| Message ID | PR3PR03MB6665EED3ECA95439D9B03CFE8FC39@PR3PR03MB6665.eurprd03.prod.outlook.com |
|---|---|
| State | Accepted |
| Commit | 0383ec88a56dcad7b2e87ff243f26d3287ba3fdb |
| Headers | show |
| Series | [FFmpeg-devel,1/5] avformat/av1dec: Set position of AVPackets given to BSF | expand |
| Context | Check | Description |
|---|---|---|
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
On 8/22/2021 10:25 AM, Andreas Rheinhardt wrote: > The low overhead OBU format provides no means to resync after performing > a byte-based seek; in other words: Byte based seeking is just not > supported. > > Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> > --- > The reason I didn't disallow this earlier was that there is one scenario > where such a seek would work: When one already knows the correct > position for the start of a frame. But otherwise it just doesn't work. > > Somehow ffplay tries byte based seeking even for formats with the > AVFMT_NO_BYTE_SEEK flag set. Line 2837 should probably be changed to no > longer do that. > > libavformat/av1dec.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libavformat/av1dec.c b/libavformat/av1dec.c > index 88a3c325e4..37f21483b9 100644 > --- a/libavformat/av1dec.c > +++ b/libavformat/av1dec.c > @@ -428,7 +428,7 @@ const AVInputFormat ff_obu_demuxer = { > .read_packet = obu_read_packet, > .read_close = av1_read_close, > .extensions = "obu", > - .flags = AVFMT_GENERIC_INDEX, > + .flags = AVFMT_GENERIC_INDEX | AVFMT_NO_BYTE_SEEK, > .priv_class = &av1_demuxer_class, > }; > #endif LGTM. Maybe also AVFMT_NOTIMESTAMPS could be added.
James Almer: > On 8/22/2021 10:25 AM, Andreas Rheinhardt wrote: >> The low overhead OBU format provides no means to resync after performing >> a byte-based seek; in other words: Byte based seeking is just not >> supported. >> >> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> >> --- >> The reason I didn't disallow this earlier was that there is one scenario >> where such a seek would work: When one already knows the correct >> position for the start of a frame. But otherwise it just doesn't work. >> >> Somehow ffplay tries byte based seeking even for formats with the >> AVFMT_NO_BYTE_SEEK flag set. Line 2837 should probably be changed to no >> longer do that. >> >> libavformat/av1dec.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/libavformat/av1dec.c b/libavformat/av1dec.c >> index 88a3c325e4..37f21483b9 100644 >> --- a/libavformat/av1dec.c >> +++ b/libavformat/av1dec.c >> @@ -428,7 +428,7 @@ const AVInputFormat ff_obu_demuxer = { >> .read_packet = obu_read_packet, >> .read_close = av1_read_close, >> .extensions = "obu", >> - .flags = AVFMT_GENERIC_INDEX, >> + .flags = AVFMT_GENERIC_INDEX | AVFMT_NO_BYTE_SEEK, >> .priv_class = &av1_demuxer_class, >> }; >> #endif > > LGTM. > > Maybe also AVFMT_NOTIMESTAMPS could be added. But isn't AV1 able to contain timestamps via its timing_info() and metadata_timecode() structure? - Andreas
On 8/22/2021 11:37 AM, Andreas Rheinhardt wrote: > James Almer: >> On 8/22/2021 10:25 AM, Andreas Rheinhardt wrote: >>> The low overhead OBU format provides no means to resync after performing >>> a byte-based seek; in other words: Byte based seeking is just not >>> supported. >>> >>> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> >>> --- >>> The reason I didn't disallow this earlier was that there is one scenario >>> where such a seek would work: When one already knows the correct >>> position for the start of a frame. But otherwise it just doesn't work. >>> >>> Somehow ffplay tries byte based seeking even for formats with the >>> AVFMT_NO_BYTE_SEEK flag set. Line 2837 should probably be changed to no >>> longer do that. >>> >>> libavformat/av1dec.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/libavformat/av1dec.c b/libavformat/av1dec.c >>> index 88a3c325e4..37f21483b9 100644 >>> --- a/libavformat/av1dec.c >>> +++ b/libavformat/av1dec.c >>> @@ -428,7 +428,7 @@ const AVInputFormat ff_obu_demuxer = { >>> .read_packet = obu_read_packet, >>> .read_close = av1_read_close, >>> .extensions = "obu", >>> - .flags = AVFMT_GENERIC_INDEX, >>> + .flags = AVFMT_GENERIC_INDEX | AVFMT_NO_BYTE_SEEK, >>> .priv_class = &av1_demuxer_class, >>> }; >>> #endif >> >> LGTM. >> >> Maybe also AVFMT_NOTIMESTAMPS could be added. > > But isn't AV1 able to contain timestamps via its timing_info() and > metadata_timecode() structure? > > - Andreas But those are not values exported by the demuxer, which is what i interpreted the flag is about. But i see other raw demuxers with similar codec level timing information that don't set it, so i guess not.
diff --git a/libavformat/av1dec.c b/libavformat/av1dec.c index 88a3c325e4..37f21483b9 100644 --- a/libavformat/av1dec.c +++ b/libavformat/av1dec.c @@ -428,7 +428,7 @@ const AVInputFormat ff_obu_demuxer = { .read_packet = obu_read_packet, .read_close = av1_read_close, .extensions = "obu", - .flags = AVFMT_GENERIC_INDEX, + .flags = AVFMT_GENERIC_INDEX | AVFMT_NO_BYTE_SEEK, .priv_class = &av1_demuxer_class, }; #endif
The low overhead OBU format provides no means to resync after performing a byte-based seek; in other words: Byte based seeking is just not supported. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com> --- The reason I didn't disallow this earlier was that there is one scenario where such a seek would work: When one already knows the correct position for the start of a frame. But otherwise it just doesn't work. Somehow ffplay tries byte based seeking even for formats with the AVFMT_NO_BYTE_SEEK flag set. Line 2837 should probably be changed to no longer do that. libavformat/av1dec.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)