[FFmpeg-devel] [PATCH 1/4] avformat/hlsenc: fix first fragment mp4 do not split bug
刘歧
lq at chinaffmpeg.org
Tue Dec 19 05:51:03 EET 2017
> On 19 Dec 2017, at 11:42, Karthick Jeyapal <kjeyapal at akamai.com> wrote:
>
>
>
> On 12/18/17 2:17 PM, Steven Liu wrote:
>> fix ticket id: 6888
>>
>> Signed-off-by: Steven Liu <lq at chinaffmpeg.org>
>> ---
>> libavformat/hlsenc.c | 72 ++++++++++++++++++++++++++++++++++++++++++++--------
>> 1 file changed, 62 insertions(+), 10 deletions(-)
>>
>> diff --git a/libavformat/hlsenc.c b/libavformat/hlsenc.c
>> index e3442c368f..f51fec1030 100644
>> --- a/libavformat/hlsenc.c
>> +++ b/libavformat/hlsenc.c
>> @@ -352,6 +352,29 @@ static void write_styp(AVIOContext *pb)
>> ffio_wfourcc(pb, "msix");
>> }
>> +static int flush_dynbuf(VariantStream *vs, int *range_length)
>> +{
>> + AVFormatContext *ctx = vs->avf;
>> + uint8_t *buffer;
>> +
>> + if (!ctx->pb) {
>> + return AVERROR(EINVAL);
>> + }
>> +
>> + // flush
>> + av_write_frame(ctx, NULL);
>> + avio_flush(ctx->pb);
>> +
>> + // write out to file
>> + *range_length = avio_close_dyn_buf(ctx->pb, &buffer);
>> + ctx->pb = NULL;
>> + avio_write(vs->out, buffer, *range_length);
>> + av_free(buffer);
>> +
>> + // re-open buffer
>> + return avio_open_dyn_buf(&ctx->pb);
>> +}
>> +
>> static int hls_delete_old_segments(AVFormatContext *s, HLSContext *hls,
>> VariantStream *vs) {
>> @@ -677,7 +700,9 @@ static int hls_mux_init(AVFormatContext *s, VariantStream *vs)
>> if ((ret = avio_open_dyn_buf(&oc->pb)) < 0)
>> return ret;
>> - if ((ret = s->io_open(s, &vs->out, vs->base_output_dirname, AVIO_FLAG_WRITE, &options)) < 0) {
>> + ret = hlsenc_io_open(s, &vs->out, vs->base_output_dirname, &options);
>> + av_dict_free(&options);
>> + if (ret < 0) {
>> av_log(s, AV_LOG_ERROR, "Failed to open segment '%s'\n", vs->fmp4_init_filename);
>> return ret;
>> }
>> @@ -1404,9 +1429,10 @@ static int hls_start(AVFormatContext *s, VariantStream *vs)
>> av_dict_free(&options);
>> if (err < 0)
>> return err;
>> - } else
>> + } else if (c->segment_type != SEGMENT_TYPE_FMP4) {
>> if ((err = hlsenc_io_open(s, &oc->pb, oc->filename, &options)) < 0)
>> goto fail;
>> + }
>> if (vs->vtt_basename) {
>> set_http_options(s, &options, c);
>> if ((err = hlsenc_io_open(s, &vtt_oc->pb, vtt_oc->filename, &options)) < 0)
>> @@ -1414,9 +1440,7 @@ static int hls_start(AVFormatContext *s, VariantStream *vs)
>> }
>> av_dict_free(&options);
>> - if (c->segment_type == SEGMENT_TYPE_FMP4 && !(c->flags & HLS_SINGLE_FILE)) {
>> - write_styp(oc->pb);
>> - } else {
>> + if (c->segment_type != SEGMENT_TYPE_FMP4) {
>> /* We only require one PAT/PMT per segment. */
>> if (oc->oformat->priv_class && oc->priv_data) {
>> char period[21];
>> @@ -1780,7 +1804,8 @@ static int hls_write_packet(AVFormatContext *s, AVPacket *pkt)
>> vs->size = new_start_pos - vs->start_pos;
>> if (!byterange_mode) {
>> - if (hls->segment_type == SEGMENT_TYPE_FMP4 && !vs->init_range_length) {
>> + if (hls->segment_type == SEGMENT_TYPE_FMP4) {
>> + if (!vs->init_range_length) {
>> avio_flush(oc->pb);
>> range_length = avio_close_dyn_buf(oc->pb, &buffer);
>> avio_write(vs->out, buffer, range_length);
>> @@ -1789,6 +1814,7 @@ static int hls_write_packet(AVFormatContext *s, AVPacket *pkt)
>> vs->packets_written = 0;
>> ff_format_io_close(s, &vs->out);
>> hlsenc_io_close(s, &vs->out, vs->base_output_dirname);
>> + }
>> } else {
>> hlsenc_io_close(s, &oc->pb, oc->filename);
>> }
>> @@ -1807,7 +1833,20 @@ static int hls_write_packet(AVFormatContext *s, AVPacket *pkt)
>> vs->number--;
>> }
>> - if (!vs->fmp4_init_mode || byterange_mode)
>> + if (hls->segment_type == SEGMENT_TYPE_FMP4) {
>> + ret = hlsenc_io_open(s, &vs->out, vs->avf->filename, NULL);
> Will it be better if you call "set_http_options()" here itself, instead of a separate patch 4/4? In that way this patch would be self contained.
