Re: [fyi] Memory leak in Piggy Bank solved!

From: David Huynh <dfhuynh_at_csail.mit.edu>
Date: Fri, 19 Aug 2005 06:31:49 -0400

Thank you! I'm really really glad you figured out the bug! :-) Methinks,
I should hide over here in China until the rest of the bugs are fixed :-D

David


Stefano Mazzocchi wrote:

> This is to inform you that as of this afternoon, the SVN trunk of the
> piggy-bank repository contains a version of PB that does not leak
> memory, at least on Deerpark.
>
> I still have to test it on firefox, but I think it should work fine
> there as well.
>
> For the technically curious, the leak was due to a run-away pointer in
> the overlay.onLocationChange() function that kept a reference of the
> tab browser object even after the tab had been closed.
>
> This meant that no tab was ever garbage collected when piggy-bank was
> in place (pretty nasty and shows why firefox became unresponsive after
> a while, only for heavy tab users since the tabs were finally garbage
> collected when the window was closed).
>
> Yes, it was all David's fault :-), but I don't blame him for having
> noticed... he's not a heavy tab users, apparently and it took me 3
> weeks of patient instrumentation and analysis of firefox internals +
> several popular extentions (greasemonkey, for example) to understand
> everything that went on.
>
> But I'm happy I did, now I understand everything about the code (and a
> ton more about firefox and how to program the moz platform) and I was
> able to fix a number of other minor issues that I found during my leak
> hunt.
>
> - o -
>
> Next in my todo list is to move on the java side of things and
> understand why PB takes so much to start when first installed and try
> to avoid it by either starting a second thread to do the calculations
> in background while firefox starts, or precalculate the indexes and
> triple store when building the distribution.
>
> In the process, I'll try to update to the latest version of Sesame,
> even if might not be that painless since SerQL is apparently back
> incompatible :-(
>
> Suggestions on this matter are highly appreciated, btw.
>
Received on Fri Aug 19 2005 - 10:27:57 EDT

This archive was generated by hypermail 2.3.0 : Thu Aug 09 2012 - 16:39:18 EDT