[vlc-devel] Getting stuck in stream_Read()
fyhuel at viotech.net
Wed Apr 4 17:56:11 CEST 2012
On Wed, Apr 4, 2012 at 5:05 PM, Casian Andrei <skeletk13 at gmail.com> wrote:
> 2012/4/4 Frederic YHUEL <fyhuel at viotech.net>:
>> On Fri, Mar 30, 2012 at 11:59 AM, Frederic YHUEL <fyhuel at viotech.net> wrote:
>>> 2012/3/30 Rémi Denis-Courmont <remi at remlab.net>:
>>> > On Fri, 30 Mar 2012 11:29:15 +0200, Frederic YHUEL <fyhuel at viotech.net>
>>> > wrote:
>>> >> Actually it gets stuck in net_Read(), and more precisely inside the
>>> >> call to poll().
>>> >> Since net_Read() calls poll() with no timeout, I'm not sure it is a
>>> >> bug. But I have a really poor understanding of that stuff.
>>> >> Is it possible to blame the server? In that case, would it be possible
>>> >> to have a timeout mechanism, like a net_Read flavour which calls
>>> >> poll() with a timeout? (and the associated stream_Read variant?)
>>> > Getting stuck in net_Read() is fine if the server is not sending any
>>> > data.
>>> > That is not a case for a timeout, unless the application layer protocol
>>> > specifies timeouts.
>>> > Of course, you are not supposed to get stuck to the point that the stop
>>> > button does not act timely.
>>> > But that is a case for some for of interruption/exception mechanism, not
>>> > for ****ing^W^H polling timers.
>>> Ok, thank you very much Rémi.
>> Very strangely, I don't have this problem on my personal machine which
>> runs Debian wheezy 64 bits. At work I have Ubuntu Oneiric 32 bits. I
>> installed the 64 bits kernel of wheezy
>> (linux-image-3.2.0-2-amd64_3.2.12-1_i386.deb) on this machine, and the
>> problem is not there anymore.
>> Is it possible that (32 bits) Ubuntu kernel have a buggy poll() ? I
>> guess that launchpad would have been flooded by bug reports in that
>> case, so I don't understand... any idea?
> I am witnessing this kind of behavior when using the internet
> connection from my campus. It's affecting only certain addresses and
> it seems it's ISP related. I have to mention that everything that uses
> http is affected.
> While debugging VLC when it tried to contact certain problematic
> addresses, it got stuck in net_Read(), just like you describe.
> So, I am suggesting that this might be an ISP issue. Any network
> debugging I tried was futile, maybe you will have more luck.
I should do more tests, but it seems to me that it is ISP independent.
This Smooth Streaming server do have some problems though: for
example, sometimes, typically after 16 min, the TCP connection is
closed and I cannot reconnect. I thought the server did that on
purpose after a certain time, because it is just a demo server, but
now I'm not sure (I got 45 min of playback this afternoon).
So I think it's mainly the server's fault, like Rémi suggested.
More information about the vlc-devel