On Sun, Aug 31, 2014 at 12:00:13PM +0200, wm4 wrote: > On Sun, 31 Aug 2014 11:50:36 +0200 > Reimar Döffinger <reimar.doeffin...@gmx.de> wrote: > > > On Sun, Aug 31, 2014 at 11:39:16AM +0200, wm4 wrote: > > > I got the semantics of avio_read wrong: short reads on EOF are in > > > fact allowed and don't return an error. Hopefully this is correct now... > > > > What I _really_ dislike about these approaches is that we hard-code EOF > > behaviour in every single demuxer. > > So some will return partial packets some won't. > > And if we wanted to make that behaviour configurable we'd have to add > > options to every single one instead of changing the code in one single > > place. > > Not to mention all the other optimizations, just look how much e.g. > > av_get_packet actually does! > > Why do you absolutely want to handle this manually via avio_read? > > You can just use av_get_packet + av_append_packet, then you only need > > to check the size once, and you also don't have to set "pkt->pos" > > manually and you still don't have to do seekback. > > I wasn't aware of av_get/append_packet() (sorry if it was mentioned > before and I ignored it)
I mentioned it but was too lazy to find out the exact function names. So I'll take part of the blame if you missed it :) > - it looks nice. I'll try to use it and send a > patch again later. It's a bit not-nice that it reallocs the packet, but > there's probably no way it matters in this case at all. I agree. _If_ it mattered at some point it would be possible to have a variant that specifies an initial allocation size - for many cases it would be possible to reasonably estimate the final packet size. Though especially for this case that would be just overkill anyway and I think going for "as simple as possible" is best. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
