[PATCH] D11727: Dtor callback emitted when msan attribute not repressed for this function.

[Evgeniy Stepanov]

eugenis added a comment.

What happens if the attribute is used with the "= default" destructor?


================
Comment at: lib/CodeGen/CGClass.cpp:1483
@@ -1483,1 +1482,3 @@
+  if (CGM.getCodeGenOpts().SanitizeMemoryUseAfterDtor
+      && CurFn->hasFnAttribute(llvm::Attribute::SanitizeMemory))
     EmitDtorSanitizerCallback(*this, Dtor);
----------------
I think it would be cleaner to use CodegenFunction::SanOpts

================
Comment at: test/CodeGenCXX/sanitize-dtor-callback.cpp:22
@@ +21,3 @@
+No_San ns;
+// Repressing the sanitization attribute results in no msan
+// instrumentation of the destructor
----------------
Maybe move this to a separate test file? This one gets cluttered; for example, tail call is a completely orthogonal issue and attribute test should not have to deal with it.

================
Comment at: test/CodeGenCXX/sanitize-dtor-callback.cpp:25
@@ +24,3 @@
+// CHECK: define {{.*}}No_SanD1Ev{{.*}} [[ATTRIBUTE:#[0-9]+]]
+// CHECK-NOT: {{^ *}}call void @__sanitizer_dtor_callback
+// CHECK: ret void
----------------
Remove {{^ }} to make the check stronger.



http://reviews.llvm.org/D11727