Merge TASK-081: fill empty-on-correct-build unit suites; delete pthread detector

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: etr

Makefile.am

@@ -255,29 +255,20 @@ check-install-layout:
 # Cross-reference: keep HEADER_HYGIENE_FORBIDDEN in sync with the
 # #ifdef ladder in test/unit/header_hygiene_test.cpp.
 #
-# TASK-020 caveat (libc++ AND libstdc++ in thread mode): <pthread.h>
-# is intentionally absent from the forbidden list below. Both
-# mainstream STLs unconditionally pull <pthread.h> in from any STL
-# container header (<vector>, <string>, <map>, etc.) when threading
-# is enabled:
-#   - libc++ (Apple's default STL on macOS) routes through
-#     <__thread/support/pthread.h>.
-#   - libstdc++ in thread-enabled mode (which is the default whenever
-#     -D_REENTRANT is set, as configure.ac does) routes through
-#     <bits/gthr-default.h>, which #include <pthread.h> directly.
-# The resulting `# N "...pthread.h"` line markers therefore appear in
-# the preprocessed output even though libhttpserver itself does not
-# include <pthread.h>. The runtime sentinel
-# test/unit/header_hygiene_test.cpp keeps the pthread guards but skips
-# them on both libc++ (_LIBCPP_VERSION) and libstdc++ in thread mode
-# (_GLIBCXX_HAS_GTHREADS), so the guards still fire on STLs that don't
-# route std::thread through pthread (e.g. MSVC's Microsoft STL).
-# pthread.h is intentionally absent from HEADER_HYGIENE_FORBIDDEN below
-# for the same reason: the grep-based check would produce false positives
-# on libc++ and libstdc++ in thread mode. The runtime sentinel in
-# test/unit/header_hygiene_test.cpp enforces the pthread invariant via
-# the #ifndef _LIBCPP_VERSION && !_GLIBCXX_HAS_GTHREADS guard ladder so
-# it still fires on STLs that do not route std::thread through pthread.
+# TASK-081 pthread-detector removal: <pthread.h> is intentionally absent
+# from the forbidden list below, AND the corresponding runtime-sentinel
+# guards in test/unit/header_hygiene_test.cpp have been removed
+# entirely. Both mainstream STLs (libc++ and libstdc++ in thread mode,
+# which is the default whenever -D_REENTRANT is set, as configure.ac
+# does) unconditionally pull <pthread.h> in from any STL container
+# header. The libhttpserver public surface uses STL containers
+# (std::string, std::vector, std::map in http_request.hpp, http_resource.hpp,
+# ip_representation.hpp, ...), so the pthread leak is structural to the
+# C++ standard library on every CI lane libhttpserver supports. The
+# detector could not be satisfied on any supported configuration without
+# rewriting the public surface to drop STL containers, which would
+# break source compatibility. See RELEASE_NOTES.md (Test infrastructure)
+# for the deletion rationale.
 # ---------------------------------------------------------------------------
 
 # NOTE: each entry matches a basename; the grep in check-hygiene anchors with a

RELEASE_NOTES.md

@@ -377,6 +377,23 @@ DR-011 ([specs/architecture/11-decisions/DR-011.md](specs/architecture/11-decisi
   CI signal. Two new CI lanes (`tls-no-cli`, plus a baseline canary)
   assert the SKIP wiring fires (or does not) as configured.
 
+- **`<pthread.h>` runtime-sentinel guard removed (TASK-081).** The
+  pthread leak detector in `test/unit/header_hygiene_test.cpp` (and
+  its companion regex slot in `Makefile.am`'s `HEADER_HYGIENE_FORBIDDEN`)
+  was unable to fire on any CI lane libhttpserver actually runs on. The
+  libhttpserver public surface uses STL container headers
+  (`std::string`, `std::vector`, `std::map`, ...) and both mainstream
+  C++ standard libraries — libc++ via
+  `<__thread/support/pthread.h>` and libstdc++ in thread-enabled mode
+  via `<bits/gthr-default.h>` — unconditionally drag `<pthread.h>` in
+  from those headers. Because the public surface cannot be rewritten
+  to drop STL containers without a source-incompatible break, the
+  pthread guard is structurally unsatisfiable on every supported
+  configuration. The detector was deleted rather than kept as
+  dead code gated on STL-implementation-detection macros. The
+  remaining hygiene sentinels (`<microhttpd.h>`, `<gnutls/gnutls.h>`,
+  `<sys/socket.h>`, `<sys/uio.h>`) continue to run on every CI lane.
+
 ## See also
 
 - [README.md](README.md) — full v2.0 introduction and worked examples.

specs/tasks/M7-v2-cleanup/TASK-081.md

@@ -16,12 +16,12 @@ Several unit suites have near-zero effective coverage on the CI lanes that matte
 Make sure every suite actually exercises code on the build configuration where the feature is *available* (the inverse of what each gate currently does).
 
