| Issue |
133056
|
| Summary |
[libc++] Instantiation of `pair` in class template argument deduction is a bit too eager
|
| Labels |
libc++
|
| Assignees |
|
| Reporter |
frederick-vs-ja
|
The following code snippet is rejected with libc++ ([Godbolt link](https://godbolt.org/z/q1x1E8Tqq)):
```C++
#include <utility>
#include <type_traits>
// note that *+[]{} dereferences a function pointer converted from the lambda _expression_
static_assert(std::is_same_v<decltype(std::pair{0, *+[]{}}), std::pair<int, void(*)()>>);
```
It seems that the related implicit deduction guide synthesized from the constructor (`pair(const T1&, const T2&) -> pair<T1, T2>`) should be excluded by SFINAE mechanisms before instantiating invalid `pair<int, void()>`, and the decaying explicit deduction guide (`pair(T1, T2) -> pair<T1, T2>`) should work.
But libc++ currently makes the instantiation a bit too eager, and then causes hard error. I've roughly verified that this can (and perhaps should) be fixed within the library implementation.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
