[Clang] Check for abstract parameters only when functions are defined.

[cor3ntin]

Summary

The C++ standard allows abstract parameters in deleted functions
and in function declarations

The type of a parameter or the return type for a function definition
shall not be a (possibly cv-qualified) class type that is
incomplete or abstract within the function body
unless the function is deleted.

Fixes https://github.com/llvm/llvm-project/issues/63012

[aaron.ballman]

clang/lib/Sema/SemaDeclCXX.cpp

6089

This doesn't seem quite right -- we skip bodies for various reasons: https://github.com/llvm/llvm-project/blob/main/clang/include/clang/Parse/Parser.h#L413

Which test case needed this change?

clang/test/CXX/class.derived/class.abstract/p3.cpp

38–42

Whole lotta trailing semi colons that should be removed (you should get all of them, not just theses).

76–77

FIXME can be removed now? Do we need a DR test case for this?

clang/test/CXX/drs/dr6xx.cpp

707–708

Is this fixme still accurate? Should we be changing the resolution from partial to full?

clang/test/SemaCXX/abstract.cpp

369

A mildly horrifying test case to consider:

struct S {
  virtual void func() = 0;
};
void S::func() {}

static_assert(__is_abstract(S));

struct T {
  void func(S) = delete;
  void other(S);
  void yet_another(S) {}
};

We currently don't do good things... https://godbolt.org/z/7vEY3Kz9c

[cor3ntin]

clang/test/CXX/drs/dr6xx.cpp

707–708

nah, cf below (L714)

[cor3ntin]

Address Aaron's comments.

Initially these changes were tracked in
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0929r2.html

I fixed the specific github issue but more work is needed to acertain conformance.

[aaron.ballman]

LGTM, though please wait a day or so to land this in case @rjmccall disagrees with my assessment of the Objective-C changes.

clang/test/SemaObjCXX/parameters.mm

14–17

I think this change makes sense for Objective-C as well (this is a declaration and not a definition, at least as I understand things), but pinging @rjmccall just in case I'm wrong.

[rjmccall]

clang/test/SemaObjCXX/parameters.mm

14–17

No, that's right, methods in @interfaces and @protocols are just declarations.

Consistency with the base standard seems like the right way to go here, even those I personally find this rule rather silly. (It's one thing to allow this for deleted methods, but allowing it for declarations?)