[FFmpeg-devel] [PATCH] lavf/mov: ensure only one tkhd per trak

[Chris Cunningham]

> But if the ID isn't coming from a tkhd (either because none exits, or
because you have truns before the tkhd appears), perhaps we can consider
invalid.

Taking a closer look at the spec, I think it actually _is valid_ to have
truns before tkhd. They "strongly recommend" that the header boxes like
tkhd be placed first, but its not a "must". No clue how often ffmpeg
encounters such a file, but today's defaulting of id = the number of
streams probably facilitates playback in such cases because the probability
of tkhd's track id matching that default is reasonably high.

Still, I'll take a stab at the approach, if just for discussion/testing.

On Thu, Dec 13, 2018 at 12:39 PM Chris Cunningham <chcunningham at chromium.org>
wrote:

> "st->id" is not necessary for demuxing AFAIK, please correct me if Im
>> wrong.
>> Would an init value to -1 for st->id work ?
>>
>
> st->id does get used here and there. For ex, mov_read_trun reads the id to
> decide which stream corresponds to the current fragment. But if the ID
> isn't coming from a tkhd (either because none exits, or because you have
> truns before the tkhd appears), perhaps we can consider invalid. Using a
> value of -1 would make that easier to detect. It risks breaking some bad
> files, but I'm fine with that if you'd support it. I'll send a new patch
> and we can see if Michael finds new breaks.
>
> Chris
>