Windows CI refused legitimate temp directories for the same reason macOS did,
one commit earlier:
cleanup() refused to remove a path outside os.tmpdir():
C:\Users\runneradmin\AppData\Local\Temp\bug-3491-7zisup
GitHub's windows runners report os.tmpdir() in the 8.3 SHORT form
(C:\Users\RUNNER~1\AppData\Local\Temp) while callers hold the expanded LONG
form. fs.realpathSync() does not reliably expand 8.3 names; only
fs.realpathSync.native() does. Casing can differ independently (C:\ vs c:\).
Two platform-specific breakages from the same check is a sign the check was
written against one spelling rather than the concept, so this stops patching
symptoms. tmpRootCandidates() collects every variant derivable from
os.tmpdir() — resolved, realpath'd, and native-realpath'd — each probe isolated
in its own try/catch so an unavailable variant contributes nothing instead of
crashing teardown, then deduped. A target is accepted under any of them, and
the comparison folds case on win32 only, where casing genuinely varies. The
error message still prints the original-case path.
On this machine three variants collapse to two: /var/folders/.../T and
/private/var/folders/.../T. Both spellings of a real temp dir are accepted and
removed.
The Linux matrix passed every one of these broken states — 30544 and then
30545, both lanes green — because /tmp has neither symlink indirection nor
short names. The platform CI shards are the only thing that has caught any of
it, which is worth stating plainly given what this branch is about.
Refs #3057
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The guard added in the previous commit refused legitimate temp directories on
macOS. os.tmpdir() returns a path under /var/folders/..., /var is a symlink to
/private/var, and path.resolve() does not resolve symlinks. So a caller that
passed the realpath'd form — via fs.realpathSync(), or via process.cwd() after
chdir-ing into a temp dir, which returns the resolved path — produced
/private/var/folders/... and failed a check written against /var/folders/...
Measured on this machine before the fix:
mkdtemp path /var/folders/.../probe-XXX allowed
fs.realpathSync of the same dir /private/var/folders/.../probe-XXX REFUSED
process.cwd() after chdir to it /private/var/folders/.../probe-XXX REFUSED
The accepted-roots set is now built from both the resolved and realpath'd forms
of os.tmpdir(), deduped — on Linux they are identical, so the set collapses to
one entry and nothing changes there. realpathSync is wrapped because it throws
if the root is momentarily missing, and a safety check must not become a new
crash. The target itself is deliberately NOT realpath'd: cleanup() is called on
already-deleted directories, where realpathSync raises ENOENT.
The roots are computed per call rather than hoisted to module scope, because two
test files override TMPDIR and a hoisted value would go stale for them.
Worth naming: the remote matrix runs linux-node22 and linux-node24 only, and it
passed 30544/30544 on the broken commit. os.tmpdir() is /tmp on Linux with no
symlink indirection, so that matrix could not have caught this at any sample
size. The regression test added here branches on whether realpath differs from
the original path, so it exercises the real case on macOS and stays meaningful
rather than vacuous on Linux.
Refs #3057
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
cleanup() computed `isTmpPath` — the exact predicate for "is this path safe to
delete" — and consulted it only inside the catch block, to classify a transient
Windows error. The rmSync above it ran unconditionally against whatever path it
was handed, with `recursive: true, force: true`. The guard existed and the
destructive call never asked it.
The chdir at the top of the function makes the failure mode worse rather than
better: a wrong target first moves the process out of the tree, then deletes it.
So the predicate moves above both, and an out-of-tmpdir path is refused before
either can run. The refusal throws and names the path; returning quietly would
reproduce the fail-open shape this wave exists to remove.
The catch keeps its `&& isTmpPath` term. It is now always true, but it states
the condition the swallow depends on rather than inheriting it from a check
twenty lines up, and it stays correct if the guard is ever relaxed.
All 300+ call sites resolve under os.tmpdir() today, including the two files
that override TMPDIR — both create their override root through the real
os.tmpdir() first — so nothing legitimate is refused.
