[clang] [Clang] Improve support for expression messages in `static_assert` (PR #73234)

[Aaron Ballman via cfe-commits]

================
@@ -179,18 +179,20 @@ static_assert(false, Message{}); // expected-error {{static assertion failed: He
 }
 
 struct MessageInvalidSize {
-    constexpr auto size(int) const; // expected-note {{candidate function not viable: requires 1 argument, but 0 were provided}}
-    constexpr auto data() const;
+    constexpr unsigned long size(int) const; // expected-note {{'size' declared here}}
+    constexpr const char* data() const;
 };
 struct MessageInvalidData {
-    constexpr auto size() const;
-    constexpr auto data(int) const; // expected-note {{candidate function not viable: requires 1 argument, but 0 were provided}}
+    constexpr unsigned long size() const;
+    constexpr const char* data(int) const; // expected-note {{'data' declared here}}
 };
 
 static_assert(false, MessageInvalidSize{});  // expected-error {{static assertion failed}} \
-                                             // expected-error {{the message in a static assertion must have a 'size()' member function returning an object convertible to 'std::size_t'}}
+                                             // expected-error {{the message in a static assertion must have a 'size()' member function returning an object convertible to 'std::size_t'}} \
----------------
AaronBallman wrote:

The diagnostic is actually slightly worse than it was before -- this *does* have a `size()` member function that returns an object convertible to `std::size_t`, the issue is that it's not viable to call because it has required parameters. The previous notes were a bit better at showing this because some users use `()` to mean "this is the name of a function not a variable" in prose and not "this is a function call accepting no arguments" -- e.g., `memcpy()` vs `errno` Same issue applies to the `data()` diagnostic below.

Can we do anything about this (without a ton of effort)?

https://github.com/llvm/llvm-project/pull/73234