[clang] [C2y] Support WG14 N3457, the __COUNTER__ macro (PR #162662)

[via cfe-commits]

================
@@ -226,7 +226,7 @@ class Preprocessor {
       LangOptions::FPEvalMethodKind::FEM_UnsetOnCommandLine;
 
   // Next __COUNTER__ value, starts at 0.
-  unsigned CounterValue = 0;
+  unsigned long CounterValue = 0;
----------------
Sirraide wrote:

> I went with `unsigned long` because the standard requires the value to fit within a `signed long` explicitly; so it was for parity with the standards wording. WDYT?

Looking at what the paper specifies, I think it’s weird no matter which one you choose because the standard both specifies that it should fit in a `signed long` and that the maximum value is `2147483647`, but `2147483647` is *not* the same as `LONG_MAX` e.g. on linux.

Also, ‘fits in a `signed long`’ is relative to whatever a `signed long` happens to be on the target, whereas using `unsigned long` in Clang itself is relative to whatever platform the compiler is running on, so there isn’t really any correlation between the two (other than that both must be >= 32 bits as specified by the standard).

I personally find `uint32_t` easier to read because it’s based on the maximum value of `2147483647`, whereas `unsigned long` just ‘happens to work’ and doesn’t really communicate intent—that said this might also just be my dislike for implicit-width integer types showing (I try to avoid using `long` and friends whenever possible).

https://github.com/llvm/llvm-project/pull/162662