On Fri, Nov 9, 2012 at 10:20 AM, Dan Dennedy <d...@dennedy.org> wrote:
> thank you for the patches...
>
> On Thu, Nov 8, 2012 at 1:08 AM, Maksym Veremeyenko <ve...@m1stereo.tv> wrote:
>> Hi,
>>
>> *0001-fix-leak-on-allocated-sax-context.patch* during initializing xml
>> parser context libxml allocate sax context variable. /producer_xml.c/
>> reallocate it again and rewrite pointer to it that cause mem leak. so
>> possible solution is restore it back before context releasing. also that
>> patch fix another context freeing and try to free a document root that is
>> related unreleased after xmlFreeParserCtxt.
>
> looks ok, but not yet tested
>
>> *0002-close-skipped-frame.patch* - valgrind mem checker slowdown code
>> execution dramatically and that cause dropping frame. that could cause
>> memory leak on loosing skipped frame, so i think we fix release them to
>> avoid memory leak.
>
> rejected, you cannot close the frame here. There is more to a frame
> than its image. Typically, when a frame is "dropped," it is only the
> image processing that is skipped but the audio is still used.
>
>> *0003-reorder-startup-stopping-preroll-thread.patch* - that patch avoid some
>> race condition that happen then /preroll/ thread is still executing, but
>> consumer is in stopping state. in that state /ScheduleNextFrame/ cause
>> running consumer thread start again.
>
> looks ok, but not yet tested
>
>> *0004-partially-fix-leak-of-service-allocated-in-on_start_.patch* - service
>> variable from stack during sax parsing is not freed. that patch fix only
>> cases i found, but i think there are more
>
> pending further review and analysis
>
>> +
>> + if ( track )
>> + {
>> + mlt_service_close( track );
>> + free( track );
>> + }
>>
>> should be added to *on_end_xxxxx* handler...
I applied patches 1, 3, and 4. I looked a little for additional cases
addressed in patch 4, but I only found one that applied to the
consumer element, which I addressed.
P.S. when running in valgrind and not checking a consumer or
frame-dropping in particular, you could use "-consumer null
real_time=0" (or negative real_time values to check those threading
models).
--
+-DRD-+
------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_nov
_______________________________________________
Mlt-devel mailing list
Mlt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlt-devel
