[PATCH] D146764: [clang] Make predefined expressions string literals under -fms-extensions

[Aaron Ballman via Phabricator via cfe-commits]

aaron.ballman added inline comments.


================
Comment at: clang/lib/Sema/SemaExpr.cpp:3586-3591
+  // MSVC treats all predefined expressions as string literals rather than char
+  // arrays.
+  if (LangOpts.MicrosoftExt)
+    return SL;
+
   return PredefinedExpr::Create(Context, Loc, ResTy, IK, SL);
----------------
aeubanks wrote:
> aaron.ballman wrote:
> > This is incorrect -- these are still predefined expressions even if they're a string literal. Otherwise, we lose AST fidelity and things like the `predefinedExpr()` AST matcher don't work (https://github.com/llvm/llvm-project/blob/main/clang/include/clang/ASTMatchers/ASTMatchers.h#L2697), and `-ast-print` will print the wrong thing (https://github.com/llvm/llvm-project/blob/main/clang/lib/AST/StmtPrinter.cpp#L1248).
> how would you structure this? an almost identical `PredefinedExpr` class that also subclasses `StringLiteral`? or special case all the places that check for `StringLiteral` to also check for `PredefinedExpr` (the case I was mostly looking at was initializing a char array with predefined expressions but presumably there are more places that matter)
We can't inherit from `StringLiteral` (it's a `final` class because it has `TrailingObjects`), so that might be the only viable approach. However, there's a fair number of places in the compiler where we care if something is or isn't a `StringLiteral`, so I can see why we'd want to avoid that if possible.

Oye, but it seems that even MSVC is not consistent with whether the predefined expression is a string literal or not: https://godbolt.org/z/jzY8cz79d


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D146764/new/

https://reviews.llvm.org/D146764