[clang] [Clang][Parse] Fix assertion when annotating a failed decltype-specifier (PR #211221)

[Aaron Ballman via cfe-commits]

================
@@ -1136,6 +1136,20 @@ void Parser::AnnotateExistingDecltypeSpecifier(const DeclSpec &DS,
   // make sure we have a token we can turn into an annotation token
   if (PP.isBacktrackEnabled()) {
     PP.RevertCachedTokens(1);
+    if (DS.getTypeSpecType() == TST_error && PP.hasCachedTokenLocation()) {
----------------
AaronBallman wrote:

I don't think this fix is correct; we should not have to track cached token locations like this. I did more investigation and it looks like we're several layers deep in hacks.

The issue is that member pointers are getting special handling which includes looking for a `decltype` token, but it doesn't care whether that `decltype` is valid or not; if the token is there, it's assumed to be parsed as a `decltype` specifier: https://github.com/llvm/llvm-project/blob/aab7e0b08d30ddca5858069a4c14c1ea3da042e6/clang/lib/Parse/ParseDecl.cpp#L6493

Adding an extra check for a `(` as the next token resolves the issues and causes no additional test failures.

(FWIW, we are finding that LLMs do a *terrible* job of correctly addressing failed assertions. I see you disclosed LLM tool use, so that's something you should be aware of for future PRs. If it's fixing a failed assertion, assume the LLM will point you in the wrong direction.)

https://github.com/llvm/llvm-project/pull/211221