599568 - physicSketch Experiment fails to retain drawn objects

Status: RESOLVED DUPLICATE of bug 597378

(Core :: JavaScript Engine, defect)

Description

[Andrew Condinho]

User-Agent:       Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b7pre) Gecko/20100924 Firefox/4.0b7pre
Build Identifier: Build identifier: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b7pre) Gecko/20100924 Firefox/4.0b7pre

Basically, once you launch the experiment, whenever you attempt to draw any shapes on the canvas it shows the lines and then they disappear. After about the 3rd attempt it stops even drawing the lines at all. 

This issue does not occur in Chrome, or Firefox 3.6, and I also tested it on 32/64-bit machines.

Reproducible: Always

Steps to Reproduce:
1. Launch experiment.
2. Draw a shape in the canvas.
3. Objects fail to appear.
Actual Results:  
The lines that you draw are visible while drawing, however when you let go of the mouse button to complete the shame they disappear. After 3 tries it stops even drawing the preliminary lines. 

Expected Results:  
The shapes you draw should stay retained and drop according to the programmed physics.

Comment 1

[Boris Zbarsky [:bzbarsky]]

It's throwing on this line:

    context.moveTo(point.x, point.y);

which means one of those is a NaN or Infinity.

Can you narrow down when this problem first appeared?  Doing that using tracemonkey nightlies might be a good idea...  What's needed is the http:// URL from about:buildconfig for the nightlies immediately before and after the problem appears.

Comment 2

[David Humphrey (:humph)]

This one uses box2d.js too, just like the Google Gravity one -- bug 595541

Comment 3

[Andrew Condinho]

I haven't used tracemonkey before how would I go about doing this?

Comment 4

[Boris Zbarsky [:bzbarsky]]

Just look for the tracemonkey directories in http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/

Comment 5

[Boris Zbarsky [:bzbarsky]]

Though given comment 2, starting with m-c nightlies might be better (and in fact just starting with the bug 595541 range and seeing whether it applies here).

Comment 6

[Andrew Condinho]

I checked the m-c nightlies at those bugs ranges and for

Mozilla Central
2010-08-31-04-mozilla-central - Experiment works
2010-09-01-04-mozilla-central - Experiment works
2010-09-01-09-mozilla-central - Experiment works

So not sure if they are related, I'll start at the latest release and go back until I find the builds where it breaks.

Comment 7

[Andrew Condinho]

Ok I've narrowed it down to these 2 m-c builds

Mozilla Central
2010-09-11-04-mozilla-central - Experiment works
2010-09-12-04-mozilla-central - Experiment doesn't work

I then tested the tracemonkey builds and had to go back another day to find one that worked

Tracemonkey
2010-09-10-04-tracemonkey - Experiment works
2010-09-11-04-tracemonkey - Experiment doesn't work

Comment 8

[Boris Zbarsky [:bzbarsky]]

ok.  For those tracemonkey builds, what are the changeset ids?  See comment 1.

Comment 9

[Andrew Condinho]

I'm hoping these are the ids you're looking for...

2010-09-10-04-tracemonkey - http://hg.mozilla.org/tracemonkey/rev/a1f43f4ef565
2010-09-11-04-tracemonkey - http://hg.mozilla.org/tracemonkey/rev/b66534953ae4

Comment 10

[Boris Zbarsky [:bzbarsky]]

Yes, perfect.

Regression range: http://hg.mozilla.org/tracemonkey/pushloghtml?fromchange=a1f43f4ef565&tochange=b66534953ae4

Which happens to include a mozilla-central merge, and not that many other things.  You're very sure about the moilla-central regression range in comment 7, right?  _That_ range has the tracemonkey merge that landed jaegermonkey in it...

Comment 11

[Benoit Jacob [:bjacob] (mostly away)]

(In reply to comment #10)
> _That_ range has the tracemonkey merge that landed jaegermonkey in it...

I can confirm that this is jaegermonkey-related: if I go to about:config and set javascript.options.methodjit.content to false (which means disable jaegermonkey), that fixes the problem.

Comment 12

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

This eventually gives:

Error: uncaught exception: [Exception... "An invalid or illegal string was specified"  code: "12" nsresult: "0x8053000c (NS_ERROR_DOM_SYNTAX_ERR)"  location: "http://physicsketch.appspot.com/js/physicsketch.js Line: 335"]

Error: (void 0) is undefined
Source File: http://physicsketch.appspot.com/js/box2d.js
Line: 292

and stops working.  The first one is triggered by itself if you draw a shape that doesn't have endpoints connected; after that you can't draw any more.  The second is triggered if you try to draw an illegal shape, or if something happens inside box2d that causes its processing to abort (e.g. I've had the physics animation just die on me in the middle of an object falling).

Comment 13

[David Humphrey (:humph)]

> Error: (void 0) is undefined
> Source File: http://physicsketch.appspot.com/js/box2d.js
> Line: 292

That's the same thing happening in bug 595541

Comment 14

[Jan de Mooij [:jandem]]

May also be related to bug 597378. The pinball demo also uses box2d...

Comment 15

[David Anderson [:dvander] - inactive, e-mail if emergency]

Thanks, Jan, I've verified this to be fixed by the same dependency as bug 597378.