[FFmpeg-devel] Issue with ogg page termination on full last page with even last segment size

[Peter Zebühr]

Ping!

I would appreciate some input on this, a bit more digging and I have found that in the lavf-ogg flake test the last packet that arrives and causes the regression with my patch looks like this:
	(AVPacket) $1 = {
	  buf = 0x0000000103105740
	  pts = 44100
	  dts = 44100
	  data = 0x0000000103105c40 ""
	  size = 0
	  stream_index = 0
	  flags = 1
	  side_data = 0x0000000103105ff0
	  side_data_elems = 1
	  duration = 0
	  pos = -1
	  convergence_duration = 0
	}

Currently wondering if it makes sense to send in a zero sized packet for muxing or not. That is, should I try to fix that or should I rather try to compensate for this in oggenc.c?

Regards,
Peter

> On 15 Dec 2020, at 14:35, Peter Zebühr <peter at zebuhr.se> wrote:
> 
> Hello!
> 
> I have stumbled on an issue with the ogg page termination in a probably quite rare case.
> If the last page ends up full (255 segments) and the last packet that is written is an even multiple of 255 that needs to be terminated with a lacing value of zero that spills over to the next page, then oggenc fails to write a last page that contains the zero lacing value and libogg will fail to detect the end of stream when parsing the produced file. Example:
> 
> 1. Last page contains 253 segments
> 2. Packet arrives of size 510 bytes
> 3. Packet it split into three segments of sizes [255, 255, 0]
> 4. Page gets the two segments [255, 255] and is full
> 5. A new page is created containing only the [0] lacing value but no actual data
> 6. The last page is ignored when completing the stream (because it contains 0 bytes data)
> 
> I have a patch that fixes this (attached) in the case that I had, but... one of the fate tests fail (lavf-ogg) and I need some help to reason about what the right thing to do is. 
> 
> The reason the test fails is because one extra page is created with my patch. The reason why one extra page is created is because the previous page is completed as it hits the "pref_duration" setting of 1 second (https://github.com/FFmpeg/FFmpeg/blob/master/libavformat/oggenc.c#L261) and then write_packet is called with one more packet of size 0 bytes. The questions I have are:
> * Is this a reasonable regression? From the ogg perspective it sounds kind of reasonable that a 0 byte packet would result in a 0 byte segment but not sure why write_packet is even called with a 0 byte packet.
> * If this is not a reasonable regression I could either:
>   - short circuit handling of a 0 byte packet and make it produce 0 segments in ogg_buffer_data
>   OR
>   - tweak the logic in the proposed change to write_trailer to make it consider segment_count only if the ongoing page is a continuation of the previous page.
>  (OR perhaps it is another bug that a 0 byte packet is written?)
> 
> 
> / Peter
> 
> <0001-Fix-ogg-page-termination-on-even-last-packets.patch>_______________________________________________
> 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".