Rongsong Shen
d26ff8870d
The buglog was one 27K-line JSON file whose fields covered only symptom,
root cause and fix, as free text: 2% of entries named the code version a
bug was found on, 12% the build profile, 36% a way to reproduce, and 58% of
the file was OpenWolf hook auto-detections ("Missing guard clause",
"Function not marked async" on .md files).
- .claude/skills/buglog/bug-record.md: the record standard. Eight
sections (Summary as the title, Description, Environment, Reproduction;
Root cause, Fix, Verification, References once fixed), each with the
content it must contain -- e.g. Environment needs the OS, the libc the
failing binary loads (from its ELF interpreter: glibc 2.42 in the nix
shell vs 2.40 on the host), toolchain, build profile and commit.
- .wolf/bugs/<id>.md: 500 records migrated from buglog.json; every field
verified to round-trip. They carry legacy: true and an explicit "not
recorded" where the old log had no such field; component derived from
file paths, severity from keywords (else unknown), fix_commits from
hashes in status_note/fix that git resolves (91 of 463 fixed).
- .wolf/buglog-archive.json: the 710 hook auto-detections, verbatim.
- .wolf/buglog.json is now an index generated from the records, keeping
the fields OpenWolf's hooks and `openwolf bug search` read.
- scripts/buglog.py rewritten: new (template + Environment collected from
the machine and --binary), check, status (refuses a status the record
doesn't support; fixed needs Root cause, Fix, Verification and
--commit hashes git resolves), note, recur, set, show, list, triage,
env, index. 29 unittest cases in scripts/tests/test_buglog.py.
- scripts/git-hooks/pre-commit runs `check --staged`: incomplete staged
records, a legacy record hand-changed to fixed without the fix-time
sections, a new record claiming legacy: true, or a stale index refuse
the commit.
- scripts/gitea-sync.py reads the records: issue body = the record's
sections plus linked fix commits (origin remote, GITEA_COMMIT_REPO to
override), and a body hash so an issue's text is refreshed when the
record changes.
- bug-1866 rewritten to the full standard as the worked example.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
3.8 KiB
| id | status | status_updated | status_note | component | severity | timestamp | last_seen | occurrences | files | tags | related | legacy | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| bug-1791 | fixed | 2026-09-24 | controller | unknown | 2026-09-09T08:50:58.000Z | 2026-09-19T09:00:00Z | 1 |
|
|
true |
bug-1791: controller_unit_test: interruptcall.c:142: xcetest_worker_main: Assertion `unlock_rc == 0' failed…
Description
controller_unit_test: interruptcall.c:142: xcetest_worker_main: Assertion `unlock_rc == 0' failed (pthread_mutex_unlock returned EPERM)
Environment
Not recorded (migrated from buglog.json).
Reproduction
Not recorded (migrated from buglog.json).
Root cause
xce-test-cli-ux task 1.5's per-connection lock registry initialized each connection's pthread_mutex_t with PTHREAD_MUTEX_ERRORCHECK, intended to catch double-unlock/never-unlocked regressions. But this design's own ownership-handoff protocol (design.md/research.md's 'per-connection lock' decision) REQUIRES a legitimate cross-thread release on the abandonment path: the calling (REPL) thread acquires the lock via pthread_mutex_trylock(), but on abandonment the still-running detached WORKER thread is the one that must release it later, once fn() finishes -- ownership of 'release the lock' passes to the worker at the same moment ownership of the heap call block does. PTHREAD_MUTEX_ERRORCHECK (and _RECURSIVE) mutexes track a single owning thread and reject pthread_mutex_unlock() from any other thread with EPERM, which fired immediately on the very first same-key-serialization test.
Fix
Switched the per-connection lock registry (xcetest_conn_lock_for() in interruptcall.c) to initialize each mutex with the DEFAULT type (pthread_mutex_init(&node->mutex, NULL)) instead of PTHREAD_MUTEX_ERRORCHECK. glibc default (PTHREAD_MUTEX_NORMAL-equivalent) mutexes do not track a per-thread owner, so a worker thread may validly unlock a mutex the REPL thread locked. Kept the assert(pthread_mutex_unlock(...) == 0) safety-net checks at every release site (computed outside the assert() expression itself, so the unlock call is never skipped under NDEBUG) as a lower-cost guard against genuine corruption, without relying on error-checking-mutex semantics that conflict with the intentional cross-thread handoff. SUPERSEDED 2026-09-09 (task 1.5 remediation round 1, independent review): an external reviewer correctly pointed out that POSIX classifies non-owning-thread pthread_mutex_unlock() as undefined behavior for PTHREAD_MUTEX_NORMAL/DEFAULT too, not just ERRORCHECK/_RECURSIVE -- NORMAL is merely permitted not to detect the violation, not declared well-defined for it. The DEFAULT-mutex fix above was therefore empirically safe on XCE's glibc/NPTL-exclusive Nix build but relied on non-portable, POSIX-UB behavior, under-documented as such. Replaced the pthread_mutex_t per-connection lock with a POSIX semaphore (sem_t, binary, sem_init(&sem, 0, 1)) throughout: sem_trywait() in place of pthread_mutex_trylock() (EAGAIN in place of EBUSY) in the interruptible poll loop, and sem_post() in place of pthread_mutex_unlock() at every release site (both the caller-normal-completion path and the worker-abandoned-cleanup cross-thread path). sem_t has no ownership concept, so this cross-thread release is now fully POSIX-defined with no UB reliance at all, at the same code complexity. All 1285 controller_unit_test tests (including the 3 task-1.5-specific tests) continued to pass unchanged after the migration.
Verification
Not recorded (migrated from buglog.json).
References
Not recorded (migrated from buglog.json).