[PATCH] D46651: [OpenCL] Support new/delete in Sema

[John McCall via Phabricator via cfe-commits]

rjmccall added inline comments.


================
Comment at: lib/Sema/SemaDeclCXX.cpp:13010
+  if (SemaRef.getLangOpts().OpenCLCPlusPlus) {
+    if (auto *PtrTy = ResultType.getTypePtr()->getAs<PointerType>()) {
+      ResultType = RemoveAddressSpaceFromPtr(SemaRef, PtrTy);
----------------
`getTypePtr()` is done automatically by `operator->` on `QualType`.  This can be fixed in multiple places in this patch.


================
Comment at: lib/Sema/SemaExprCXX.cpp:2030
+      }
+    }
 
----------------
Anastasia wrote:
> svenvh wrote:
> > rjmccall wrote:
> > > I think a better interpretation of this rule would be to just error on attempts to use the standard non-placement operator new/delete instead of trying to outlaw the operator declarations.  For example, I don't know why a user-defined non-global operator new would be problematic.
> > Good point, I have raised this with Khronos, so I will hold this off until we have clarification.
> The decision by Khronos is to allow all user defined new/delete operators (even global). I have submitted the change to the spec. The next publishing date is however in July.
Okay.  I agree with your decision in this patch to treat this as a defect in the spec and thus to go ahead and do the right thing now.


https://reviews.llvm.org/D46651