steakhal wrote: The CFG construction doesn't seem to be the right place to propagate noreturn attributes. It is already fairly complicated and the annotations seem like an orthogonal concern.
Adding deduced attributes should be a separate step in the ASTConsumers pipeline. We take the AST, do some analysis and mutate the AST by adding the deduced attributes to the relevant AST nodes, and only to those where it makes sense. This worked for us downstream for a similar use-cases. This also side-steps patching the CFG construction, because by the time we get there, we are already done with the annotation phase thus, it would see the implict (deduced) `analyzer_noreturn` attributes. We build the CallGraph and iterate it in Post-order to make sure leaf calls are inferred first, thus the callers would get to see the deduced `analyzer_noreturn` attributes too and build from the bottom up. If I had time, I could probably try to upstream this part. The beauty of this approach is that its a separate component, and we would not need to piggyback on a default bool parameter of the `analyzer_noreturn` attribute. https://github.com/llvm/llvm-project/pull/146355 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
