[FFmpeg-devel] [PATCH v2 1/4] avformat: add av_program_copy()

Gyan Doshi ffmpeg at gyani.pro
Thu Feb 16 09:58:10 EET 2023



On 2023-02-16 12:22 pm, Andreas Rheinhardt wrote:
> Gyan Doshi:
>> Helper to transfer programs from one muxing context to another.
>> ---
>>   doc/APIchanges         |  3 ++
>>   libavformat/avformat.c | 70 ++++++++++++++++++++++++++++++++++++++++++
>>   libavformat/avformat.h | 14 +++++++++
>>   libavformat/version.h  |  2 +-
>>   4 files changed, 88 insertions(+), 1 deletion(-)
>>
>> diff --git a/doc/APIchanges b/doc/APIchanges
>> index 6baf914760..4916d1abe8 100644
>> --- a/doc/APIchanges
>> +++ b/doc/APIchanges
>> @@ -14,6 +14,9 @@ libavutil:     2021-04-27
>>   
>>   API changes, most recent first:
>>   
>> +2023-02-xx - xxxxxxxxxx - lavf 59.39.100 - avformat.h
>> +  Add av_program_copy()
>> +
>>   2023-0x-xx - xxxxxxxxxx - lavc 59.63.100
>>     Allow AV_CODEC_FLAG_COPY_OPAQUE to be used with decoders.
>>   
>> diff --git a/libavformat/avformat.c b/libavformat/avformat.c
>> index 19c7219471..d3c8def170 100644
>> --- a/libavformat/avformat.c
>> +++ b/libavformat/avformat.c
>> @@ -359,6 +359,76 @@ void av_program_add_stream_index(AVFormatContext *ac, int progid, unsigned idx)
>>       }
>>   }
>>   
>> +int av_program_copy(AVFormatContext *dst, AVFormatContext *src, int progid, int overwrite)
>> +{
>> +    AVProgram *src_prog = NULL;
> Should be const.
>
>> +    AVProgram *dst_prog = NULL;
>> +    int i, j, ret;
>> +    int idx = -1;
>> +
>> +    for (i = 0; i < src->nb_programs; i++)
> 1. nb_programs is unsigned and therefore the iterator should be, too.
> 2. Use a smaller scope for the iterator.
> (Same for all other iterators.)
>
>> +        if (src->programs[i]->id == progid)
>> +            src_prog = src->programs[i];
>> +
>> +    if (!src_prog) {
>> +        av_log(src, AV_LOG_ERROR, "source program not found: id=0x%04x\n", progid);
>> +        return AVERROR(EINVAL);
>> +    }
>> +
>> +    for (i = 0; i < dst->nb_programs; i++) {
>> +        if (dst->programs[i]->id == progid)
>> +            idx = i;
>> +    }
>> +
>> +    if (idx >= 0 && !overwrite) {
>> +        av_log(src, AV_LOG_ERROR, "target muxer already has program with id=0x%04x; not overwriting\n", progid);
>> +        return AVERROR(EINVAL);
> This should not be an error.
>
>> +    }
>> +
>> +    av_log(src, AV_LOG_TRACE, "%s program: id=0x%04x\n", idx >= 0 ? "overwriting" : "copying", progid);
>> +
>> +    if (idx >= 0) {
>> +        dst_prog = dst->programs[idx];
>> +        av_dict_free(&dst_prog->metadata);
>> +        av_freep(&dst_prog->stream_index);
>> +        dst_prog->nb_stream_indexes = 0;
>> +    } else {
>> +        dst_prog = av_mallocz(sizeof(*dst_prog));
>> +        if (!dst_prog)
>> +            return AVERROR(ENOMEM);
>> +        ret = av_dynarray_add_nofree(&dst->programs, &dst->nb_programs, dst_prog);
> av_dynarray_add_nofree() presumes that dst->programs points to a buffer
> for a power-of-two pointers (or to NULL); what if a user has reallocated
> dst->programs himself and moved AVPrograms from another AVFormatContext
> manually? Do we treat this as API violation?

That scenario should be treated the same way as in av_new_program().
But I presume that if a user knows of and uses this helper, they won't 
have done that.

I can add it in the doxy as a caveat.


