On Mon, Oct 17, 2016 at 3:47 PM, Michael Niedermayer
> On Mon, Oct 17, 2016 at 12:01:56PM +0200, wm4 wrote:
>> On Mon, 17 Oct 2016 04:24:40 +0200
>> Michael Niedermayer <mich...@niedermayer.cc> wrote:
>> > On Sun, Oct 16, 2016 at 06:01:54PM -0400, Helmut K. C. Tessarek wrote:
>> > > On 2016-09-27 09:30, Michael Niedermayer wrote:
>> > > > Its long since FFmpeg 3.1, so its time to make 3.2
>> > > > ill branch release/3.2 off master and make 3.2 in maybe about a week or
>> > > > 2 unless something delays it
>> > >
>> > > What happened to 3.2?
>> > the AVFrame.pts field has been changed in master redefining ABI/API
>> > without soname bump.
>> > this complicates the 3.2 release.
>> > ATM we have 3.0 and 3.1 releases with the same sonames as git master
>> > but they are not fully compatible due to the AVFrame.pts change
>> > (on top of that i was a bit more busy with random stuff than i
>> > expected)
>> > the obvious solution is to bump major version of all libs. (this is
>> > not wanted i belive though)
>> > the alternative is to remove the (useless) use of AVFrame.pts from
>> > 3.0 and 3.1 and make new releases
>> > then document in release notes that the 3.2 release cannot co-exist
>> > with 3.0 or 3.1 prior to the last release and assume no other
>> > application use AVFrame.pts
>> > Ill look at making the 3.0 and 3.1 release without AVFrame.pts use in
>> > ffmpeg ASAP. These wont do any harm either way
>> > Maybe someone can write the release notes ?
>> > [...]
>> The change was backwards-compatible, as nobody should have used the pts
>> field for decoding before.
> The field was a normal public and documented field
> Documented in our public doxygen for 3.0 and 3.1
> The AVFrame.pts doxy links to examples, which use the field:
The documentation is hardly very enlightining on any level. Which
time_base does it refer to, even? AVFrames can come from various
sources, and there is various time_bases involved. Or why does it not
document that its basically unused for decoding, and only used for
encoding and filtering?
Not to mention that the field was 99% of the time unset during
decoding, and when it was set (by like 2 video decoders that didnt
know any better), it was actually using the pkt timebase (ie. a clone
ffmpeg-devel mailing list