The regression test targets tests/ itself: a real directory that must never be
deleted, so nothing is created and nothing needs tearing down. It asserts the
throw names the path, that cwd is unchanged (the chdir hazard), and that a known
file inside still exists — proving the directory was not emptied rather than
merely still present.
Whether tests/ is outside os.tmpdir() is environment-dependent: os.tmpdir() is
/tmp on Linux, so a checkout under /tmp would put it inside, and there a correct
guard would delete this directory rather than refuse it. The test asserts that
precondition before calling cleanup, so that environment fails loudly instead of
destructively.
Refs #3057
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Pay down pre-existing error→warn lint debt. Removes dead imports/vars, unused functions, redundant regex/string escapes, and stale eslint-disable directives; converts unused `catch (_e)` to optional catch binding (src/*.cts).
No behavior change. Lint 345→125 warnings (0 errors); deferred categories (n/no-process-exit, test-sleeps, control-regex) tracked in #732 for follow-up. Full test suite green (0 failures); code-review verified all removals unused and all escape fixes semantics-preserving.
Closes#732
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The windows-test-parity ratchet greps test source for fs.rmSync-without-
maxRetries (and six other Windows-portability anti-patterns), failing when an
integer offender COUNT exceeds a frozen baseline (rmSync: 95). A count ratchet
is a Goodhart metric: fixing one offender and adding another keeps the count
constant, so a new defect slips through green. Replace it — and every other
count ratchet in the repo — with a layered, masking-proof design.
Behavioral seam test
- tests/helpers-cleanup.test.cjs proves helpers.cleanup() carries the Windows
EBUSY retry budget. cleanup() delegates retries to Node's fs.rmSync via
maxRetries (it owns no loop), so the test asserts the option contract
(recursive/force/maxRetries>0/retryDelay>0) + real-FS removal + the cwd-guard,
rather than a loop that does not exist. The EBUSY risk is now tested ONCE at
the helper, not approximated textually at every call site.
Write-time ESLint rule (AST-accurate, replaces the grep)
- eslint-rules/no-raw-rmsync-in-tests.cjs (error in tests/**/*.test.cjs) bans
raw fs.rmSync, steering to cleanup(). Catches member, computed (fs['rmSync']),
destructured and aliased forms; escape hatch is inline
`// eslint-disable-next-line local/no-raw-rmsync-in-tests -- <reason>` only.
- Migrated 336 raw fs.rmSync teardown calls across ~116 test files to cleanup().
~18 genuinely load-bearing sites (mid-test SUT/fault-injection removals,
error-swallowing or name-colliding local teardown helpers) keep the raw call
with an inline eslint-disable + reason.
Shared anti-ratchet primitive
- scripts/lib/allowlist-ratchet.cjs:
- assertWithinAllowlist: fails on NOVEL ids (new offender introduced) AND on
STALE ids (a known offender was fixed but not pruned) — identity, not count,
and a ratchet DOWN toward zero.
- assertTightCeiling: a size/length budget whose ceiling must stay within a
grace band of the high-water mark, so budgets may only tighten, never creep.
Ratchets converted onto the primitive
- windows-test-parity-guard.test.cjs: rmSync rule deleted (now ESLint-enforced);
the remaining six patterns moved from integer baselines to named-set
allowlists with ratchet-down.
- scripts/lint-test-file-count.{cjs,allowlist.json}: per-module integer counts →
named filename sets (closes the swap-a-file-keep-the-count blind spot); a
module dropping under cap now FAILS to force pruning its allowlist entry.
- enh-2790 skill-count `<= 63` → named skill allowlist (ratchets toward ~58).
Size budgets hardened (tighten-only)
- agent-size / workflow-size / feat-3039 help-tiered: ceilings lowered to the
current high-water mark and an assertTightCeiling anti-creep check added per
tier. Fixed external-contract limits (description ≤100 chars, agent ≤100 KB)
are intentionally left as-is — they are not grandfathered creeping budgets.
No user-facing behavior change (tests + tooling only); no USER_FACING_PREFIXES
touched, so no changeset fragment is required.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
