jcsxky wrote:
@mizvekov After looking into the code, I think we should transform qualifier in
TST. Consider the following code:
```cpp
template <typename T>
t1<T>::template t2<T> f1();
void f2() {
f1<bool>();
}
```
`TemplateSpecializationType` of `t2` whose `Template` is a
`QualifiedTemplateName`(`t1<T>::`) will be dependent if we didn't transform the
qualifier since its qualifier and `DeclContext` are both dependent and it
wouldn't be specialized in `Sema::CheckTemplateIdType` due to its name
dependency. Transform the qualifier and rebuild the `TemplateName` as a
`DependentTemplateName` is also consistent with what it is before cpp20(`t2`
is a `DependentTemplateName`) which needs a `typename` keyword(`typename
t1<T>::template t2<T>`).
> The fact that this change somehow affects program semantics is still
> unexpected and unexplained.
The case in `clang/test/SemaCXX/many-template-parameter-lists.cpp` is affected
is because the qualifier didn't be transformed and we can't find the
instantiation of the declaration(`X`) which will not happen after this patch.
This case is also accepted by gcc. The other cases are expected or explained in
the comments.
>If I had to take a guess, there is something wrong in the dependency
>computation for the NNS. Have you checked that?
I didn't find some problems in dependency computation for the NNS.
https://github.com/llvm/llvm-project/pull/94725
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
