[webkit-gtk] webkit-gtk - memory leak?

[Toni Koski]

On 11/02/2010 01:22 PM, Sergio Villar Senin wrote:
> En 02/11/10 09:23, Sergio Villar Senin escribiu:
>    
>> En 02/11/10 08:33, Toni Koski escribiu:
>>      
>>> No effect with g_free(). Still leaking.
>>> Yes, I have tested with uzbl also.  With "www.hs.fi" I can see that the
>>> used memory increases when the page is reloaded again and again :-(
>>>        
>> Maybe the problem is related to the new HTTP cache we added recently.
>> Even if you do not use it, you'll be using the new stream based loading
>> code needed by cache. I can help with that, just ping me (sergio) on IRC.
>>      
> After investigating a little bit it seems that it's not cache's issue.
> BTW running your example with valgrind gives me the following outcome
> after 5 reloads:
>
> ==23248== LEAK SUMMARY:
> ==23248==    definitely lost: 8,804 bytes in 36 blocks
> ==23248==    indirectly lost: 21,540 bytes in 1,051 blocks
> ==23248==      possibly lost: 1,206,326 bytes in 9,118 blocks
> ==23248==    still reachable: 758,669 bytes in 4,566 blocks
> ==23248==         suppressed: 14,516 bytes in 270 blocks
>
> This includes all the libraries used by webkitgtk. There is only 8k of
> definitely lost memory. After 20 reloads I got this:
>
> ==23488==    definitely lost: 12,817 bytes in 86 blocks
> ==23488==    indirectly lost: 28,500 bytes in 1,387 blocks
> ==23488==      possibly lost: 1,404,947 bytes in 9,443 blocks
> ==23488==    still reachable: 1,051,558 bytes in 4,774 blocks
> ==23488==         suppressed: 15,988 bytes in 297 blocks
>
> So 12k of definitely lost memory after 20 reloads. Most of the possibly
> lost stuff is related to font handling as far as I saw. What is the
> "huge" leak you identified? What tools are you using?
>
> Br
> _______________________________________________
> webkit-gtk mailing list
> webkit-gtk at lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-gtk
>    

I have not made any "professional" memory tracking. I have just 
monitoring RSS size with "ps" -command and top. Some of test I have made 
with "smem.pl" Perl script, with is reading memory usage from 
/proc/$PID/smap file and generates human readable reports.
Yes, I know RSS is not whole truth. But what I can see in real world, is 
that the test-program seems to use more and more merory until system 
starts to swap and freeze. Website "www.hs.fi" seems to be very bad in 
point of view mine tests. If I loop my program without any "reload", 
"reload_bypass_cache" or "load_uri", I cant see any memory leak.

/toni