alexfh wrote: > Is the proposed plan of action marking the functions `[[noreturn]]` ? As per > [#145166 > (comment)](https://github.com/llvm/llvm-project/pull/145166#issuecomment-3004769089) > , marking the functions ``[[noreturn]]` was discouraged.
There are two problems here: 1. the problem demonstrated by the example in https://gcc.godbolt.org/z/bv1ac475d, i.e. that the `InferredNoReturnAttr` is attached to the explicit specialization of a function that does actually return (I think, it's not controversial that this is an incorrect behavior and should be fixed). 2. the general approach with deriving an implicit noreturn attribute from a function definition - this is debatable, and @erichkeane seems to agree with your approach, but I believe cross-function diagnostics in general have a potential for inconsistent behavior, e.g. when moving functions across translation units. I'd prefer to rely on explicitly stated noreturn annotations rather than deriving them from function definitions. IMO `[[noreturn]]` is a part of the function API, since it affects how callers can use it. I'd like @erichkeane and @AaronBallman to chime in on the second part - maybe this should be reconsidered. IMO, the motivating issue for this PR should be solved in the user code rather than in the compiler (https://github.com/llvm/llvm-project/issues/144952#issuecomment-3100190298). Now, getting back to the issue # 1: if you have a trivial fix for this, please go ahead and implement it (assuming the decision on # 2 is to leave this feature in Clang). If the fix is not obvious, I'd suggest to revert the commit and re-send it for review together with the fix. https://github.com/llvm/llvm-project/pull/145166 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