set_http_options is add option to http operation, so separate two patches.
And this patch just fix ticket
>> + if (ret < 0) {
>> + av_log(NULL, AV_LOG_ERROR, "Failed to open file '%s'\n",
>> + vs->avf->filename);
>> + return ret;
>> + }
>> + write_styp(vs->out);
>> + ret = flush_dynbuf(vs, &range_length);
>> + if (ret < 0) {
>> + return ret;
>> + }
>> + ff_format_io_close(s, &vs->out);
> Again is it better to call hlsenc_io_close() here itself instead of patch 3/4. Again just from self-containment perspective.
Replace API to hlsenc_io_close is a new operation, so separate two patch
And this patch just fix ticket
>> + }
>> ret = hls_append_segment(s, hls, vs, vs->duration, vs->start_pos, vs->size);
>> vs->start_pos = new_start_pos;
>> if (ret < 0) {
>> @@ -1861,6 +1900,7 @@ static int hls_write_trailer(struct AVFormatContext *s)
>> AVFormatContext *vtt_oc = NULL;
>> char *old_filename = NULL;
>> int i;
>> + int ret = 0;
>> VariantStream *vs = NULL;
>> for (i = 0; i < hls->nb_varstreams; i++) {
>> @@ -1873,11 +1913,25 @@ static int hls_write_trailer(struct AVFormatContext *s)
>> if (!old_filename) {
>> return AVERROR(ENOMEM);
>> }
>> -
>> + if ( hls->segment_type == SEGMENT_TYPE_FMP4) {
>> + int range_length = 0;
>> + ret = hlsenc_io_open(s, &vs->out, vs->avf->filename, NULL);
>> + if (ret < 0) {
>> + av_log(NULL, AV_LOG_ERROR, "Failed to open file '%s'\n", vs->avf->filename);
>> + return AVERROR(ENOENT);
>> + }
>> + write_styp(vs->out);
>> + ret = flush_dynbuf(vs, &range_length);
>> + if (ret < 0) {
>> + return ret;
>> + }
>> + ff_format_io_close(s, &vs->out);
>> + }
> Will it make sense it make this code as a separate function in order to avoid code repetition?
That is optimize operation, this patch just fix ticket, so i will send a new patch to merge two workflow to on function.
Thanks
Steven
>> av_write_trailer(oc);
>> if (oc->pb) {
>> vs->size = avio_tell(vs->avf->pb) - vs->start_pos;
>> + if (hls->segment_type != SEGMENT_TYPE_FMP4)
>> ff_format_io_close(s, &oc->pb);
>> if ((hls->flags & HLS_TEMP_FILE) && oc->filename[0]) {
>> @@ -2238,10 +2292,8 @@ static int hls_init(AVFormatContext *s)
>> }
>> }
>> - if (hls->segment_type != SEGMENT_TYPE_FMP4 || hls->flags & HLS_SINGLE_FILE) {
>> if ((ret = hls_start(s, vs)) < 0)
>> goto fail;
>> - }
>> }
>> fail:
> Thanks,
> Karthick
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
More information about the ffmpeg-devel
mailing list