Port video.Window to C

[yunline]

Split from #2070

• Port video.Window to C
• _sdl2.video.Window (old) -> _sdl2.video._Window and _sdl2.window.Window

[Starbuck5]

One thing I'd like to address besides the code is the namespace considerations.

Right now

_sdl2.video.Window -> _sdl2.video._Window
*** -> window.Window

I would like this new window to be in _sdl2 for now as well, until we decide it's ready for a public release (do a lot more integration work, add surface rendering functionality)

It also breaks every _sdl2 program to get rid of the Window class being where it is.

So instead, I suggest

_sdl2.video.Window -> _sdl2.video._Window
*** -> _sdl2.window.Window
    |  _sdl2.video.Window

I suggest aliasing a _sdl2.window.Window to _sdl2.video.Window. This way we can use existing _sdl2 programs and users to test the new Window implementation. Since the old window is still around (for the moment), we can also switch to that at runtime to test differences.

---

Also if you're looking for test code, https://mcpalmer1980.itch.io/flyboy is a complete game in _sdl2.

[Starbuck5]

Tests are failing because of the relative mouse mode not being in the type stub?

[Starbuck5]

This is a very important PR to get the Window API up and running, thank you so much for tackling it!

I think it looks good now.

Roadmap in my head:

• Two window APIs coexist for a release (for back testing purposes)
• Window tests!
• Old _sdl2.window cython code is removed
• New window features: (display interop, software rendering, etc)
• Release window API publicly

[MyreMylar]

Things I've noticed:

1. No snake icon in the title bar top left corner of the new C window versus the other ways to create a window.
2. When expanding a resizable new window to the bottom right some white lines are left behind.
3. If you .destroy() a new window with .grab = True your mouse is still grabbed to the now destroyed window's area.
4. Going directly from minimised to maximised with the functions does not expand the black area of the window in the same way it does if you go, for example: minimise() -> restore() ->maximise()
5. calling .show() on a hidden window turns the background white.

All this is tested on a Windows 11 PC, results may vary on other platforms as we are dealing with window management here. Some of these issues may be pre-exisiting in the old cython window.

[Starbuck5]

@MyreMylar Thanks for testing this out.

Number 1 is consistent with the cython implementation. Number 2, number 4, and number 5 are also consistent with cython, and seem like the product of not clearing the screen with a Renderer. Even display surfaces have some weird artifacts when you don't fill them every frame (when resized).

I want this PR to be consistent with the cython implementation. It's "port window," not "improve window."

I can't replicate number 4.

[MyreMylar]

@MyreMylar Thanks for testing this out.

Number 1 is consistent with the cython implementation. Number 2, number 4, and number 5 are also consistent with cython, and seem like the product of not clearing the screen with a Renderer. Even display surfaces have some weird artifacts when you don't fill them every frame (when resized).

I want this PR to be consistent with the cython implementation. It's "port window," not "improve window."

I can't replicate number 4.

For me the snake icon does appear on the old cython window but never on the new C window.

This is my test program:

import pygame
from pygame._sdl2.window import Window as OldWindow
from pygame._sdl2.video import Window as NewWindow

pygame.init()

old_window = OldWindow(title="Old Window", position=(20, 100))
new_window = NewWindow(title="New Window", position=(1000, 100))


running = True
while running:
    for event in pygame.event.get():
        if event.type == pygame.QUIT or event.type == pygame.KEYDOWN and event.key == pygame.K_ESCAPE:
            running = False

I also figured out that it is the first window created that gets the black background, any subsequent windows get a white background so that explains those differences somewhat. This is consistent between cython window and C window. Not really an issue for this PR.

Anyway it basically works being the important thing 👍

[yunline]

I copied some code from display.c to set the default icon.
I think copying is not a good choice, so I opened #2222, made a new API to get resource from pkgdata

[Starbuck5]

I’m unconvinced that the old cython code did this.

Also if I understand the imports correctly, Myre’s test code is testing the new window twice. Haven’t fully verified yet, just how it looks.
Edit: I get it now, I think. It's strange to have this new interface for the old window. I thought it would just be available as _sdl2.video._Window. I haven't tested either way, all of my testing has involved switching back and forth versions.

[yunline]

I’m unconvinced that the old cython code did this.

pygame-ce/src_c/cython/pygame/_sdl2/video.pyx

Lines 263 to 269 in f5b674f

	import pygame.pkgdata
	surf = pygame.image.load(pygame.pkgdata.getResource(
	'pygame_icon.bmp'))
	surf.set_colorkey(0)
	self.set_icon(surf)

It really does.

[Starbuck5]

Ok, it's because my test code had the window go fullscreen first, which leads SDL to lose track of the icon and the resulting non fullscreen window not having any icon.

[oddbookworm]

I forgot to actually leave a review when I looked at it a couple days ago, LGTM!