Re: Memory leaks when embedding and using JavaXPCOM

Hi,

I have made some improvements in JavaXPCOM.
They are related to a more efficient way of disposing native resources
from Java proxies when they are garbage collected.

I've been following the guidelines of this nice article:
http://java.sun.com/developer/technicalArticles/javase/finalization

How can I contribute them back to the community?
Should I open a bug and attach a patch ? If I do so, will the change
be merged with trunk, after being evaluated?


On 14 Jul., 09:50, David P=E9rez <[email protected]> wrote:
> Hi Benjamin,
>
> First of all, thanks for your valuable answer.
>
> On 1 Jul., 16:33, Benjamin Smedberg <[email protected]> wrote:
>
> > On 7/1/09 5:47 AM, David P rez wrote:
>
> > > We ve made a simple XULRunner app, coded in Java.
> > > We ve observed that it has huge memory leaks.
> > > Rewriting the same app in Javascript, we observe no leaks.
>
> > That's certainly possible. JavaXPCOM doesn't have an owner and I may
> > consider removing it in the next release if we can't find an owner.
>
> JavaXPCOM, even when leaky, is very useful, so I wouldn't drop it in
> the next release.
>
>
>
> > > So, we suspect there is some problem with JavaXPCOM. =A0Maybe the
> > > problem is related with the cycle collector.
> > >https://developer.mozilla.org/En/Interfacing_with_the_XPCOM_cycle_col.=
..
>
> > Is your code creating object cycles. In most cases if you are in contro=
l of
> > the code you can manually avoid object cycles by nulling references when
> > they are no longer needed. I suspect adding cycle-collection to the
> > JavaXPCOM interface would be very difficult, because it would require
> > knowing all the object edges through the Java GC heap.
>
> I have made simply an app that loads an URL in a browser, waits
> through events till the page is loaded, and emulates mouse clicks in
> certains part of the page. =A0In Java code I need to make XPCOM calls to
> objects. =A0For reacting to events (page loaded), I need to create Java
> objects that implements certain nsIXXXX interface.
>
> Maybe the cycling collector isn't the real problem, it's just a
> supposition.
>
> I have tried to apply this patch and the memory leaks have improved a
> lot.https://bugzilla.mozilla.org/show_bug.cgi?id=3D501930
>
> > --BDS