03.07.11 23:06, Dan Dennedy написав(ла): > On Sun, Jul 3, 2011 at 12:48 PM, Maksym Veremeyenko<ve...@m1stereo.tv> wrote: >> On Sun, July 3, 2011 10:01 pm, Dan Dennedy wrote: >>> 2011/7/3 Maksym Veremeyenko<ve...@m1stereo.tv>: >>>> Hi, >>>> >>>> playing files in a loop mode in melted cause a leak of file descriptors >>>> that >>>> seen well from /proc/<process id>/fd/<file no>. >>>> >>>> the problem comes from avformat's producer re-open code that lost >>>> release of >>>> audiostream context. >>>> >>>> attached patch fix this behaviour... >>> >>> This patch is wrong in a couple of ways. It completely alters the >>> locking mechanisms, and I will be surprised if it works under >>> multi-threaded access such as with kdenlive or with parallel consumer. >>> Your attempt to refactor to common code is noble but naive with >>> respect to the locking requirements, which are different depending >>> upon the state of the locks when these similar actions are invoked. >>> Secondly, it breaks the user-selected audio and video tracks because >>> in producer_open() they are reset to their defaults. That is why they >>> were saved before producer_open() and re-applied afterwards. >> >> updated patch attached.... > > Please test my latest commits. In my analysis and testing while > working on these fixes, I found it is not at all safe to close the > audio format and codec contexts at this point - think parallel.
there are no leaks noticed with latest git, thanks. > > No offense to you, but I am not too comfortable with other people's > hands getting into avformat producer. That is a very intricate piece > of work, and there are a lot of special requirements that only I know > about. Thank you for trying, though! It will be nice for someone else > to become knowledgeable about this code. If you look at the git log, > you will see that I have put major effort to refactoring its code > periodically in order to tame this wild beast. > If you want to contribute something here, I would accept a patch that > refactors my recent change that added a take_lock parameter to > producer_open() into a new function producer_open_with_lock() that > inside calls producer_open() - similar approach to your patch. > ok -- ________________________________________ Maksym Veremeyenko ------------------------------------------------------------------------------ All of the data generated in your IT infrastructure is seriously valuable. Why? It contains a definitive record of application performance, security threats, fraudulent activity, and more. Splunk takes this data and makes sense of it. IT sense. And common sense. http://p.sf.net/sfu/splunk-d2d-c2 _______________________________________________ Mlt-devel mailing list Mlt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/mlt-devel
