On 02/19/2017 03:23 PM, John Stebbins wrote:
> initial_padding was getting added to the edit list indirectly due to
> initial negative dts.  But in cases where the audio is delayed,
> all or part of initial_padding would be unaccounted for.  This patch
> makes initial_padding explicit.
> ---
>  libavformat/movenc.c | 53 
> +++++++++++++++++++++++++++++++++++++---------------
>  1 file changed, 38 insertions(+), 15 deletions(-)
>
> diff --git a/libavformat/movenc.c b/libavformat/movenc.c
> index 689291d..1f29716 100644
> --- a/libavformat/movenc.c
> +++ b/libavformat/movenc.c
> @@ -1698,9 +1698,28 @@ static int mov_write_edts_tag(AVIOContext *pb, 
> MOVMuxContext *mov,
>                                        track->timescale, AV_ROUND_UP);
>      int version = duration < INT32_MAX ? 0 : 1;
>      int entry_size, entry_count, size;
> -    int64_t delay, start_ct = track->start_cts;
> -    delay = av_rescale_rnd(track->start_dts + start_ct, MOV_TIMESCALE,
> +    int64_t delay;
> +    int64_t mediatime;
> +    int64_t skip = 0;
> +
> +    delay = av_rescale_rnd(track->start_dts + track->start_cts, 
> MOV_TIMESCALE,
>                             track->timescale, AV_ROUND_DOWN);
> +
> +    if (track->par->codec_type == AVMEDIA_TYPE_AUDIO &&
> +        track->par->initial_padding > 0) {
> +        /* Adjust delay so that initial_padding gets recorded in the
> +         * MediaTime of an edit list entry even in the case that
> +         * delay is positive. I.e. we don't want initial_padding to be
> +         * absorbed and hidden in the delay. MediaTime must contain
> +         * initial_padding in order to know where the actual media
> +         * timeline begins. A player should drop samples until MediaTime
> +         * is reached */
> +        delay += av_rescale_rnd(track->par->initial_padding, MOV_TIMESCALE,
> +                                track->par->sample_rate, AV_ROUND_DOWN);
> +        skip = av_rescale_rnd(track->par->initial_padding,
> +                              track->timescale,
> +                              track->par->sample_rate, AV_ROUND_DOWN);
> +    }
>      version |= delay < INT32_MAX ? 0 : 1;
>  
>      entry_size = (version == 1) ? 20 : 12;
> @@ -1731,33 +1750,37 @@ static int mov_write_edts_tag(AVIOContext *pb, 
> MOVMuxContext *mov,
>          }
>          avio_wb32(pb, 0x00010000);
>      } else {
> -        /* Avoid accidentally ending up with start_ct = -1 which has got a
> -         * special meaning. Normally start_ct should end up positive or zero
> -         * here, but use FFMIN in case dts is a a small positive integer
> -         * rounded to 0 when represented in MOV_TIMESCALE units. */
> -        start_ct  = -FFMIN(track->start_dts, 0);
> -        /* Note, this delay is calculated from the pts of the first sample,
> -         * ensuring that we don't reduce the duration for cases with
> -         * dts<0 pts=0. */
> -        duration += delay;
> +        /* Avoid accidentally ending up with mediatime = -1 which has got a
> +         * special meaning. skip and -track->start_dts are guaranteed to be
> +         * positive here, so it is not possible mediatime to be -1 */
> +        skip = FFMAX(skip, -track->start_dts);

I was doing some additional testing (or rather I recruited someone to help me 
test ;) and found an error in my math. 
The above calculation should be

skip = FFMAX(skip, -track->start_dts - track->start_cts);

Sending an update to fix this.

-- 
John      GnuPG fingerprint: D0EC B3DB C372 D1F1 0B01  83F0 49F1 D7B2 60D4 D0F7


Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to