jcsxky wrote:
> So, looked into this more. You are correct that `Function` needs not be
> dependent because we only ever show the name of the function unqualified.
> However, `FunctionSig` does. Here is an example of what we are trying to
> preserve https://godbolt.org/z/M3eTa8nn5 (the clang 17 behavior was incorrect)
>
> I apologize for not realizing earlier that your change was actually sensible
> in the case of `Function`
Never mind! After looking into the code a little bit more, `FunctionSig` indeed
needs to be dependent.
```cpp
template<int N>
constexpr int g() {
return N;
}
template <typename T>
struct S {
void f(auto) {
constexpr const char* F = __builtin_FUNCSIG();
static_assert(__builtin_strlen(F)==g<__builtin_strlen(F)>());
}
};
void f(){
S<int>{}.f(0);
}
```
However, it still crash when we changed from `__builtin_FUNCTION` to
`__builtin_FUNCSIG` in the testcase from
https://github.com/llvm/llvm-project/issues/80210. The underlying issue is that
`CurContext` is `FunctionTemplateDecl` which is dependent and we used it as
`ParentContext` of `SourceLocExpr` when transforming `decltype` of function
`construct_at`. This makes transformed return type of `construct_at` dependent
which is not what we want. I will close this patch since it doesn't resolve
the underlying issue. Thanks for your time for the review! @cor3ntin
https://github.com/llvm/llvm-project/pull/94942
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
