OpenGL Notes

Arcade is using OpenGL for the underlying rendering. OpenGL functionality is given to use through pyglet when a window is crated. The underlying representation of this is an OpenGL context. Arcade’s representation of this context is the arcade.Window.ctx. This is an ArcadeContext.

Working with OpenGL do add some challenges we need to be aware of.

Initialization

Certain operations can’t be done before a window is created. In Arcade we do deferred initialization in many of our types to make this as painless as possible for the user. SpriteList can for example be built before window creation and will be initialized internally in the first draw call.

TextureAtlas on the other hand cannot be crated before the window is created, but Texture can freely be loaded at any time since these only manage pixel data with Pillow and calculate hit box data on the cpu.

Threads

OpenGL is not thread safe meaning doing actions from anything but the main thread is not possible. You can still use threads with arcade, but they cannot interact with anything that affects OpenGL objects. This will throw an error immediately.

Another problem with threads is that the garbage collector might decide to run in a non-main thread for various reasons. A way to combat this is to change the garbage collection mode for OpenGL resources

# Enable "context_gc" mode (default is "auto") during
# window creation
Window(gc_mode="context_gc")

# From now on you need to manually call ctx.gc()
# for OpenGL resources to be deleted. This can be
# done very frame if needed or in shorter intervals
window.ctx.gc()

Note that if vsync is enabled all threads will stall when all rendering is done and OpenGL is waiting for the next vertical blank. The only way to combat this is to disable vsync or use subprocesses.

SpriteLists can be created in threads if they are created with the lazy=True parameters. This ensures OpenGL resources are not created until the first draw() call or initialize() is called. This can be especially useful when using spatial hashing because the initial calculations can be costly.