My work with the entity system is pretty much finished for now. Everything seems to be finally working and the code is reasonably clean. I don’t want to get stuck polishing the code while it’s still missing most of the planned features. I didn’t get to gameplay / user interface stuff yet, but I hope to do so next week. Time to get something concrete done for a change.
I did run into some problems with collision detection though. Shells that are fired travel relatively fast compared to their size, which leads to the infamous bullet-through-wall effect when they collide with ground. In other words: they go straight through it. I tried fixing this both by making the terrain collision shape consist of polygons instead of line segments, and decreasing the step size so that shells wouldn’t jump over the line segments. Both of these were too much for pymunk though, and the game ground to a halt. Note this isn’t a problem with Python either, since all the physics number crunching is done in C code.
I now have a couple of options. First, I’ll try pyBox2d, based on the Box2D physics library (which chipmunk is based on, which pymunk is based on…) Box2D has continuous collision detection, meaning that objects can’t jump through each other between steps, but the pyBox2d bindings are based on SWIG, which is much slower than ctypes, so it might have performance problems of its own… Additionally, I just (literally a minute ago) stumbled on Elements, which is another Python 2D physics library based on Box2D, so I’ll take a look at that as well. A backup plan is faking continuous collision detection by making the collision shapes of shells longer than they actually are. Because the action in Artillery Brawl isn’t very dynamic, this shouldn’t cause any noticable errors in collision detection, but I’ll have to see about that if I’m not satisfied with pyBox2d.
EDIT: At least these results look promising for Box2d.
Tags: Artillery Brawl, Box2d, entity system, physics, pymunk