cor3ntin wrote: For the record, I am fairly opposed to a warning on normal use of a standard function.
I'm not very concerned about users using different target flags across TUs because this is already a massive foot gun and I do not buy that we are making things materially worse. Instead, by having opinionated on-by-default warnings we encourage users to not care or proactively distrust our warnings. Users will use that functions in headers as a lot of people like to put most of their implementation in headers. Don't get me wrong, I think diagnosing actual ODR-violation is a worthwhile goal, and as @AaronBallman mentioned there are ways to do that (modules, asan, build system warning on inconsistent flags), but from within a single TU we don't have a way to detect actual ODR issues, so we would end up forcing our opinions on users, which I'm not sure is marginally better than not implementing the feature at all. I don't think this feature warrants deviating from our policy that on-by default warnings should be actionable and have a low false-positive rate, nor the complexity to actually produce the warning. If we get users reporting issue with that feature, then we can always emit a deprecation warning. https://github.com/llvm/llvm-project/pull/89446 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
