[FFmpeg-devel] [RFC] General design for subtitles demuxers

[JULIAN GARDNER]

----- Original Message -----
> From: Clément Bœsch <ubitux at gmail.com>
> To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> Cc: 
> Sent: Wednesday, 13 June 2012, 9:02
> Subject: Re: [FFmpeg-devel] [RFC] General design for subtitles demuxers
> 
> On Wed, Jun 13, 2012 at 08:53:27AM +0200, Nicolas George wrote:
>>  Le sextidi 26 prairial, an CCXX, Clément Bœsch a écrit :
>>  > Yet another subtitles thread, addressing this time several issues in a 
> row.
>>  > 
>>  > All the text subtitles formats are small files where it is interesting 
> to
>>  > consider them as such, and not as a stream. In practice I propose the
>>  > following:
>>  > 
>>  > 1) Parse the whole file
>>  > 
>>  >    Just like lavf/assdec or lavf/jacosubdec, the whole file should be 
> parsed on
>>  >    the read_header() callback.
>>  > 
>>  >    Why is it done like this? Because in "format" oriented 
> ("ASS"), the events
>>  >    can be in any order, require looking ahead, or preprocessing the 
> file; when
>>  >    they are muxed ("SSA") the timeline must makes sense, and 
> it doesn't matter
>>  >    since this demuxing belongs to the video demuxer.
>>  > 
>>  >    JACOsub also has that issue, but SubRip as well (remember the event
>>  >    numbers?), and actually all the others (see below).
>>  > 
>>  > 2) Set missing durations
>>  > 
>>  >    Since we now have the complete list of events in a list, it's 
> easy to work
>>  >    on them, and we can just set the packets duration when missing that
>>  >    information: this actually solves in a very simple way the
>>  >    last-to-next-subtitle feature I mentioned several times.
>>  > 
>>  > 3) Sort the events
>>  > 
>>  >    We should then export the sorting of all these subtitles (PTS 
> based) raw
>> 
>> 
>>  Until now, I agree at 100%.
>> 
> 
> Great. I'll see what I can do on this ASAP then.
> 
>>  >    events from lavf/assdec.c.
>> 
>>  This part I am not sure I understand. Do you mean that the demuxer is
>>  supposed to translate into ASS?
>> 
> 
> I'm just refering here to:
> 
>     static int event_cmp(uint8_t **a, uint8_t **b)
>     {
>         return get_pts(*a) - get_pts(*b);
>     }
> 
>     [...]
> 
>     qsort(ass->event, ass->event_count, sizeof(*ass->event), 
> (void*)event_cmp);
> 
>>  >    We should take out read_seek2() from lavf/assdec.
>> 
>>  Looks reasonable.
>> 
> 
> We just need to find a better way of storing the events & pts; maybe to
> something simple like
> 
>     typedef struct {
>         int64_t pts;
>         uint8_t *data;
>     } FFSubtitle;
> 
> We will BTW be able to have a generic read_packet() as well, and greatly
> simplify again all this stuff.
> 

Dont we already have that information in the AVSubtitle structure, this is what i used, i created a FIFO of pointers to structs as all subtitles are alreardy read in fully, which is what caused me problems in getting the conversion to dvb subtitles.

joolz