[MPlayer-users] -ss sometimes jumping dramatically in some mpg/mpeg2 files

[Heitzso]

For what it is worth, all but one of my 60 or so video files now seek
correctly by looking for that last pts at -1000000.  The one file
that does not work in that case works when I back up to -1500000
and then start looking for the pts.

More elegant solution might be to look for final pts in last 500000, then
if not found, go looking from -1000000 to -500000, then if not found look
from -1500000 to -1000000, etc. 

Spec on these files is they are 4500 video bit rate mpeg2 with empty
vobs created w/ transcode&mplex.  The last 10 to 15 seconds are one
or two static clips with words on them, e.g. "END OF INTERVIEW" type
things.  I'm guessing the static nature of the last 10 or 15 seconds
alters the pts pattern.

> I've been digging through the mplayer code to see why some -ss seeks
> go bonkers in some of my video files, but not in most of them.
>
> The problem arises from my mpg(2) files not having a pts for mplayer
> to find in the last 500,000 bytes of the file.  This causes mplayer to 
> fallback
> to guess-timating where to seek into the file based on the video bit 
> rate.
> But mplex stuck the max bit rate into the video file header (assume, 
> based
> on other emails in transcode list) which is 9000, but that's twice the
> average bit rate and therefore mplayer easily falls off the cliff (so 
> to speak)
> when trying to seek into the latter part of the video file.
>
> I backed up that 500,000 to 1,000,000 and now that class of seek problem
> went away.  This is any easy fix in my case because I know my video 
> files are
> all larger than 1,000,000 bytes.  A more elegant fix would be to 
> insure that
> we haven't pushed into a negative number.  Anyway, the fix should be
> applied to demux_mpg.c, around line 110:
>
> int demux_mpg_open (demuxer_t * demuxer) {
>    stream_t *s = demuxer->stream;
>    off_t pos = stream_tell (s);
>    off_t end_seq_start = demuxer->movi_end - 1000000;    // 500000 is 
> a wild guess
>
> Note this is not patchable, I reformatted the C files to correct
> indenting.  But the above mod is certainly simple enough for
> someone to change by hand.
>
> Note the "500000 is a wild guess".  That was certainly a clue for me.
>
> ANYWAY,
>
> A better code sequence would be to calculate the end_seq_start then
> test to see if it is less than 0, and, if it is, to recalculate as ?? 
> 1% of total
> length, or some other? algorithm.  IF the powers that be would like
> me to mod that one line in an official not-re-formatted version of this
> file and submit a simple patch I will.  Let me know off this list and 
> I'll
> set it up and change the wild guess number at the same time.
>
> If anyone knows why backing up to 1,000,000 char before the end and
> then looking for the pts is problematic please let me know.
>
> Thanks,
> Heitzso
>
> _______________________________________________
> MPlayer-users mailing list
> MPlayer-users at mplayerhq.hu
> http://mplayerhq.hu/mailman/listinfo/mplayer-users
>