[FFmpeg-devel] [FFmpeg-cvslog] Merge commit '81306fd4bdeb5c17d4db771e4fec684773b5790f'

[Hendrik Leppkes]

On Mon, Mar 14, 2016 at 10:12 AM, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
> On Mon, Mar 14, 2016 at 9:58 AM, Hendrik Leppkes <h.leppkes at gmail.com> wrote:
>> On Mon, Mar 14, 2016 at 5:07 AM, Simon Thelen <ffmpeg-dev at c-14.de> wrote:
>>> On 16-02-16 at 17:27, Derek Buitenhuis wrote:
>>>> ffmpeg | branch: master | Derek Buitenhuis <derek.buitenhuis at gmail.com> | Tue Feb 16 16:26:49 2016 +0000| [d0fc5de3a643fe7f974ed14e410c2ac2f4147d7e] | committer: Derek Buitenhuis
>>>>
>>>> Merge commit '81306fd4bdeb5c17d4db771e4fec684773b5790f'
>>>>
>>>> * commit '81306fd4bdeb5c17d4db771e4fec684773b5790f':
>>>>   hls: eliminate ffurl_* usage
>>>>
>>>> Merged-by: Derek Buitenhuis <derek.buitenhuis at gmail.com>
>>>>
>>>> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d0fc5de3a643fe7f974ed14e410c2ac2f4147d7e
>>>> ---
>>> At least this change in hls.c seems to be incorrect, it breaks setting
>>> the user-agent option for hls. Initially the correct user agent is used
>>> when querying the hls, but when refreshing the playlist the user agent
>>> is reset to Lavf/57.25.100. Reverting this specific hunk seems to fix
>>> the issue. I assume the other changed call to update_options in open_url
>>> is also broken, but didn't test.
>>>
>>
>> Your version seems to be old, the current code has s->pb->opaque as an
>> AVIOInternal in hls.c, and uses another struct entirely to store
>> options.
>> Can you verify that current git master still suffers from the same problem?
>>
>
> After some closer look, it appears there were some more wrong changes made:
>
> 9f9ed79d4cb40e5d9093899f8a79086ff23da844 should just be flat out
> reverted. It introduces some new weird context thats never set
> anywhere, so its always NULL and does nothing?
> It also misses the key point that it actually needs to talk to the
> underlying http protocol to get the user-agent and cookie options out
> of it.
>
> And the update_options calls in question should just use s->pb as
> their context, and update_options changed to automatically search
> child contexts, which should then reach all the way down to the
> protocol and retrieve the appropriate options.
>

Posted a patch set to address a bunch of HLS regressions.

- Hendrik