jach / lgame Goto Github PK

There are currently two problems with using cl-sdl-ttf.

1. Since cl-sdl-ttf depends on libffi, and since cffi-libffi currently won't load lib version >= 8 (>= 3.4.2-r1) until cffi/cffi@7f0a5ed is part of a quicklisp distro, trying to load lgame currently fails. Fortunately a restart is available to let you use the value ("libffi.so") and load that, but this isn't acceptable.
2. The CL lib has a bunch of issues and the owner seems unresponsive. One is the use of finalizers seems to not be correct/safe. Another is that calling the underlying functions is kind of impossible, see my bad wrapper functions to try and regain some control.

1 can be solved by waiting, I think, cffi is more active and the fix is already in, even if not yet distributed to quicklisp. 2 is more tricky. I think I may need to do the foreign bindings myself. Well, whatever, this issue is just for my notes. I'll close it if at least 1 is resolved and I no longer have an error when quickloading. On the off chance that I need to make this public before it's fixed, I'll find some other workaround, like dropping sdl-ttf entirely...

After seeing the SDL3 migration guide I'm pretty confident I don't want to touch it. I'd be surprised if pygame smoothly migrates to it anytime soon, though the pygame-ce project is worth keeping an eye on. I think it's time to give up on SDL.

Probably should follow Kandria and use OpenGL / GLFW as a base layer. Or perhaps finally dig into Trial and see if a pygame-esque interface can be easily built on top of it? (And how about using godot as a backend instead?)

Well, for the time being, I'll probably do nothing. SDL2 isn't going anywhere.

If I do do anything, I should prioritize using fewer direct SDL calls, symbols, and data objects and instead go through the trouble of making Lisp native versions that get translated later.

e.g. an actual rect class instead of sdl-rect, whose int behavior has given me trouble before, actual Lisp render functions that can be inlined to the sdl render ones but also potentially in the future behave as just staging a render to be executed on a separate correct thread (that won't crash), and actual Lisp keyword symbols for events instead of the constants brought in from the sdl header (that are renamed now in sdl3). I get a lot of productivity by being able to tab-complete, though.