[FFmpeg-devel] [PATCH] sdp: Add a framesize attribute to H263 SDP descriptions

Luca Abeni lucabe72
Mon Dec 20 15:57:58 CET 2010

On 12/20/2010 02:08 PM, Martin Storsj? wrote:
> Hi Luca,
> On Mon, 20 Dec 2010, Luca Abeni wrote:
>> On 12/20/2010 01:42 PM, Martin Storsjo wrote:
>>> While not mentioned in RFC 4629, this is required for H263 in
>>> 3GPP TS 26.234.
>> Is it really mandatory? I seem to remember it was optional...
>> But I might be wrong.
> Quoting from the 3GPP spec:
>      For H.263 streams, a PSS server shall include the "a=framesize" field
>      at the media level for each stream in SDP, and a PSS client should
>      interpret this field, if present. Clients should be ready to receive SDP
>      descriptions without this attribute.
> So depending on the interpretation of "shall", it's more or less
> mandatory.
According to RFC 2119, "shall" means that the "a=framesize" attribute is
mandatory. So, I still think that it would be nice to be able to control
the generation of this attribute (something like "make a PSS compliant
SDP"), but ok.

>> Anyway, I think we already discussed this some time ago...
>> I think setting framesize in this way is not correct (it should
>> be the maximum possible frame size in the session).
> Yes, we've discussed it before, but then I didn't know about any players
> actually requiring it - now I've stumbled upon such ones.
Well, if there are players who need this attribute, this must be done.
I just suggest to add a small comment (to avoid confusing other people
reading the code).

> Do we btw normally support muxing streams with varying frame size? We've
> got the same issue for all other formats that transmit the frame size
> implicitly in extradata in the SDP (such as H264 or MPEG4).
If you use in-band global headers, it works ok. I tried it in the past
with H.264 and in-band PPS/SPS, and it worked (I think this kind of
things are done by DVB-H for changing the resolution on advertisements,
or similar).


More information about the ffmpeg-devel mailing list