Blade window transparency

[kvark]

Release Notes:

• N/A

Following up to #10880
TODO:

• create window as transparent
  • X11
  • Wayland
  • Windows
  • MacOS (when used with Blade)
• enable GPU surface transparency
• adjust the pipeline blend modes
• adjust shader outputs

Blurred results from @jansol (on Wayland), who contributed to this work:

[ThatOneCalculator]

Progress? 👀 (Hyprland with wlroots)

[kvark]

Pushed the X11 changes here, but it currently doesn't start. Vulkan surface creation fails with VK_ERROR_OUT_OF_HOST_MEMORY. Looking at the Mesa source, it appears to throw this error pretty much on anything that goes wrong, without necessarily logging the reason anywhere.

[kvark]

It doesn't like the Visual I'm choosing, for some reason. Here is the visual selected by default, where it works:

Using Visual id: 66, class: TRUE_COLOR, depth: 24, bits_per_component: 8, colormap: 65, masks: 0xff0000, 0xff00, 0xff

And here is the opaque Visual I got by scanning the compatible ones:

Using Visual id: 975, class: DIRECT_COLOR, depth: 24, bits_per_component: 8, colormap: 0, masks: 0xff0000, 0xff00, 0xff

Also, here is the transparent Visual I got, which also doesn't work:

Using Visual id: 751, class: TRUE_COLOR, depth: 32, bits_per_component: 8, colormap: 0, masks: 0xff0000, 0xff00, 0xff

I can see that it's not about TRUE_COLOR versus DIRECT_COLOR, and not about depth, and not about color masks (which are the same in all cases). The only strange thing is - default format having a color map.
I suspect that not all of the visuals are compatible with Vulkan surface, but I don't know how to detect it.

[kvark]

I got stuck for too long with this issue on X11, which turned out to be quite an adventure!

I looked closely at all of the existing X11 examples and also at Blade examples, which worked. Discovered that winit is still producing RawWindowHandle::Xlib even though using Xcb - rust-windowing/winit#3681
I tried following the cairo example by querying formats and seeing what visual IDs are associated.
I tried different Visual IDs by hand and confirmed that I'm failing on the same ones that work via different paths...
I tried compiling RadV (AMD's driver) locally and running with it, stepping through in a debugger, to discover that it failed on simple queries about the given window (like, who is your father?).
This led me to double check all of the things we do at window creation and conclude that the window was created as invalid in the first place, and we weren't properly checking for it. Of course creating a Vulkan surface would fail in this case!

Finally, I found out that border_pixel setting was missing comparing to winit's path. Also related to this StackOverflow discussion. This was the last missing piece, and now everything appears to work in order, finally!
Small achievement, but at great cost :)

[mikayla-maki]

Looks great, thank you so much @kvark

Also, yikes, that X11 bug looks gnarly. Thanks for tracking that one down :D

[mikayla-maki]

Merging without windows support as we're not officially working on Windows yet. If folks want to follow up, feel free!

[ThatOneCalculator]

Nice! One minor issue I'm noticing is that the selected item in the file sidebar shows as solid white

  "experimental.theme_overrides": {
    "background.appearance": "blurred",
    "background": "#191724BB",
    "editor.background": "#00000030",
    "editor.gutter.background": "#00000030",
    "tab_bar.background": "#00000030",
    "terminal.background": "#00000030",
    "toolbar.background": "#00000030",
    "tab.active_background": "#26233a90",
    "tab.inactive_background": "#00000000",
    "status_bar.background": "#00000030",
    "title_bar.background": "#00000030",
    "panel.background": "#00000030",
    "border.variant": "#00000000",
    "scrollbar.track.border": "#00000000",
    "editor.active_line.background": "#00000000"
  },