[PATCH] D112765: [AST] injected-class-name is not a redecl, even in template specializations

[Aaron Ballman via Phabricator via cfe-commits]

aaron.ballman accepted this revision.
aaron.ballman added a comment.
This revision is now accepted and ready to land.

Aside from the cast request, this LGTM, thank you!



================
Comment at: clang/lib/Sema/SemaTemplateInstantiateDecl.cpp:1841-1842
+                                   /*DelayTypeCreation=*/IsInjectedClassName);
+  if (IsInjectedClassName)
+    SemaRef.Context.getTypeDeclType(Record, cast<CXXRecordDecl>(Owner));
 
----------------
sammccall wrote:
> hokein wrote:
> > aaron.ballman wrote:
> > > Why is this call needed? (It seems strange to me that we call it and ignore the return value.)
> > my understanding is
> >   - `getTypeDeclType` is more than a getter, it also has a side-effort -- if the type of the passed `Record` is empty, it creates a type, and propagates the type to `Record->TypeForDecl`;
> >   - from the above line, we delay the type creation when `IsInjectedClassName` is true;
> >   - so we need to create a type for the `Record` by invoking `getTypeDeclType`;
> > 
> > might be worth a comment.
> > 
> That's exactly right. The injected-class-name is an independent decl but yields the same RecordType, and getTypeDeclType has a hacky interface so we can force this link.
> (It wasn't needed in the old version of the code, because when it was a redecl getTypeDeclType would yield the same type in any case)
> 
> I should have mentioned this is all derived from the injected-class-name handling in Sema::ActOnStartCXXMemberDeclarations(). It lacks comments, but I'll add a few while here.
Ah, thank you for the explanation! I think a `(void)` on the result will also help make it clear that we intentionally are ignoring that result despite it looking like a getter.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D112765/new/

https://reviews.llvm.org/D112765