On 19 August 2015 at 19:17, Gerd Hoffmann <kra...@redhat.com> wrote: > On Do, 2015-08-06 at 20:28 +0200, Kővágó, Zoltán wrote: >> Currently the gcc specific version only evaluates the arguments once, >> while the generic version evaluates one argument twice, which can cause >> debugging headaches when an argument has a side effect. > > The answer to that is "don't do that". Do we have macro calls with side > effects in the tree? > >> This patch at least provides consistent behavior between compilers. > > Makes sense. > >> -#else >> #define audio_MIN(a, b) ((a)>(b)?(b):(a)) >> #define audio_MAX(a, b) ((a)<(b)?(b):(a)) >> -#endif > > include/qemu/osdep.h already provides MIN/MAX macros. > > I think we should either define audio_MIN (and audio_MAX) to those, or > simply do s/audio_MIN/MIN/ in audio/*.c
My vote is for the latter. Incidentally we already assume both typeof and statement expr support in our compilers, so we could upgrade our local MIN/MAX implementations to use them if we really needed. (We'd have to rename them though, since the system implementation likely does the eval-twice thing.) A quick grep doesn't show any audio_MIN/MAX which need to avoid multiple-evaluation, though. thanks -- PMM
