New Version of Phys2D

[CommanderKeith]

That looks amazing, well done, really impressive. 

I can't wait to see it in action!

[gideon]

A small update: I am still working on the collision detection, it's mostly a long and painfull process of debugging. There are quite a lot of problems with floating point rounding errors which are not easy to solve. While many examples work perfectly, like the gear system I showed before, however some simple examples exhibit very strange (jumpy) behaviour.

Movable line support is coming along nicely, although I will have to seriously alter the collision detector's structure to properly implement line-line collisions and allowing one way pass-though lines. For both problems, I need to know on what side (of a line) the other body was in the last frame. Just making the collision detector statefull isn't enough; it will have to be updated each frame as well. But I'm talking to myself.

Currently I'm planning to
- get the most apparant bugs fixed or work around them
- get movable line support working except for line-line collisions and pass-through lines
.. and release a patch. After that I'll continue with optimizations and getting the line-stuff working, but that will be another patch.

I still update my project's svn fairly regularly: http://sourceforge.net/svn/?group_id=117163, in case you can't wait any longer.

[darkprophet]

Sorry to state this if you already knew, but you could do all your calculations as doubles then cast the result...I know Intel FPUs have 80bit precision so a double takes roughly the same time as a float to calculate (there are a few other issues).

DP

Games published by our own members! Check 'em out!

[gideon]

Using doubles only makes the problem more rare: very small penetrations will simply always be there, I have to deal with them some way or another.

[Marvin Fröhlich]

Using doubles only makes the problem more rare: very small penetrations will simply always be there, I have to deal with them some way or another.

Did you check the performance impact of using doubles instead of floats? (Of course correctness should be of higher priority than performance.)

Marvin

[gideon]

(Of course correctness should be of higher priority than performance.)

Why should it? As long as everything looks 'natural' I'm ok with it. In any case, I haven't had any need for doubles yet; the problems that I encounter are usually more related to picking the right collision normal and penetration depth.

There might be some trouble up ahead though. I haven't properly tested the accuracy yet, for example placing object very far from the origin and using very small objects. Most certainly things will go wrong somewhere, if it turns out we need more 'space' (determined by the minimum objects size and maximum distance from the origin) switching to floats isn't really big a deal.

Maybe fixed point math could be useful. Does anyone know whether fixed point math actually is faster than floating point math? On modern processors like my dual core amd64 it's probably completely irrelevant, but what about mobile processors?

[darkprophet]

fixed point is faster on mobile phones than floating point. IIRC, a few games use fixed point to accelerate 3D rendering on the NGage platform.

DP

[t_larkworthy]

I checked out Phys2D yesterday with the idea of intergrating it with JOODE. It looks really nice, but at the moment there are not really enough new 2D primatives to really warrent the work. So just waiting for polgon support... hehe. I would personally like accuracy to be there, with a switch to turn it off, though our tri mesh support also has problems with very pointy thin geometies (ported from Collada I think).
Tom