[FFmpeg-devel] [PATCH] http: try to detect live akamai streams and dont enable seeking for them

[Reimar Döffinger]

On Mon, Jun 04, 2012 at 03:14:20PM +0200, Michael Niedermayer wrote:
> On Mon, Jun 04, 2012 at 07:45:48AM +0200, Reimar Döffinger wrote:
> > On 4 Jun 2012, at 02:28, Michael Niedermayer <michaelni at gmx.at> wrote:
> > > On Mon, Jun 04, 2012 at 02:25:57AM +0200, Michael Niedermayer wrote:
> > >> On Sun, Jun 03, 2012 at 10:38:14PM +0200, Reimar Döffinger wrote:
> > >>> On Sun, Jun 03, 2012 at 06:43:33PM +0200, Clément Bœsch wrote:
> > >>>> On Sun, Jun 03, 2012 at 06:36:34PM +0200, Michael Niedermayer wrote:
> > >>>>> Fixes ticket1320
> > >>>>> 
> > >>>>> Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
> > >>>>> ---
> > >>>>> libavformat/http.c |    6 +++++-
> > >>>>> 1 file changed, 5 insertions(+), 1 deletion(-)
> > >>>>> 
> > >>>>> diff --git a/libavformat/http.c b/libavformat/http.c
> > >>>>> index 9a5ca62..28a777c 100644
> > >>>>> --- a/libavformat/http.c
> > >>>>> +++ b/libavformat/http.c
> > >>>>> @@ -57,6 +57,7 @@ typedef struct {
> > >>>>>     int multiple_requests;  /**< A flag which indicates if we use persistent connections. */
> > >>>>>     uint8_t *post_data;
> > >>>>>     int post_datalen;
> > >>>>> +    int is_akamai;
> > >>>>> } HTTPContext;
> > >>>>> 
> > >>>>> #define OFFSET(x) offsetof(HTTPContext, x)
> > >>>>> @@ -318,7 +319,8 @@ static int process_line(URLContext *h, char *line, int line_count,
> > >>>>>                 if ((slash = strchr(p, '/')) && strlen(slash) > 0)
> > >>>>>                     s->filesize = atoll(slash+1);
> > >>>>>             }
> > >>>>> -            h->is_streamed = 0; /* we _can_ in fact seek */
> > >>>>> +            if (!s->is_akamai || s->filesize != 2147483647)
> > >>>>> +                h->is_streamed = 0; /* we _can_ in fact seek */
> > >>>>>         } else if (!av_strcasecmp(tag, "Accept-Ranges") && !strncmp(p, "bytes", 5)) {
> > >>>>>             h->is_streamed = 0;
> > >>>>>         } else if (!av_strcasecmp (tag, "Transfer-Encoding") && !av_strncasecmp(p, "chunked", 7)) {
> > >>>>> @@ -333,6 +335,8 @@ static int process_line(URLContext *h, char *line, int line_count,
> > >>>>>         } else if (!av_strcasecmp (tag, "Connection")) {
> > >>>>>             if (!strcmp(p, "close"))
> > >>>>>                 s->willclose = 1;
> > >>>>> +        } else if (!av_strcasecmp (tag, "Server") && !av_strcasecmp (p, "AkamaiGHost")) {
> > >>>>> +            s->is_akamai = 1;
> > >>>>>         }
> > >>>>>     }
> > >>>> 
> > >>>> This reminds me of a similar issue Reimar fixed in MPlayer by adding
> > >>>> another exception (IIRC it was Youtube); this might be a fairly common
> > >>>> issue with some webservers, so maybe a more generic attribute could be
> > >>>> used? (broken_httpserver or something).
> > >>> 
> > >>> I'm not sure if FFmpeg's http implementation can detect it in some other
> > >>> way, however it is true that in MPlayer we have two other broken cases
> > >>> (and we lack this one). YouTube and MakeMKV are those.
> > >>> (see stream/http.c:758).
> > >> 
> > >> do you have testcases for the 2 ?
> > >> so we can check if ffmpeg needs such checks too ?
> > > 
> > > s/we/i/
> > 
> > Not sure about YouTube, though I guess any video would have to do?
> 
> tried but it seems the flv stream contains no index and our http
> doesnt support seeking by time
> mplayer itself fails to seek as well
> 
> not sure if i did something wrong ...

I admit I don't remember the test-case. My first suggestion would be to
try with the HTML5/webm versions of the videos if you haven't.
Though I would have thought that the flv streams should have an index,
otherwise how does flash player do seeking?