akash-manna-sky wrote: > Your PR removed the assertion. Please provide a clear and widely accepted > reason. > > BTW, was LLM involved in this PR?
You're right, removing the assertion wasn't the way to go — I've updated the PR to keep it. The assertion assumed the address_space attribute is always stored on the declarator chunk being visited. That's what #196982 disproves: an attribute written after the declarator-id stays on the Declarator itself and wraps the outermost type, so the chunk-only search came up empty and the unreachable fired. The fix now searches all the attribute lists a declarator carries (chunk, declarator, decl-spec, declaration), skipping invalid/malformed ones since those never produce a type. With that, the attribute is always findable and the original llvm_unreachable stays as-is. And yes, I did use an AI assistant while working on this. I've reviewed, built, and tested the change myself before push on github. Please have a look @TPPPP72 https://github.com/llvm/llvm-project/pull/216348 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
