On Saturday 22 March 2025 15:35:48 Lasse Collin wrote: > Things I noticed about your patches but I didn't make any changes: > > - With MSVCRT, _stat32 sets timestamps to -1 if the time doesn't fit > into 32-bit time_t. The 32i64 wrappers truncate the timestamps > instead. Maybe this doesn't matter, but I mention it in case you > think it does. With file size it's different anyway: _stat32 uses a > truncated st_size if file is too large.
This is really a problem. Different msvcrt versions behaves differently. And POSIX stat() requires that it returns -1/EOVERFLOW when the file size does not fit into the st_size field. I tested 32-bit Linux stat() and it really behaves in this way. And also msvcr100+ _stat32() and _stat64i32() behaves in this way. So for pre-msvcr100 is needed different POSIX stat() wrapper than for msvcr100+ builds. timestamps behaves across all msvcrt versions in the same way, overflow is always indicated by the -1 value. I will fix POSIX stat wrappers for st_size and also non-POSIX 32i64 wrappers for timestamps to be consistent. > - The new file mingw-w64-crt/include/filetime_to_time64.h isn't listed > in mingw-w64-crt/Makefile.am. It should be if "make dist" needs to > work. > > - Commit message typos: > > * "Additionaly define struct stat64 for LFS" > Additionally > > * "Function _time64 is avaulable since msvcr70.dll." > available > > Thanks! > > -- > Lasse Collin These points should be already fixed (in my worktree version). _______________________________________________ Mingw-w64-public mailing list Mingw-w64-public@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
