Jordi - IlMaestro - Ministral wrote:
> - No more leaks... really ? Well, I'm not sure. This distribution does not leak
> in the examples ( those that work ) as it used to do. Making sure all
> eventListeners are removed on layer deletion seems enought to fix the problem.
> There was no need for MEthod.destroy() either. However, I have tested it with a
> very complex website and it continues leaking. A little less, but still a lot.
> Depending on where does my investigation takes me, this method may return. I'd
> appreciate anyone running memory-leaking tests.
I actually run on a modified version of the dynapi which has added code
to kill memory leaks. As you said, it only leaks when you do more
advanced stuff - just creating layers doesn't leak at all. Mouseevents
leak, although that's easy to kill. I'm guessing there might be some
other slight leaks around related to circular references in dynapi or
back-references in HTML elements, or possibly widget inheritance. Our
production website runs on my modified version of dynapi and hardly
leaks at all. It chews up about 1MB more memory on the /first/ refresh,
then keeps pretty much stable on memory use. This is on a real heavy
website with a lot of dynamic content, which used to leak about 6MB on
every refresh. I'm gonna continue to look at how things work in dynapi
regarding references and stuff, to see if I can pound out the initial
1MB leak as well. Time is the killer here.
Basically, all I've done is remove references in HTML-objects to dynapi
objects (including eventhandlers), as well as remove some circular
references within dynapi.
As soon as DynAPI X is stable and integrated with DynAPI 2, I'll want to
start using that for development and see what can be done about leaks
there.
Cheers,
Daniel
--
Daniel Aborg <[EMAIL PROTECTED]>
T: 0207 445 447 M: 07720 29 44 40
_______________________________________________
Dynapi-Dev mailing list
[EMAIL PROTECTED]
http://lists.sourceforge.net/lists/listinfo/dynapi-dev