>
>> +        if (ret < 0) {
>> +            av_free(dst_prog);
>> +            return AVERROR(ENOMEM);
>> +        }
>> +    }
>> +
>> +    /* public fields */
>> +    dst_prog->id          = src_prog->id;
>> +    dst_prog->flags       = src_prog->flags;
>> +    dst_prog->discard     = src_prog->discard;
>> +    dst_prog->program_num = src_prog->program_num;
>> +    dst_prog->pmt_pid     = src_prog->pmt_pid;
>> +    dst_prog->pcr_pid     = src_prog->pcr_pid;
>> +    dst_prog->pmt_version = src_prog->pmt_version;
>> +
>> +    for (i = 0; i < dst->nb_streams; i++) {
>> +        for (j = 0; j < src_prog->nb_stream_indexes; j++)
>> +            if (dst->streams[i]->id == src->streams[src_prog->stream_index[j]]->id)
> The documentation should mention that AVStream.id is used for
> stream-matching.
>
>> +                av_program_add_stream_index(dst, dst_prog->id, i);
> This involves a realloction whose success can't be checked due to a
> design bug in av_program_add_stream_index(). Add an
> ff_program_add_stream_index() without this design bug and turn
> av_program_add_stream_index() into a wrapper for it.

Isn't it better to deprecate this and add av_program_add_stream_index2() 
without the bug?


>
>> +    }
>> +
>> +    av_dict_copy(&dst_prog->metadata, src_prog->metadata, 0);
> Missing error check.


Ok, but as a note, most usages (80%+) in the codebase omit the check, 
including
lavu/opt.c, lavu/tests/dict.c,  lavf/mux.c and fftools/*

Do I restore the destination program (when overwriting) or bow out?


>
>> +
>> +    /* private fields */
>> +    dst_prog->start_time         = src_prog->start_time;
>> +    dst_prog->end_time           = src_prog->end_time;
>> +    dst_prog->pts_wrap_reference = src_prog->pts_wrap_reference;
>> +    dst_prog->pts_wrap_behavior  = src_prog->pts_wrap_behavior;
> These private fields are demuxer-only; there is no need to copy them for
> muxers.
>
>> +
>> +    return 0;
>> +}
>> +
>>   AVProgram *av_find_program_from_stream(AVFormatContext *ic, AVProgram *last, int s)
>>   {
>>       for (unsigned i = 0; i < ic->nb_programs; i++) {
>> diff --git a/libavformat/avformat.h b/libavformat/avformat.h
>> index 1d97d56ac5..2b7ed3abd8 100644
>> --- a/libavformat/avformat.h
>> +++ b/libavformat/avformat.h
>> @@ -1996,6 +1996,20 @@ uint8_t *av_stream_get_side_data(const AVStream *stream,
>>   
>>   AVProgram *av_new_program(AVFormatContext *s, int id);
>>   
>> +/**
>> + * Copy an AVProgram from one AVFormatContext to another.
>> + *
>> + * @param dst           pointer to the target muxer context
>> + * @param src           pointer to the source muxer context
>> + * @param progid        ID of the program to be copied
>> + * @param overwrite     whether to overwrite if target muxer already
>> + *                      contains a program with the same ID
>> + *
>> + * @return  0 in case of success, a negative AVERROR code in case of
>> + *          failure
> Better make >= 0 for success for extensibility.
>
>> + */
>> +int av_program_copy(AVFormatContext *dst, AVFormatContext *src, int progid, int overwrite);
>> +
> 3. Why is this public? Which user is supposed to need this?

Someone implementing tee or similar manually.

Ok to the rest.

Regards,
Gyan



> 4. src should be const.
>
>>   /**
>>    * @}
>>    */
>> diff --git a/libavformat/version.h b/libavformat/version.h
>> index 134cdb2b89..9aba356e09 100644
>> --- a/libavformat/version.h
>> +++ b/libavformat/version.h
>> @@ -31,7 +31,7 @@
>>   
>>   #include "version_major.h"
>>   
>> -#define LIBAVFORMAT_VERSION_MINOR  38
>> +#define LIBAVFORMAT_VERSION_MINOR  39
>>   #define LIBAVFORMAT_VERSION_MICRO 100
>>   
>>   #define LIBAVFORMAT_VERSION_INT AV_VERSION_INT(LIBAVFORMAT_VERSION_MAJOR, \
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".



More information about the ffmpeg-devel mailing list