[FFmpeg-devel] [PATCH] mxfdec: calculate the index in display order

[Marton Balint]

On Tue, 14 Jul 2015, Tomas Härdin wrote:

> On Sat, 2015-07-11 at 18:54 +0200, Marton Balint wrote:
>> This should fix seeking for open GOP files as well.
>>
>> Signed-off-by: Marton Balint <cus at passwd.hu>
>> ---
>> @@ -1411,8 +1417,7 @@ static int mxf_compute_ptses_fake_index(MXFContext *mxf, MXFIndexTable *index_ta
>>                  break;
>>              }
>>
>> -            index_table->fake_index[x].timestamp = x;
>
> Why is this removed?

It is only moved to the index calculation loop below, it looked cleaner to
fill all fields of the fake index in one place.

>
>> -            index_table->fake_index[x].flags = !(s->flag_entries[j] & 0x30) ? AVINDEX_KEYFRAME : 0;
>> +            flags[x] = !(s->flag_entries[j] & 0x30) ? AVINDEX_KEYFRAME : 0;
>>
>>              if (index < 0 || index >= index_table->nb_ptses) {
>>                  av_log(mxf->fc, AV_LOG_ERROR,
>> @@ -1421,11 +1426,20 @@ static int mxf_compute_ptses_fake_index(MXFContext *mxf, MXFIndexTable *index_ta
>>                  continue;
>>              }
>>
>> +            index_table->offsets[x] = offset;
>>              index_table->ptses[index] = x;
>>              max_temporal_offset = FFMAX(max_temporal_offset, offset);
>>          }
>>      }
>>
>> +    /* calculate the fake index table in display order */
>> +    for (x = 0; x < index_table->nb_ptses; x++) {
>> +        index_table->fake_index[x].timestamp = x;
>> +        if (index_table->ptses[x] != AV_NOPTS_VALUE)
>> +            index_table->fake_index[index_table->ptses[x]].flags = flags[x];
>
> ptses are checked to be in range, right?
>

Sure :). ptses are only set to x in the first loop, and x is between 0 and 
nb_ptses.

> Anyway, the patch is probably OK. I recall doing something similar at
> work in another mxf library. Do we have test files with B-frames? If not
> then this may be a good time to add them.

tests/data/lavf/lavf.mxf already has B-frames, but probably a more 
thorough seek test for the file would not hurt... I plan to work on this 
later.

Thanks,
Marton