[PATCH] D12022: Refactored dtor sanitizing into EHScopeStack

[Evgeniy Stepanov via cfe-commits]

eugenis added inline comments.

================
Comment at: lib/CodeGen/CGCXX.cpp:42-44
@@ -33,1 +41,5 @@
 bool CodeGenModule::TryEmitBaseDestructorAsAlias(const CXXDestructorDecl *D) {
+  // If sanitizing memory to check for use-after-dtor, do not emit as
+  // an alias, unless it has no fields or has only fields with non-trivial
+  // destructors.
+  if (getCodeGenOpts().SanitizeMemoryUseAfterDtor &&
----------------
rsmith wrote:
> I assume the rationale here is that the field and base class destructors will mark their storage as destroyed, so there's nothing else to mark? This may mean that vptrs and padding bytes don't get marked as destroyed, is that OK?
Yes, each (sub-)object poisons its storage in the base destructor, and only does it for fields that don't have their own destructors to avoid double-poisoning.

Poisoning vptr would be interesting. Yes, current implementation fails to do this. Padding - less so, but it would make the tool a little bit more predictable.


http://reviews.llvm.org/D12022