AaronBallman wrote: > Just to confirm: is this the intentional outcome of this patch?
Before C++20, left shift of -1 does not produce a mathematical result that's within the range of representable values for the result type, so it's undefined behavior, and therefore not a core constant expression, and therefore invalid as the initializer for an enumeration constant. So I believe the behavior is expected and correct; you can see the UB diagnostic in other contexts: https://gcc.godbolt.org/z/bqj6xhzzr and GCC agrees with Clang in the enumeration case: https://gcc.godbolt.org/z/qjKzeaYTz > the diagnostic - "expression is not an integral constant expression" - > doesn't sound helpful (how is it not constant or not integral?). And what's > worse, is that it's an error rather than a warning that can be disabled. It has integral constants but it's not an integral constant expression as far as the language is concerned. The note tells you why it's not a valid integral constant expression. As for warning vs error; the standard makes the code ill-formed and we treat that as an error generally speaking. If you can't enable C++20 mode, you can still get the same value in C++17 and earlier with well-formed code: https://gcc.godbolt.org/z/3ra83joqs https://github.com/llvm/llvm-project/pull/70307 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
