[FFmpeg-devel] [PATCH] lavf/ape: Fix last packet duration calculation

[Paul B Mahol]

On 8/25/12, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Sat, Aug 25, 2012 at 12:20:38PM +0000, Paul B Mahol wrote:
>> Signed-off-by: Paul B Mahol <onemda at gmail.com>
>> ---
>>  libavformat/ape.c |    6 ++----
>>  1 files changed, 2 insertions(+), 4 deletions(-)
>>
>> diff --git a/libavformat/ape.c b/libavformat/ape.c
>> index c4a390b..9c05cee 100644
>> --- a/libavformat/ape.c
>> +++ b/libavformat/ape.c
>> @@ -45,7 +45,6 @@ typedef struct {
>>      int nblocks;
>>      int size;
>>      int skip;
>> -    int64_t pts;
>>  } APEFrame;
>>
>>  typedef struct {
>> @@ -344,8 +343,7 @@ static int ape_read_header(AVFormatContext * s)
>>
>>      pts = 0;
>>      for (i = 0; i < ape->totalframes; i++) {
>> -        ape->frames[i].pts = pts;
>> -        av_add_index_entry(st, ape->frames[i].pos, ape->frames[i].pts, 0,
>> 0, AVINDEX_KEYFRAME);
>> +        av_add_index_entry(st, ape->frames[i].pos, pts, 0, 0,
>> AVINDEX_KEYFRAME);
>>          pts += ape->blocksperframe;
>>      }
>>
>
>> @@ -394,7 +392,7 @@ static int ape_read_packet(AVFormatContext * s,
>> AVPacket * pkt)
>>      AV_WL32(pkt->data + 4, ape->frames[ape->currentframe].skip);
>>      ret = avio_read(s->pb, pkt->data + extra_size,
>> ape->frames[ape->currentframe].size);
>>
>> -    pkt->pts = ape->frames[ape->currentframe].pts;
>
> this causes a significant slowdown on slow media, as much more data
> will be read in the hope to find a valid PTS (that will not be found
> before EOF or max_analyze_duration)

That is because of extremly big superframes in ape, I really doubt there
is real slowdown with this patch, because avformat reads whole packet anyway.
>
> you can see this with -v 99
> with pts:
> All info found
> File position after avformat_find_stream_info() is 134672
>
> without pts:
> max_analyze_duration 5000000 reached at 5015510
> File position after avformat_find_stream_info() is 1048576
>

I think this is unrelated bug, current packet duration caclulation is
wrong with ape.

It have many frames in single packet and this does not work correctly at all.