__is_pointer used both as a trait and an identifier seems inconsistent

[pinskia]

Take:

template<typename _Tp, bool _IsPtr = __is_pointer(_Tp)>
  struct __is_pointer {};

template<typename _Tp>
inline bool t = __is_pointer(_Tp); // is accepted

template<bool>
struct ts{};

template<typename _Tp>
  struct is_pointer : ts<__is_pointer(_Tp)> {}; // is rejected

int tt = t<int>;

I would have expected the variable template t would have been rejected the same way as the template argument for the base class is_pointer being rejected but it is inconsistent here.

This shows up now in libstdc++'s headers (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115497).

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Andrew Pinski (pinskia)

Details

Take: ```

template<typename _Tp, bool _IsPtr = __is_pointer(_Tp)>
struct __is_pointer {};

template<typename _Tp>
inline bool t = __is_pointer(_Tp); // is accepted

template<bool>
struct ts{};

template<typename _Tp>
struct is_pointer : ts<__is_pointer(_Tp)> {}; // is rejected

int tt = t<int>;

I would have expected the variable template t would have been rejected the same way as the template argument for the base class is_pointer being rejected but it is inconsistent here.

This shows up now in libstdc++'s headers (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115497).
</details>

[mrdaybird]

std::replace_if fails to compile because of this(#96278) Maybe check if it works before closing this issue.