[PATCH] D14811: [tsan] Handle dispatch_once on OS X

[Dmitry Vyukov via llvm-commits]

dvyukov added a comment.

LGTM with nits


================
Comment at: lib/tsan/rtl/tsan_libdispatch_mac.cc:30
@@ +29,3 @@
+// GCD's dispatch_once implementation contains a fast path that contains an
+// intentionally racy read and it's inlined into user's code. Furthermore, this
+// fast path doesn't establish a proper happens-before relations between the
----------------
I can't imagine how undefined behavior can be intentional. Also, whether it is intentional or not is irrelevant here. So just drop "intentionally".

================
Comment at: lib/tsan/rtl/tsan_libdispatch_mac.cc:45
@@ +44,3 @@
+  SCOPED_TSAN_INTERCEPTOR(dispatch_once, predicate, block);
+  atomic_uint32_t *a = static_cast<atomic_uint32_t*>((void *)predicate);
+  u32 v = atomic_load(a, memory_order_acquire);
----------------
C++ and C cast chained look weird. Do either just C cast, or just reinterpret_cast.

================
Comment at: lib/tsan/rtl/tsan_libdispatch_mac.cc:65
@@ +64,3 @@
+  SCOPED_TSAN_INTERCEPTOR(dispatch_once_f, predicate, context, function);
+  WRAP(dispatch_once)(predicate, ^{ function(context); });
+}
----------------
For my education: how do these blocks work? Are they allocated with malloc, and then atomically reference-counted?


http://reviews.llvm.org/D14811