[webkit-gtk] Memory leaks creating and removing webviews

Brian Holt brian.holt at samsung.com
Fri Jun 14 01:43:33 PDT 2013


>> Also, running a page making XMLHttpRequests on the DRT under valgrind 
>> shows that there are no "leaks" in the sense that memory is no longer
available.
>> What is far more likely is that the references are being held in some 
>> structure (HashTable?) and upon exit are properly freed.
>
>This is interesting, maybe it's just a matter of bad refcounting. Will try
to take a look at it.

Reading that again I realised I didn't explain that very well.  

Valgrind shows no leaks, meaning that the memory is being properly freed
upon exit.  But that doesn't help people who see a steady increase in the
amount of memory consumed by WK while doing these XMLHttpRequests.  So, what
is most likely happening is that the refs are held in some structure and not
being freed at the right point.

I'm already trying to look into the issue, but as they say, patches are
welcome :-)

Cheers
Brian

-----Original Message-----
From: Sergio Villar Senin [mailto:svillar at igalia.com] 
Sent: 14 June 2013 09:35
To: Brian Holt
Cc: webkit-gtk at lists.webkit.org
Subject: Re: [webkit-gtk] Memory leaks creating and removing webviews

En 14/06/13 10:29, Brian Holt escribiu:
>> I am not saying that there are no leaks, but it's better to use tools 
>> like
> valgrind as it can properly point to the sources of potential memory
leaks.
> 
> I've just finished running all the LayoutTests (30K+) with the 
> DumpRenderTree under valgrind
> (https://bugs.webkit.org/show_bug.cgi?id=116317) and the good news is 
> that there are only a few leaks in WK Gtk, mostly one-off leaks in 
> external libraries like enchant, fontconfig, GL etc.

Cool

> Also, running a page making XMLHttpRequests on the DRT under valgrind 
> shows that there are no "leaks" in the sense that memory is no longer
available.
> What is far more likely is that the references are being held in some 
> structure (HashTable?) and upon exit are properly freed.

This is interesting, maybe it's just a matter of bad refcounting. Will try
to take a look at it.

BR




More information about the webkit-gtk mailing list