[clang] clang: Relax LangOpts checks when lexing quoted numbers during preprocessing (PR #95798)

[via cfe-commits]

================
@@ -2068,7 +2068,8 @@ bool Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
   }
 
   // If we have a digit separator, continue.
-  if (C == '\'' && (LangOpts.CPlusPlus14 || LangOpts.C23)) {
+  if (C == '\'' &&
+      (LangOpts.CPlusPlus14 || LangOpts.C23 || ParsingPreprocessorDirective)) {
----------------
Sirraide wrote:

> Sure, but we don't have to expose that language extension to users, it can remain internal to the compiler.

Yeah, that’s fine. What I meant was if we were to just use `ParsingPreprocessorDirective`, we would be exposing it to users, so that’s probably not the way to go imo.

> The scanner works in two steps:
> 
> 1. The virtual file system "minimizes" source files into sequence of tokens that are relevant for dependency discovery. Your code fragment would be transformed into something like `[if, ident, gte, ident, then, include, string, else, include, string, endif]`.
> 2. The scanner instance takes those tokens and the TU `LanguageOptions` and runs full preprocessor on those tokens. This means that the condition should evaluate correctly.

I see; in that case being more permissive in what we accept during step 1 makes perfect sense imo. I don’t think there are *that* many language features that affect this step either since that’s basically just preprocessor directives from what I gather, so adding language options for whatever features we need to enable for this seems reasonable.

https://github.com/llvm/llvm-project/pull/95798