Hi On Sat, May 9, 2009 at 1:39 AM, Lennart Poettering <lennart at poettering.net> wrote: > On Fri, 08.05.09 00:45, Marc-Andr? Lureau (marcandre.lureau at gmail.com) wrote: > >> From: Marc-Andr? Lureau <marc-andre.lureau at nokia.com> >> >> While trying to keep ABI compatibility, I introduce a bitfield for the >> boolean volume_is_absolute. We can't extend the struct because the >> API is using pa_ext_stream_restore_info data[] directly.. > > >> +#include <pulsecore/macro.h> > > No! pulsecore is private API. pulse/ext-stream-restore.h is public > API. We cannot include this here! > > > pa_bool_t is used internally only. In the public API we only expose > ints. pa_bool_t is defined as _Bool on C99 and int on other > compilers. This discrepancy should not be visible to outside due to > ABI stability issues. That is one of the reason why I wanted to resend. (I didn't see in first place because I have private headers installed, a bit like the Intel guys patch :) > > Please don't use :1 for this. This breaks ABI. Unfortunately, it does :( > > A dirty trick would be to encode this flag in device[strlen(device)+1] > or so and hide that in a macro. But that would break if device is > NULL. > I see, you don't even consider using some of the bits of "int mute". That is understandable. > Hmm, this situation really sucks I must admit. > > I am tempted to suggest introducing ext-stream-restore2.h which > includes a much cleaned up and more extensible interface and copies > the old interface mostly 1:1 except where it doesn't make sense. > Looks like there is no other choice... > If we need to redesign this API then let's make it flexible enough to > cater for the features Colin and I discussed on this list a while > back. Ok, I will look back in archives. Do you have already recommendations to make, like use only array of pointers of structs for a start? thanks! -- Marc-Andr? Lureau
