================
@@ -1751,9 +1751,20 @@ static bool
IsStructurallyEquivalent(StructuralEquivalenceContext &Context,
// fulfill the preceding requirements. ... Otherwise, the structure, union,
// or enumerated types are incompatible.
- if (!NameIsStructurallyEquivalent(*D1, *D2)) {
+ // Note: "the same tag" refers to the identifier for the structure; two
+ // structures without names are not compatible within a TU. In C23, if either
+ // declaration has no name, they're not equivalent. However, the paragraph
+ // after the bulleted list goes on to talk about compatibility of anonymous
+ // structure and union members, so this prohibition only applies to top-level
+ // declarations, not members.
+ if (Context.LangOpts.C23 && (!D1->getIdentifier() || !D2->getIdentifier()) &&
+ (D1->getDeclContext()->isTranslationUnit() ||
----------------
AaronBallman wrote:
> Does lexical decl context come into play here at all?
C doesn't have the notion of out-of-line declarations, so no.
> Also, and this might be my C++-brain firing here, but is this sufficient? I
> know in C++ we typically check `isFileContextDecl` but IIRC that is just
> namespaces...
`isFileContext()` checks for TU and namespaces, but C has no namespaces so only
the TU is needed.
> Also-also-- what about structs who are defined 'inline' (or whatever y'all
> call it)... are those never equal too? Something like defined in a param
> list, or in an enum-initializer, or function scope?
I think existing tests already cover that, such as the one for `func1` and
`func2` in this file. Keep in mind that C is a bit odd when you come from a C++
background:
```
struct S {
struct T {
int x;
} t;
};
struct T wahoo = {}; // Perfectly fine in C, not fine in C++
```
> So I wonder if we want to invert this and check if decl-context is a
> RecordDecl?
I think the logic will work either way, but now I need to add more tests; what
happens if the structures are within a function? :-D
https://github.com/llvm/llvm-project/pull/141783
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
