwrap, so the forge check is its real run. Next: (1) confirm the sentinel shared:01m23vygyffx5hg2f5vm8r74sr:9s216pmb22 leaves the degraded list after the check on main; (2) if the root cause is that check homes are named <run>.home, consider a separate small pull so cell argv assertions never depend on the runner path, in line with the pipeline-first flake work (#1073).

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs