1.2.2b kills IE on accordion demo page

[joern]

I get weird errors in IE 6/7 on the accordion demo page, starting with an illegal argument, sometimes also a out of memory error.

Debugger says the illegal argument error is thrown in line 1077 (elem[ name ] = value;), with name="height" and value="NaNpx". Looks like a wrong height is calculated during an animation, resulting in a NaN.

To reproduce: Checkout trunk/plugins/accordion, update lib/jquery.js to latest, load accordion/demo/ in IE and click on top left accordion.

Works fine with 1.2.1.

[brandon]

The css.diff is the patch I'm currently testing locally to fix some width/height issues. Can you test to see if it fixes this ticket as well?

[brandon]

The height issue is fixed but I consistently get the "out of memory error" at page unload. It seems the unbind code is using to much of the limited memory during the unload event.

[brandon]

Actually it doesn't even just happen at unload. If I run the unload code at any time on the demo, I get the "outer of memory error".

[joern]

I can confirm that - height issue fixed, memory issue still there.

The unload handler gets all elements first:

jQuery(window).bind("unload", function() {
	jQuery("*").add(document).unbind();
});

Could it help to somehow reduce the selection? If I replace the "*" with eg. "a" on the demo page, the error doesn't occur anymore - but it leaks a lot of memory.

1.2.1 leaks a lot, too, but we handled that well enough in earlier versions. I've lost track of all the changes, but maybe we can backtrack some of them to see where the regression occured.

[brandon]

We used to keep a global cache of elements that had events bound to them. That is what we would use to fix the memory leak. However, if I remember correctly (and I know John does) it was causing a slew of its own issues in regards to memory and slow downs. We definitely need to find a way to limit the scope of the final unbind call.

[brandon]

John mentioned just going directly to the expando data cache and killing all the handlers that way. I think I'll give this a try when I get the chance ... just not sure when that will be. Feel free to try it out as well.

[davidserduke]

I'm not sure how that can work. The cache doesn't hold a reference to the elem does it? Or are you thinking about adding that?

[davidserduke]

The memory bug appears to be fixed with [4226]. If there is still an issue please reopen with a new test case.