 **Action Items:**
-- [ ] `webserver_ws_unavailable_test.cpp`: invert the guard so the suite exercises the `HAVE_WEBSOCKET`-off path (or rename + add a paired `webserver_ws_available_test.cpp` for the on-path). Goal: each build flag has a unit suite that runs.
-- [ ] `webserver_dauth_unavailable_test.cpp`: same pattern.
-- [ ] `header_hygiene_test.cpp`: investigate why the pthread leak detector was disabled on both stdlibs. Either fix the detector so it works (preferred — this is the entire purpose of the test) or delete the test with a documented `RELEASE_NOTES.md` entry. Coordinated with TASK-007.
-- [ ] `http_request_operator_stream_test.cpp`: split the credential-redaction tests into two — the `HAVE_BAUTH`-on variant pins redaction-when-set; the `HAVE_BAUTH`-off variant pins that the auth-related fields are absent from the stream.
-- [ ] `body_test.cpp` / `http_response_factories_test.cpp` / `iovec_entry_test.cpp`: write a Windows-shaped variant of the pipe/iovec tests using Windows native equivalents (CreateFileMapping for the fd source, etc.) or document the gap in `test/PORTABILITY.md`.
-- [ ] `webserver_register_ws_smartptr_test.cpp`: same pattern as `webserver_ws_unavailable_test`.
+- [x] `webserver_ws_unavailable_test.cpp`: invert the guard so the suite exercises the `HAVE_WEBSOCKET`-off path (or rename + add a paired `webserver_ws_available_test.cpp` for the on-path). Goal: each build flag has a unit suite that runs.
+- [x] `webserver_dauth_unavailable_test.cpp`: same pattern.
+- [x] `header_hygiene_test.cpp`: investigate why the pthread leak detector was disabled on both stdlibs. Either fix the detector so it works (preferred — this is the entire purpose of the test) or delete the test with a documented `RELEASE_NOTES.md` entry. Coordinated with TASK-007.
+- [x] `http_request_operator_stream_test.cpp`: split the credential-redaction tests into two — the `HAVE_BAUTH`-on variant pins redaction-when-set; the `HAVE_BAUTH`-off variant pins that the auth-related fields are absent from the stream.
+- [x] `body_test.cpp` / `http_response_factories_test.cpp` / `iovec_entry_test.cpp`: write a Windows-shaped variant of the pipe/iovec tests using Windows native equivalents (CreateFileMapping for the fd source, etc.) or document the gap in `test/PORTABILITY.md`.
+- [x] `webserver_register_ws_smartptr_test.cpp`: same pattern as `webserver_ws_unavailable_test`.
 
 **Dependencies:**
 - Blocked by: None
@@ -37,4 +37,4 @@ Make sure every suite actually exercises code on the build configuration where t
 **Related Requirements:** PRD-FLG-REQ-003 (`features()` coverage)
 **Related Decisions:** None new
 
-**Status:** Backlog
+**Status:** Done

specs/tasks/M7-v2-cleanup/_index.md

@@ -45,7 +45,7 @@ TASK-093).
 | TASK-078 | Resolve commented-out CONNECT-method test bodies | HIGH | S | Done |
 | TASK-079 | Drive nonce/opaque state machine in v2 digest-auth integ tests | MED | M | Done |
 | TASK-080 | Tighten threadsafety_stress latency gate back from 100× to 10× | MED | M | Done |
-| TASK-081 | Fill empty-on-correct-build unit suites and re-enable pthread leak detector | MED | M | Backlog |
+| TASK-081 | Fill empty-on-correct-build unit suites and re-enable pthread leak detector | MED | M | Done |
 | TASK-082 | Tighten static-size bounds in `http_resource_test` and `webserver_pimpl_test` | MED | S | Backlog |
 | TASK-083 | Wire real CI gates into benchmarks | MED | M | Backlog |
 | TASK-084 | Re-measure libstdc++/Linux v1 baseline for `get_headers` ns/call | MED | S | Backlog |