incorrect overload resolution with deducing this

[Tsche]

Hey,
@Waffl3x mentioned a possible bug in clang earlier and asked me to report it.

template<typename T>
concept AlwaysTrue = true;

struct S {
  int f(AlwaysTrue auto) { return 1; }; // f1
  int f(this S&&, auto) { return 2; };  // f2
};

int main() {
  return S{}.f(0);
}

https://clang.godbolt.org/z/x516PWsvd

According to [basic.scope.scope]/3 f1 and f2 should have corresponding signatures. Considering [over.match.funcs.general]/5 I believe f1 should be called here since it is more constrained. Clang however calls f2.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: None (Tsche)

Hey, @Waffl3x mentioned a possible bug in clang earlier and asked me to report it.

template<typename T>
concept AlwaysTrue = true;

struct S {
  int f(AlwaysTrue auto) { return 1; }; // f1
  int f(this S&&, auto) { return 2; };  // f2
};

int main() {
  return S{}.f(0);
}

https://clang.godbolt.org/z/x516PWsvd

According to [[basic.scope.scope]/3](https://eel.is/c++draft/basic.scope.scope#3) f1 and f2 should have corresponding signatures. Considering [[over.match.funcs.general]/5](https://eel.is/c++draft/over.match.funcs.general#5.sentence-3) I believe f1 should be called here since it is more constrained. Clang however calls f2.

[shafik]

I believe this is cwg2789: Overload Resolution with implicit and explicit object member functions which is not available yet but should be soon. I believe clang is wrong here.

[Waffl3x]

I believe this is cwg2789: Overload Resolution with implicit and explicit object member functions which is not available yet but should be soon. I believe clang is wrong here.

https://cplusplus.github.io/CWG/issues/2789.html

No, that isn't relevant here, the bullet being changed refers to non-template functions, these are templates.
"F1 and F2 are non-template functions"
The linked issue fixes constraints on non-template member functions, which funny enough already works in clang.
https://godbolt.org/z/fWveeqTqb