anvil/udev: Dynamic support for pixman rendererer

[ids1024]

With the changes here, the anvil udev backend will use pixman for rendering if ANVIL_USE_PIXMAN is set.

To do:

• Figure out how to deal with Send bound on MemoryRenderBuffer, and no Send for PixmanTexture
• Fix build with debug feature (need to abstract texture)
• Make render node optional, and automatically use pixman if hardware acceleration isn't present
• Pixman support in x11/winit backends

It's not entirely clear how best to abstract renderers dynamically:

• Should an enum wrapper implement the Renderer trait?
  • How does that impact performance or binary size compare to dispatching to a render function twice like this currently does?
  • Will also need enums over texture, frame, error; and probably a panic if these types are mismatched in some way
• Would any changes to Renderer and related traits make it easier to abstract over, and dispatch at runtime to different renderers?

[cmeissl]

I tried to create a wrapper enum implementing the renderer traits. Bind is the most "problematic" as this can be different per renderer.

Macro based implementation: https://github.com/cmeissl/smithay/blob/3834756520cfb8db5c21b9b5485d16bdc7cdab6e/src/backend/renderer/auto.rs
Usage of the enum: https://github.com/cmeissl/smithay/blob/3834756520cfb8db5c21b9b5485d16bdc7cdab6e/anvil/src/render.rs#L28

[ids1024]

I was thinking a macro might make sense. That looks handy. Although macro rules macros like that get a bit complicated.

[ids1024]

Updated on top of #1600.

Testing this I see a crash:

2024-12-04T18:08:44.814186Z  WARN anvil::udev: Error during rendering: ContextLost(BufferDestroyed)
thread 'main' panicked at anvil/src/udev.rs:1613:30:
Rendering loop lost: The underlying buffer has been destroyed

I guess that's an existing issue with the pixman renderer, and not something changed here? Hm.

[ids1024]

Updated so the debug feature builds, and all the feature combinations tested on CI now work. Still needs a fix for that crash.

Exactly how best to abstract over multiple renderers remains an open question.

[ids1024]

This seems to be working correctly, including with the debug feature.

We may want to change how renderers are abstracted dynamically in the future (and provide some helper for this in Smithay), but this seems good for now, to have a way to test Pixman rendering properly.