[FFmpeg-devel] [PATCH] avcodec: add HDMV Text Subtitle decoder

[Hendrik Leppkes]

On Wed, Feb 1, 2017 at 1:20 PM, Paul B Mahol <onemda at gmail.com> wrote:
> On 2/1/17, wm4 <nfxjfg at googlemail.com> wrote:
>> On Wed,  1 Feb 2017 12:12:51 +0100
>> Paul B Mahol <onemda at gmail.com> wrote:
>>
>>> Signed-off-by: Paul B Mahol <onemda at gmail.com>
>>> ---
>>>  libavcodec/Makefile    |   1 +
>>>  libavcodec/allcodecs.c |   1 +
>>>  libavcodec/textstdec.c | 409
>>> +++++++++++++++++++++++++++++++++++++++++++++++++
>>>  3 files changed, 411 insertions(+)
>>>  create mode 100644 libavcodec/textstdec.c
>>>
>
> [...]
>
>>> +
>>> +static void decode_presentation_segment(AVCodecContext *avctx,
>>> GetByteContext *gb, AVSubtitle *sub)
>>> +{
>>> +    TextSTContext *s = avctx->priv_data;
>>> +    unsigned ii, palette_update_flag, region_count;
>>> +    int64_t start_pts, end_pts;
>>> +
>>> +    start_pts = decode_pts(gb);
>>> +    end_pts   = decode_pts(gb);
>>> +
>>> +    sub->pts = start_pts * 100 / 9;
>>
>> I don't think this will work. You should use packet timestamps, not get
>> them from the packet contents. Otherwise we'd be back to the situation
>> with other text subtitles, which used to embed timestamps in the packet
>> data.
>>
>
> pts are stored in packet, they are not available from packet timestamp.
>
> And this succesfully converts TextST to ass format.

In-band timestamps are fragile and don't work half the time. Many
use-cases rely on reading and potentially modifying packet timestamps,
and that just doesn't work if they then get overriden by the decoder -
ie. all sorts of code would need special cases to handle this kind of
stream.
Better to have something on our end that handles them and rips out the
in-band timestamps and puts them into the AVPacket straight out of the
demuxer.

- Hendrik