Crash with concepts in diagnoseWellFormedUnsatisfiedConstraintExpr

[zyn0217]

The following code hits an assertion from diagnoseWellFormedUnsatisfiedConstraintExpr:

template <typename _Iterator> class normal_iterator {};

/// is_convertible
template <typename From, typename To> struct is_convertible {};

template <typename From, typename To>
inline constexpr bool is_convertible_v = is_convertible<From, To>::value;

template <typename From, typename To>
concept convertible_to = is_convertible_v<From, To>;

template <typename IteratorL, typename IteratorR>
  requires requires(IteratorL __lhs, IteratorR __rhs) {
    { __lhs == __rhs } -> convertible_to<bool>;
  }
constexpr bool operator==(normal_iterator<IteratorL> __lhs,
                          normal_iterator<IteratorR> __rhs) {
  return __lhs.base() == __rhs.base();
}

template <typename T> struct Vector {
  normal_iterator<T> begin() const { return {}; }
  normal_iterator<T> end() const { return {}; }
};

class Object;

void function(Vector<Object *> objects) { objects.begin() == objects.end(); }

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

See clangd/clangd#1726 for analysis.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

[zyn0217]

This appears to be a duplicate of #64172, as my local patch would also resolve it.

[zyn0217]

Proposed fix: https://reviews.llvm.org/D158061

[tru]

@AaronBallman is this something we should fix in 17.x? I see you added as a reviewer on the Phab diff.

[llvmbot]

@llvm/issue-subscribers-c-20

[AaronBallman]

We've been asserting on that code since Clang 10.0: https://clang.godbolt.org/z/Y6drTe37z so it's not a regression. I don't think we need to block 17.x on a fix for this (esp because @erichkeane is still out for a bit longer, so we won't hear from the concepts code owner before putting out the next rc).