[Ffmpeg-devel] HTTP/1.1 Breaks ffserver/ffmpeg combo

[Ryan Martell]

On Jan 15, 2007, at 3:27 AM, Michael Niedermayer wrote:

> Hi
>
> On Mon, Jan 15, 2007 at 10:19:12AM +0100, Guillaume POIRIER wrote:
>> Hi,
>>
>> On 1/15/07, C.Ren Boca <crboca32 at yahoo.com> wrote:
>>> I think the problem is in libavformat/http.c
>>>
>>> I diffed the files between the two svn revisions that I have,    
>>> 7486, and
>>> 7529, and there are major differences. ffserver and ffmpeg show  
>>> no diff.
>>>
>>> I really like the ffmpeg/ffserver combo. At one point I had it  
>>> working
>>> pretty damn good.
>>
>> Please use binary search to find the exact revision that broke it.  
>> You
>
> my guess would be that it was the http seeking patch which broke  
> it ...
> if so maybe ronald has a clue why?
>
> [...]

Not sure if this is relevant, but this may be the reason that my  
parsing wasn't working either.  http.c sends 1.1 headers now, whereas  
before it was 1.0.  In my case, that meant that the server was  
closing the connection after sending an entire .asx (playlist) file.   
In the new version, it wasn't closing the connection, so I was  
hanging waiting to read data from the connection in the format probe...

My patch to http.c that would respect the Content-Length: header and  
generate an eof was rejected.

The only other thing that might be able to fix this (if this is  
indeed the same issue) is adding "Connection: Close" to the get  
request, although that might screw up the http_seeking stuff.   
Changing the version back to 1.0 would make the seeking stuff be in  
non-conformance.

For more on this, see  [Ffmpeg-devel] HTTP probing issue... [PATCH].

Just my 2 cents.

-Ryan