[clang] [C23] Implement WG14 N3037 (PR #132939)

[Aaron Ballman via cfe-commits]

AaronBallman wrote:

> However, we do have some prior art -- we're not inventing the whole idea of decl-merging from scratch right now. So maybe we can follow that prior art? For example, top-level attributes on a struct's definition can be inherited from a declaration in the same scope. It seems plausible that the same would be true for redefinition, and then maybe we'd also extend that same declaration-attribute-merging to fields within the struct.

Well, yes and no. We have prior art of doing this when we are checking valid redeclaration, but this is also about type merging as much as declaration merging. And with redefinitions, you run into problems you wouldn't be able to run into with redeclarations. e.g.,
```
struct S {
  char c;
  int x;
};

void func(struct S s) { ... } // Expects an unpacked structure

struct S {
  [[gnu::packed]] char c;
  [[gnu::packed]] int x;
};

int main() {
  struct S s = {};
  func(s); // Passes a packed structure
}
```
You couldn't run into this situation with redeclarations because you previously could not redeclare the structure fields.

The logic would sort of work out for standard attributes, at least by happenstance. There could be observable differences with `[[reproducible]]` and `[[unsequenced]]` but those are much more contrived situations given the nature of the attributes.

> So...yeah...

Yeah, there's a fair amount of moving parts. This is why I want to go with the most conservative approach of rejecting unless the attributes are the same. At some point, we'll get around to being able to tablegen more of this logic (likely leaning on the existing tablegen for some things like inheritable declaration attributes, and introducing new concepts like inheritable type attributes) and we can then relax the restrictions. But if we release in an unrestricted form, users *will* begin to rely on invalid code being accepted and we'll be in a pickle.

Curiously, GCC doesn't merge the standard attributes either, it seems to do last-one-wins: https://godbolt.org/z/j3W7ej5Kq

https://github.com/llvm/llvm-project/pull/132939