[PATCH] D44883: [Sema] Extend -Wself-assign and -Wself-assign-field to warn on overloaded self-assignment (classes)

[Arthur O'Dwyer via Phabricator via cfe-commits]

Quuxplusone added a comment.

> The warning is fundamentally about dataflow, but this doesn't apply to unevaluated expressions. There are plenty of cases where a user might want to ask if assignment is well formed on noexcept using only one variable. For example:
> 
>   template <class T> void foo(T& value) noexcept(noexcept(value = value)) { /* perform an assignment somewhere */ }

My impression is that this case will not trigger the warning because the variable `value` is dependently typed.
I agree there should be a regression test for this exact case just to make sure of that.

@EricWF, is it important IYO that this warning not trigger in unevaluated contexts even for non-dependently-typed variables?
This is the case that seems to be coming up in practice in libc++ tests, but is hard to reason about because it's "only" deliberately contrived test code.

  auto foo(std::exception& value)
      noexcept(noexcept(value = value)) // ok to diagnose?
      -> decltype(value = value)  // ok to diagnose?
  {
      static_assert(noexcept(value = value));  // ok to diagnose?
  }


Repository:
  rC Clang

https://reviews.llvm.org/D44883