Commit Graph

3 Commits

Author SHA1 Message Date
pku-xht
2471e2b2bb fix review finding: exitsWithin cleans up its listener and timer on both arms
Each timed-out wait used to leave the once('exit') listener from its inner
waitForExit attached to the child; the dispose ladder accumulates at most a
couple, but in a shared library a caller polling exitsWithin in a loop would
pile listeners onto one child (MaxListenersExceededWarning at 11) and retain
their closures. The race now owns its wiring: the timeout arm removes the
exit listener, the exit arm clears the (still unref'ed) grace timer, and an
already-exited child short-circuits true without attaching anything. Tests
pin listenerCount('exit') === 0 after every outcome.
2026-07-09 10:42:34 +08:00
pku-xht
7ccf31a59b fix review finding: root-portable rm-failure injection in the config-dir test
The best-effort-remove test provoked a real EACCES via a chmod-000 subtree,
which only fails for unprivileged users — under root, recursive rm ignores
permission bits, deleting the subtree: the existsSync assertion goes red and
the swallow branch loses coverage, failing the per-file gate. The rejection
is now injected deterministically at the node:fs/promises boundary (rm
wrapped with a real-passthrough vi.fn; one test queues a single rejection),
the fs-failure boundary being exactly the non-deterministic seam the testing
policy sanctions mocking. Everything else in the suite stays on the real
filesystem, and the swallow contract stays error-kind agnostic.
2026-07-09 10:20:42 +08:00
pku-xht
cdd1ce2ad4 feat(subagent): extract dsh-subagent-process shared out-of-process machinery
The credential env scrub (SENSITIVE_ENV_PATTERN/buildChildEnv), the
spawn-failure capture (spawnFailure), the child-exit waits
(waitForExit/exitsWithin), and the stdin-EOF -> SIGTERM -> SIGKILL dispose
ladder move out of subagent-acp into a new pure library package (the
subagent-inprocess shape), with the ladder taking its two grace periods as
parameters — defaults stay in the plugin Config. New isolated-config-dir
helpers (mkdtemp create, best-effort remove; a pinned dir is never removed)
land alongside for the CLAUDE_CONFIG_DIR / CODEX_HOME redirection the RFC
names.

The ACP backend migrates onto the library with no semantic change: its suite
passes with import-path edits only. bash-local keeps its sibling copy, per
the RFC's blast-radius call.

RFC: docs/rfc/proposed/feature/2026-07-07-claude-code-and-codex-subagent-backends.md
2026-07-08 17:09:53 +08:00