[Clang] Avoid misleading 'conflicting types' diagnostic with no-prototype decls.

[cishida]

Summary

Clang has recently started diagnosing prototype redeclaration errors like rG385e7df33046

This flagged legitimate issues in a codebase but was confusing to resolve because it actually conflicted with a function declaration from a system header and not from the one emitted with "note: ".

This patch updates the error handling to use the canonical declaration's source location instead to avoid misleading errors like the one described.

[cishida]

Restrict using the canonical decl's source location to only when the prototype was inferred from previously seen decl.

[aaron.ballman]

LGTM, thank you for catching this!

clang/lib/Sema/SemaDecl.cpp

3914–3918

I don't think it's safe to assign this new value to Old as that object is what we eventually merge the declaration with (so we'll get the decl chain order wrong, I believe).

[cishida]

clang/lib/Sema/SemaDecl.cpp

3914–3918

Ah, I hadn't realized we still can merge decls together in a fatal error case.

[aaron.ballman]

clang/lib/Sema/SemaDecl.cpp

3914–3918

Oh shoot, good point, we early return here anyway, so the assignment won't matter. So the code LGTM as-is, thank you!