[FFmpeg-devel] [PATCH 1/4] avformat/nutenc: don't use header_count to store different variables

[Andreas Rheinhardt]

Andriy Gelman:
> On Mon, 09. Nov 00:04, Andreas Rheinhardt wrote:
>> Andriy Gelman:
>>> From: Andriy Gelman <andriy.gelman at gmail.com>
>>>
>>> Currently, header_count is used to store both the elision header count
>>> and the header repetition count (number of times headers have been written
>>> to output). Fix this by using a separate variable to store repetition
>>> count.
>>>
>>> Signed-off-by: Andriy Gelman <andriy.gelman at gmail.com>
>>> ---
>>>  libavformat/nut.h    | 3 ++-
>>>  libavformat/nutenc.c | 4 ++--
>>>  2 files changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/libavformat/nut.h b/libavformat/nut.h
>>> index a4409ee23d..a990d3832e 100644
>>> --- a/libavformat/nut.h
>>> +++ b/libavformat/nut.h
>>> @@ -103,7 +103,8 @@ typedef struct NUTContext {
>>>      unsigned int time_base_count;
>>>      int64_t last_syncpoint_pos;
>>>      int64_t last_resync_pos;
>>> -    int header_count;
>>> +    int header_count;           // elision header count
>>> +    int header_rep_count;       // number of times headers written
>>>      AVRational *time_base;
>>>      struct AVTreeNode *syncpoints;
>>>      int sp_count;
>>> diff --git a/libavformat/nutenc.c b/libavformat/nutenc.c
>>> index 1dcb2be1b1..87adef6f7e 100644
>>> --- a/libavformat/nutenc.c
>>> +++ b/libavformat/nutenc.c
>>> @@ -684,7 +684,7 @@ static int write_headers(AVFormatContext *avctx, AVIOContext *bc)
>>>      }
>>>  
>>>      nut->last_syncpoint_pos = INT_MIN;
>>> -    nut->header_count++;
>>> +    nut->header_rep_count++;
>>>  
>>>      ret = 0;
>>>  fail:
>>> @@ -988,7 +988,7 @@ static int nut_write_packet(AVFormatContext *s, AVPacket *pkt)
>>>          data_size += sm_size;
>>>      }
>>>  
>>> -    if (1LL << (20 + 3 * nut->header_count) <= avio_tell(bc))
>>> +    if (1LL << (20 + 3 * nut->header_rep_count) <= avio_tell(bc))
>>>          write_headers(s, bc);
>>>  
>>>      if (key_frame && !(nus->last_flags & FLAG_KEY))
>>>
>> 1. You are not the first to notice this:
>> https://patchwork.ffmpeg.org/project/ffmpeg/patch/20200505141657.10787-2-andreas.rheinhardt@gmail.com/
> 
> The goal of my set was to help user in [1] by adding option to insert periodic
> headers (see patch 3).
> http://ffmpeg.org/pipermail/ffmpeg-user/2020-November/050690.html
> 
>> 2. Your patch is incomplete: header_count is also used in write_trailer.
>> If you use the correct value there, you will have to update lots of
>> fate-tests (see the above commit). (The fate updates of that old patch
>> need to be updated (and were incomplete anyway because I did not ran
>> tests for gpl-components*).
> 
> It has been dead code since the header elision support was added.
> Would it make sense to remove that code in write_trailer()? This would simplify
> your patch.
> 

It would simplify the patch, but it is not how it is supposed to be.

>> 3. The reason the above patch (or rather patchset) has not been applied
>> is that there is actually a subtle bug with chapters: Users are allowed
>> to add new chapters after writing the header (some muxer then write the
>> chapters when writing the trailer). Yet this doesn't work with nut right
>> now: See the commit message of the preceding patch
>> (https://patchwork.ffmpeg.org/project/ffmpeg/patch/20200505141657.10787-1-andreas.rheinhardt@gmail.com/)
>> for details.
> 
> I'd have to look at your set in more detail, but I thought repeated headers had
> to be the same as original ones. From the spec:
> 
> "Headers may be repeated, but if they are, then all mandatory headers MUST be
> repeated and repeated headers MUST be identical."
> 
For the specs, chapters are info_packets and not part of a mandatory
header, so the above is no restriction.

- Andreas