[FFmpeg-devel] [PATCH 04/12] avformat/mxfdec: compute both essence_offset and essence_length in mxf_compute_essence_containers

[Tomas Härdin]

ons 2018-06-13 klockan 21:58 +0200 skrev Marton Balint:
> 
> On Wed, 13 Jun 2018, Tomas Härdin wrote:
> 
> > sön 2018-06-10 klockan 12:36 +0200 skrev Marton Balint:
> > > Also compute the correct essence_offset and essence_length for all clip wrapped
> > > essences.
> > > 
> > > > Signed-off-by: Marton Balint <cus at passwd.hu>
> > > 
> > > ---
> > >  libavformat/mxfdec.c | 108 +++++++++++++++++++++++++++------------------------
> > >  1 file changed, 57 insertions(+), 51 deletions(-)
> > > 
> > > diff --git a/libavformat/mxfdec.c b/libavformat/mxfdec.c
> > > index 1ab34f4873..67b0028e88 100644
> > > --- a/libavformat/mxfdec.c
> > > +++ b/libavformat/mxfdec.c
> > > @@ -98,6 +98,7 @@ typedef struct MXFPartition {
> > >      int pack_length;
> > >      int64_t pack_ofs;               ///< absolute offset of pack in file, including run-in
> > >      int64_t body_offset;
> > > +    KLVPacket first_essence_klv;
> > >  } MXFPartition;
> > >  
> > >  typedef struct MXFCryptoContext {
> > > @@ -2782,47 +2783,76 @@ static int mxf_parse_handle_partition_or_eof(MXFContext *mxf)
> > >      return mxf->parsing_backward ? mxf_seek_to_previous_partition(mxf) : 1;
> > >  }
> > >  
> > > +static int64_t round_to_kag(int64_t position, int kag_size)
> > > +{
> > > +    /* TODO: account for run-in? the spec isn't clear whether KAG should account for it */
> > > +    /* NOTE: kag_size may be any integer between 1 - 2^10 */
> > > +    int64_t ret = (position / kag_size) * kag_size;
> > > +    return ret == position ? ret : ret + kag_size;
> > 
> > This should probably just be
> > 
> > return ((position + kag_size - 1) / kag_size) * kag_size;
> > 
> > but maybe there's some fine point I'm overlooking. What happens if
> > kag_size is a smallish value like 16?
> > 
> > A check for position > INT64_MAX - kag_size might also be appropriate.
> 
> Well, this is your code as well :), I just moved it to avoid a forward 
> declaration.
> 
> > 
> > > +}
> > > +
> > > +static MXFWrappingScheme mxf_get_wrapping_by_body_sid(AVFormatContext *s, int body_sid)
> > > +{
> > > +    for (int i = 0; i < s->nb_streams; i++) {
> > > +        MXFTrack *track = s->streams[i]->priv_data;
> > > +        if (track && track->body_sid == body_sid && track->wrapping != UnknownWrapped)
> > > +            return track->wrapping;
> > > +    }
> > > +    return UnknownWrapped;
> > > +}
> > > +
> > >  /**
> > >   * Figures out the proper offset and length of the essence container in each partition
> > >   */
> > > -static void mxf_compute_essence_containers(MXFContext *mxf)
> > > +static void mxf_compute_essence_containers(AVFormatContext *s)
> > >  {
> > > +    MXFContext *mxf = s->priv_data;
> > >      int x;
> > >  
> > > -    /* everything is already correct */
> > > -    if (mxf->op == OPAtom)
> > > -        return;
> > > -
> > >      for (x = 0; x < mxf->partitions_count; x++) {
> > >          MXFPartition *p = &mxf->partitions[x];
> > > +        MXFWrappingScheme wrapping;
> > >  
> > >          if (!p->body_sid)
> > >              continue;       /* BodySID == 0 -> no essence */
> > >  
> > > -        if (x >= mxf->partitions_count - 1)
> > > -            break;          /* FooterPartition - can't compute length (and we don't need to) */
> > > +        /* for non clip-wrapped essences we compute essence_offset
> > > +         * for clip wrapped essences we point essence_offset after the KL (usually klv.offset + 20 or 25)
> > > +         */
> > >  
> > > -        /* essence container spans to the next partition */
> > > -        p->essence_length = mxf->partitions[x+1].this_partition - p->essence_offset;
> > > +        wrapping = (mxf->op == OPAtom) ? ClipWrapped : mxf_get_wrapping_by_body_sid(s, p->body_sid);
> > >  
> > > -        if (p->essence_length < 0) {
> > > -            /* next ThisPartition < essence_offset */
> > > -            p->essence_length = 0;
> > > -            av_log(mxf->fc, AV_LOG_ERROR,
> > > -                   "partition %i: bad ThisPartition = %"PRIX64"\n",
> > > -                   x+1, mxf->partitions[x+1].this_partition);
> > > +        if (wrapping == ClipWrapped) {
> > > +            p->essence_offset = p->first_essence_klv.next_klv - p->first_essence_klv.length;
> > > +            p->essence_length = p->first_essence_klv.length;
> > 
> > Much nicer
> > 
> > > +        } else {
> > > +            int64_t op1a_essence_offset =
> > > +                p->this_partition +
> > > +                round_to_kag(p->pack_length,       p->kag_size) +
> > > +                round_to_kag(p->header_byte_count, p->kag_size) +
> > > +                round_to_kag(p->index_byte_count,  p->kag_size);
> > 
> > Is this really always the case? I guess with OP1a it isn't a huge
> > concern since the demuxer will find the next essence packet anyway. But
> > still..
> > 
> > I'm also fairly sure this is my code originally so.. :)
> 
> I think this tends to work well, if kag_size is not guessed. However, if 
> it _is_ guessed, then there might be problems, and having a 0 kagsize is 
> valid (only means an unknown KAG). So I am inclined to simply 
> use p->first_essence_klv.offset unconditionally, as you suggested in
> 
> http://ffmpeg.org/pipermail/ffmpeg-devel/2018-February/225629.html

Yes, past me seems to have the right idea :)

> See the attached patch.

I think we should still parse kag_size even if we don't end up using it
for much. Especially since there's special cases for Sony files in
there. Did we have a sample for that? I forget.

Are there cases where we might not run across an essence KLV during
scanning but where there would be one anyway? I suspect not, not even
if we're reading an unseekable file

/Tomas