On Sat, Jun 3, 2017 at 2:07 PM, Jan Engelhardt <jeng...@inai.de> wrote: > > On Saturday 2017-06-03 13:34, Hendrik Leppkes wrote: >>On Sat, Jun 3, 2017 at 1:18 PM, Jan Engelhardt <jeng...@inai.de> wrote: >>> >>>The layout of struct AVCodec changed. Four function pointers were >>>inserted in the middle, shifting void (*flush) to a new address. >>>The size of AVCodec also changed, which would have required a full >>>MAJOR bump. >> >>Our ABI is stable in one direction, not in both. [...] >>The same applies to structs. The size of structs is not part of the >>ABI, and as such we are free to grow it to accomodate new fields >>without breaking the ABI. > > Size changes are breakers in both directions. > > You claim your ABI is stable in one direction, which means that a > program built with 3.0 headers should work with a 3.3 library on a > separate system. However, a 3.0 program which invokes > someAVcontext->flush(...) would, under 3.3, invoke someAVc->send_frame > because of the changed layout. So it's not ABI compatible in that one > direction that ffmpeg is supposed to be compile, either.
And if it would do that, it would violate our documented ABI/API, because we have clear documentation stating that those parts of the struct are not part of public ABI/API. It may not be the cleanest thing in the world, but it is what it is, and we have used this in various structs in the past. - Hendrik _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
