@@ -158,6 +158,8 @@ typedef struct mkv_cues {
int num_entries;
} mkv_cues;
+struct MatroskaMuxContext;
+
typedef struct mkv_track {
int write_dts;
int has_cue;
@@ -171,6 +173,13 @@ typedef struct mkv_track {
int64_t duration_offset;
int64_t codecpriv_offset;
int64_t ts_offset;
+ /* This callback will be called twice: First with a NULL AVIOContext
+ * to return the size of the (Simple)Block's data via size
+ * and a second time with the AVIOContext set when the data
+ * shall be written.
+ * The callback shall not return an error on the second call. */
+ int (*reformat)(struct MatroskaMuxContext *, AVIOContext *,
+ const AVPacket *, int *size);
} mkv_track;
typedef struct MatroskaMuxContext {
@@ -2433,6 +2442,8 @@ static int mkv_write_block(AVFormatContext *s, AVIOContext *pb,
#endif
if (par->codec_id == AV_CODEC_ID_AV1) {
err = ff_av1_filter_obus_buf(pkt->data, &data, &size, &offset);
+ } else if (track->reformat) {
+ err = track->reformat(mkv, NULL, pkt, &size);
} else
data = pkt->data;
@@ -2483,9 +2494,13 @@ static int mkv_write_block(AVFormatContext *s, AVIOContext *pb,
put_ebml_num(pb, track_number, track->track_num_size);
avio_wb16(pb, ts - mkv->cluster_pts);
avio_w8(pb, (blockid == MATROSKA_ID_SIMPLEBLOCK && keyframe) ? (1 << 7) : 0);
+ if (track->reformat) {
+ track->reformat(mkv, pb, pkt, NULL);
+ } else {
avio_write(pb, data + offset, size);
if (data != pkt->data)
av_free(data);
+ }
if (blockid == MATROSKA_ID_BLOCK && !keyframe)
put_ebml_sint(pb, MATROSKA_ID_BLOCKREFERENCE, track->last_timestamp - ts);
Matroska uses variable-length elements and in order not to waste bytes on length fields, the length of the data to write needs to be known before writing the length field. Annex B H.264/5 and WavPack need to be reformatted to know this length and this currently involves writing the data into temporary buffers; AV1 sometimes suffers from this as well. This commit aims to solve this by adding a callback that is called twice per packet: Once to get the size and once to actually write the data. In case of WavPack and AV1 (where parsing is cheap due to length fields) both calls will just parse the data with only the second function writing anything. For H.264/5, the position of the NALUs will need to be stored to be written lateron. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- libavformat/matroskaenc.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)