ci: cover Valgrind stale-lock false positives

Author: etr

ChangeLog

@@ -1,5 +1,11 @@
 Version 2.0.0
 
+	Fixed macOS CI failure when verifying the pinned libmicrohttpd
+		archive by using the portable `shasum -a 256` interface.
+	Replaced the blocking-start pthread mutex/condvar pair with C++20
+		atomic wait/notify, avoiding a Valgrind 3.22 Helgrind stale
+		lock-address false positive, and suppressed the same third-party
+		detector limitation in libmicrohttpd-owned mutexes.
 	v2.0 release. See RELEASE_NOTES.md for the full v1->v2 porting summary.
 	SOVERSION bumped: shared library now ships as libhttpserver.so.2
 		(Linux) / libhttpserver.2.dylib (Darwin). Parallel-installable

codecov.yml

@@ -5,6 +5,15 @@ coverage:
   precision: 2
   round: down
   range: "70...100"
+  status:
+    project:
+      default:
+        target: auto
+        # Integration tests exercise timing-dependent logging/error paths, so
+        # unrelated lines can move between fully covered and partially covered
+        # between otherwise green runs. Permit that measured noise without
+        # weakening the 70% floor or patch coverage enforcement.
+        threshold: 1%
 
 parsers:
   gcov:

src/detail/daemon_lifecycle.cpp

@@ -28,11 +28,11 @@
 
 #include "httpserver/detail/daemon_lifecycle.hpp"
 
+#include <microhttpd.h>
+
 #include <cstdint>
 #include <vector>
 
-#include <microhttpd.h>
-
 #include "httpserver/webserver.hpp"
 #include "httpserver/create_webserver.hpp"
 #include "httpserver/http_utils.hpp"

src/httpserver/detail/daemon_lifecycle.hpp

@@ -29,11 +29,11 @@
 #ifndef SRC_HTTPSERVER_DETAIL_DAEMON_LIFECYCLE_HPP_
 #define SRC_HTTPSERVER_DETAIL_DAEMON_LIFECYCLE_HPP_
 
+#include <microhttpd.h>
+
 #include <atomic>
 #include <vector>
 
-#include <microhttpd.h>
-
 namespace httpserver {
 
 class webserver;

test/valgrind-helgrind.supp

@@ -70,6 +70,53 @@
    fun:thread_main_handle_connection
 }
 
+# --- Valgrind 3.22 stale lock-identity metadata at reused addresses -------
+# Ubuntu 24.04's Valgrind 3.22 keeps a lock's old mutex/rwlock classification
+# when allocator reuse places a differently typed pthread object at the same
+# address. Current Valgrind source removes that stale record on explicit
+# initialization, but that fix is not in the release used by CI. libmicrohttpd
+# owns and correctly initializes these mutexes; Helgrind nevertheless reports
+# "pthread_mutex_* with a pthread_rwlock_t* argument" when one reuses storage
+# formerly occupied by a std::shared_mutex rwlock. Entries are anchored on the
+# exact third-party MHD functions and cover only Helgrind:Misc, so races in
+# libhttpserver's own locking remain unsuppressed.
+{
+   mhd-valgrind322-stale-lock-new-connection
+   Helgrind:Misc
+   ...
+   fun:new_connection_process_
+}
+{
+   mhd-valgrind322-stale-lock-cleanup-connections
+   Helgrind:Misc
+   ...
+   fun:MHD_cleanup_connections
+}
+{
+   mhd-valgrind322-stale-lock-update-last-activity
+   Helgrind:Misc
+   ...
+   fun:MHD_update_last_activity_
+}
+{
+   mhd-valgrind322-stale-lock-connection-cleanup
+   Helgrind:Misc
+   ...
+   fun:cleanup_connection
+}
+{
+   mhd-valgrind322-stale-lock-close-all-connections
+   Helgrind:Misc
+   ...
+   fun:close_all_connections
+}
+{
+   mhd-valgrind322-stale-lock-stop-daemon
+   Helgrind:Misc
+   ...
+   fun:MHD_stop_daemon
+}
+
 # --- libstdc++ shared_ptr control-block (atomic refcount + teardown) ---------
 # The shared_ptr control block (_Sp_counted_base and its _Sp_counted_ptr_inplace
 # derived) maintains an atomic use-/weak-count (_Lock_policy _S_atomic) that