On Mon, Jan 25, 2010 at 08:23:22PM +0100, Stefan Weil wrote: > Michael S. Tsirkin schrieb: > > On Mon, Jan 25, 2010 at 06:32:06PM +0100, Stefan Weil wrote: > >> Michael S. Tsirkin schrieb: > >>> On Sun, Jan 24, 2010 at 09:23:41PM +0000, Herve Poussineau wrote: > >>>> Replace %lld occurrences by PRId64. > >>> This is wrong. > >>> long long values should be printed with %lld. > >>> size_t - with %zd. PRId64 is for int64_t. > >>> > >> size_t => %zu, ssize_t => %zd might be better. > >> > >> And none of them works on win32, so using them > >> there can result in a crash: > >> > >> size_t st = 4711; > >> fprintf(stderr, "st=%zu, %s\n", st, "test"); > >> > >> printf functions on win32 don't know %z. > >> They run > >> > >> fprintf(stderr, "st=zu, %s\n", st, "test"); > >> > >> which results in an memory access fault when printf > >> wants to read the memory at address 0x4711. > >> > >> Regards, > >> Stefan Weil > > > > Let's just implement a compliant printf? > > Or format the harddisk and install linux? > Maybe that would be the better option :-) > > Of course you can add a printf to qemu, or to mingw32. > No need to implement it - there are lots of good free > implementations. > > The mingw developers are aware of the problem > (http://www.mail-archive.com/mingw-w64-pub...@lists.sourceforge.net/msg00416.html). > > If there is an easy solution, they will fix the problem. > > I don't think the problem can be fixed easy: > there is not only printf but a lot of functions which use > format strings. They are implemented in msvcrt.dll. > Replacing single functions in a dll is difficult. > Telling code which printf in which dll is the correct > one is difficult, too. > > There are easy solutions for QEMU: type cast > size_t values to unsigned or uint32_t in printf > or use a new macro (for example PRIsize) in > format strings. That macro would be different for > mingw32 and standard conforming systems.
The link above suggests adding -D__USE_MINGW_ANSI_STDIO=1 why don't we do just do this with mingw? People should also build with -Werror, then it would be a build error not a crash. -- MST