[dojox.Chart2D][dojox.gfx] Confirmed memory leak - all browsers

[randycasburn]

Hi Eugene, et.al.

It seems some but not all of the entities created by these two classes do not get destroyed, removed, nullified, or otherwise expunged from the bowels of the interpreter 'onUnload()'.

The memory leak is real and measurable across three browsers I tested which were IE, FF, and Chrome.

Here is the test:

1)Using your favorite leak analyzer (I used sIEve), load this url:

​http://download.dojotoolkit.org/release-1.3.0/dojo-release-1.3.0/dojox/charting/tests/test_event2d.html

2)DO NOT roll over any of the "event triggers". You'll notice no orphans or circular references! If you navigate away from the page you WILL NOT be left with any leaks.

3)Navigate back to the test page and rollover the "event triggers" on the charts. This time when you navigate from the page, you'll notice the leaks.

This leads me to believe the event generator in dojox.Chart2D is causing the problem. I attempted to correct this problem by patching the connectToPlot() method as follows:

this.stackconnections = []; this was initialized with others above

connectToPlot: function(name, object, method){
  return name in this.plots ? this.stackconnections[name] = this.stack[this.plots[name]].connect(object, method) : null;
		},

and by adding a disconnect reference to the destroy method for the class thus:

dojo.forEach(this.stackconnections, function(con){dojo.disconnect(con);});

Clearly my analysis was incorrect as this patch has no effect on this problem. If you have any other insight, I don't mind helping you trouble shoot this problem as this is quite troubling.

thanks for taking a look into this.

Regards,

Randy

[Eugene Lazutkin]

It may be a general problem, not specific to charts.

[JayZ(zhouxiang)]

Besides, if we use shape.disconnect instead of dojo.disconnect, the problem still exists(still cause memory leak).

e.g. var k = shape.connect("onmouseover", this, function(e){

....

}); shape.disconnect(k);

[ben hockey]

i was playing with sieve and wanted to see what open tickets we had which referenced sieve. i came across this ticket and it looks like this problem went away in dojo 1.3.2 (it still existed in 1.3.1). was wondering if this ticket could be closed?

[Eugene Lazutkin]

Let's wait for a week, if there is a different evidence, and close it.

[cjolif]

I think the week delay is more than over :) Shall this be closed?