[MPlayer-dev-eng] Re: [PATCH] -endpos support in mplayer::3rd try

[Alban Bedel]

On Thu, 1 Dec 2005 02:23:55 +0100
Paul TT <paultt at hackerjournal.it> wrote:

> patch adding that option to mplayer and externalizing parse_end_at
> function and field, this last passage thanx to alban bedel :-)
> 
> Albeu wrote:
> 
> >Adding up the duration of the frames that (should) have been displayed
> >is probably the only way to get a reliable absolute time value in
> >mplayer. The video timestamp can (and do on many dvd) jump around so
> >it's not really usable imho.
> 
> sorry this simply doesn't work. tried it and it's less reliable. if i
> seek forward / backward, it count frames, but if i wanna stop after 12
> seconds it's wrong to play a certain amount of frames, i should indeed
> stop after 12s, which my code does.

As said sh_video->pts is in no way an absolute time value, so you may
very well have a 5H movie where the pts never go above 5 min. In such
(a bit extreme but very well possible) case having -endpos 10:00 would
do nothing and the movie would play 5H. At the very least the doc should
warn that, in mplayer, this option is not reliable with some containers,
in particular DVD.

BTW note that i wrote "Adding up the _duration_ of the frames", this is
because the frame duration can vary and in such case -frames is not
really usable. Having something equivalent to -frames but which take a
time could be usefull imho. No need to know the fps and it would work
reliably on VFR content and lame container.

Both approch seems usefull depending on the situation. The tricky part
would be to find sensible and not too confusing names for the options if
we go for both.

> the problem is still with -ss
> option enabled.
> -ss option doesn't work reliably, i think i'll put my hands on this
> after my patches'll be applied, i can't take care of too much code
> simultaneously-or-how-ever-it's-written....

Having -ss to be exact in all case is strictly impossible. Having it exact
when possible is still a big task. Decoding must start on a keyframe so the
demuxer must seek to one. Look simple, however with some format it is not
possible to do efficiently, so the best you can do is seek somewhere that
should be near the wanted position :(

Now even if you ignore this and assume that the demuxer of sane formats
will seek to the keyframe before the wanted point, you still need some
kind of "decode to /dev/null" which is probably not trivial to implement.

Good luck ;)

> +           // FIXME: add size based support for -endpos
> +           if ( end_at.type == END_AT_TIME && end_at.pos < sh_video->pts )
> +                break;

This is wrong, you must set eof instead.

	Albeu