@erichkeane Opinions?

@erichkeane

I guess my first question is 'why'?

Per the comment you mentioned, it is redundant. Switching to DependentFunctionTemplateSpecializationInfo also results in constructs such as:

template<typename T>
struct A {
  template<>
  void f();
};

being diagnosed prior to instantiation. Class scope explicit specializations currently exist in their own broken little corner of the frontend, so unifying their representation in the AST would be the first step towards properly fixing them.

My second is 'if this is unnecessary, why was it here in the first place?`.

ClassScopeFunctionSpecializationDecl was originally added in this commit to "parse MSVC standard C++ headers, MFC and ATL code." Back then, class scope explicit specializations were a Microsoft extension, but since CWG727 they are valid ISO C++. As for why this was not originally implemented using DependentFunctionTemplateSpecializationInfo, I cannot say.

Ping @erichkeane & @shafik -- I left comments responding to your feedback

I don't have any comments, though someone with knowledge of teh clang-tools-extra stuff would be nice to take a look, as well as someone familiar with the serialization. @AaronBallman .

ALSO, this might need a release note and does this need to be mailed to the 'breaking changes' vendor list since it is significantly going to change the AST?

@AaronBallman

Can you be sure to file an issue or add a test case + FIXME comment so we don't lose track of the needed follow-up work, or are you planning to do that work immediately after this lands?

I will address that once this gets merged

CC @llvm/clang-vendors for awareness of the AST changing to remove a node.

Updated the release notes. I didn't go into exhaustive detail, but I can add more if anyone here deems it necessary @erichkeane @AaronBallman

Do you need someone to land these changes on your behalf?

@AaronBallman I don't have write access, so that would be appreciated 🙂. I was putting off rebasing incase there were more responses to my comments, but I presume that there won't be so I will get it rebased now

✅ With the latest revision this PR passed the C/C++ code formatter.

If Aaron doesn't get to it, feel free to ping me on Monday. Its too late on a Friday for me to be able to merge this for you, but monday morning would be fine for me.