AaronBallman wrote: > The validation of %b and %B was done without taking into account the current > language mode, which means that %b and %B could have been used in format > strings before they were supported.
On the one hand, yes, this is correct. On the other hand, a C23-based standard library is not going to know that the host application was compiled in C17 mode and it may very well support these modifiers: https://godbolt.org/z/G7KdeKqEx So the issue is that in C23 mode, we can assume the standard library is C23-compatible, but in pre-C23 modes, we can't make assumptions one way or the other. So I think we don't want to treat the modifiers as being unknown. Instead, I think we should issue a diagnostic along the lines `format specifier '%b' requires a C standard library that is compatible with C23; data argument may not be used by format`. I think this is a different kind of incompatibility than we've had before, so it may require more invasive changes. However, we'll need this for other format specifiers like `%wN` and `%wfN`. https://github.com/llvm/llvm-project/pull/126694 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
