infinity appearing in p2 physics

[Cryptonomicon]

A really hard problem to trace so please put up with my inability to post a simple example, but I am getting NaN's appearing in debug mode both in firefox an chrome that trace themselves to (so far) to infinities appearing deep within the Gauss Seidel GSSolver in p2 physics.

 

It appears when I have a body on the screen and then I click on the display. The click throws the body to 0,0 which I assume is some form of default for NaN's that appear - that are in turn created by infinity in the solver.

 

It does not appear in normal mode - only when I am debugging at around line 66290 in phaser .js 

 

Anyway has anyone had problems with bodies teleporting to the origin when the screen is clicked when using p2.js - I suspect it is an initialization problem NaN's and infinity should not appear in a maths solver - or at least without an exception

 

0

Equation { minForce=-1.7976931348623157e+308, maxForce=1.7976931348623157e+308, bodyA={...}, more...}  

G

Float32Array { 0=-1, 1=0, 2=-4.032659530639648, more...}  

a

Infinity  

b

0.9411764705882353  

bodyA

Object { id=5, world={...}, shapes=[6], more...}  

bodyB

Object { id=6, world={...}, shapes=[1], more...}

 

[Cryptonomicon]

The cause is line 69398 in phaser.js

 

/**

* @method Phaser.Physics.P2#update

*/

update: function () {

 

if (this.useElapsedTime)

{

this.world.step(this.game.time.physicsElapsed);

}

else

{

this.world.step(this.frameRate);

}

 

},

 

which when you are debugging has a

 

this.game.time.physicsElapsed that is 0.

 

game

Object { id=0, width=960, height=640, more...}  

lastTime

1398305112131  

msMax

388  

msMin

0  

now

1398305112131  

pauseDuration

12263  

pausedTime

0  

physicsElapsed

0  

time

1398305112131  

timeToCall

0

 

 

 

That blows the following in the

 

this.epsilon equation

 

in the function around line 62918 . (h is zero in the denominator)

 

/**

* Compute SPOOK parameters .a, .b and .epsilon according to the current parameters. See equations 9, 10 and 11 in the <a href='http://www8.cs.umu.se/kurser/5DV058/VT09/lectures/spooknotes.pdf">SPOOK notes</a>.

* @method update

*/

Equation.prototype.update = function(){

var k = this.stiffness,

d = this.relaxation,

h = this.timeStep;

 

this.a = 4.0 / (h * (1 + 4 * d));

this.b = (4.0 * d) / (1 + 4 * d);

this.epsilon = 4.0 / (h * h * k * (1 + 4 * d));   <-- This is the culprit. It expects h to be non zero

 

this.needsUpdate = false;

};

 

As it only occurs when debugging I might just give it a timeStep++  somewhere

 

 

 

Am not sure if it is a P2 bug or phaser bug so will let wiser minds than me work that out.

[george]

Can you provide a minimal example for this?

Could you then try one of those settings?

 

//You use useElapsedTime at the moment? Try setting it to false:game.physics.p2.useElapsedTime = false//or with useElapsedTime enabled:game.stage.disableVisibilityChange = truegame.time.deltaCap = 1/30

Regards

George

[Cryptonomicon]

I found I needed elapsed time due to variances in browser speed - I was getting anything from 10fps to 50fps - and elapsed time fixed that.

 

in  the end I overrode the default update

 

this.game.physics.startSystem(Phaser.Physics.P2JS);  //start physics as normal
this.game.physics.p2.update=this.replacementP2PhysicsUpdate;  //swap out the standard update

 

 

//and swapped in my own as below - the change is bold in blue

 

my.Class.prototype.replacementP2PhysicsUpdate=function () {
    
    if (this.useElapsedTime)
    {
        this.world.step(this.game.time.physicsElapsed+0.0001);
    }
    else
    {
        this.world.step(this.frameRate);
    }
};

 

/* hack ends here */

 

and all seems fine.

 

Tell you what though, in thirty years of coding phaser.js is probably the nicest to read through I have ever seen. To even be able to trace such an obscure error in 60,000 lines of code using the **cough**  debugger  **cough** in firefox in a couple of hours is saying a lot given it is written in the untyped mess that is JavaScript.

[george]

Well, you just have created a so called monkey patch

By overriding the core of phaser you exclude yourself from updating phaser or you have to maintain all of your 'monkey patches'. The main problem here is that 

physicsElapsed has a value of zero. That's clearly a bug an could be related to the first timer tick of phaser. Is the error happening just after the start ? Then you could try to defer the creation of your physics for some milliseconds with a timer/setTimeout? Could you please try this ? Or maybe create an issue on phaser's github repository? If it's a bug it will happen for other people so we should fix it- maybe even with that type of fix you just created

[george]

That's funny. Just now I got a related error with physics elapsed. I provided a fix that may also fix your problem.

https://github.com/photonstorm/phaser/pull/758

 

Regards George

[Cryptonomicon]

that is how the universe works today must be physics timing day.

 

It seems to be only in debug mode - the debugger "freezes time" somewhat when you enter the debugger and when you reenter the browser comes back to where it left off.

 

If it was C++ I would say it looks like the environment getting pushed onto a stack on debug entry and off again on exit - but this ain't C++ so I wont say that.

 

maintaining this one will be easy - has a great big //TODO: in it telling me to check future releases.

 

thanks