basic winit port

[xorgy]

This removes (for now) accessibility, as well as the system menu.

[dfrg]

This is much cleaner than I would have expected. Overall LGTM and I’m comfortable merging now and making further changes incrementally.

[xorgy]

This is much cleaner than I would have expected. Overall LGTM and I’m comfortable merging now and making further changes incrementally.

It's not really clean, because there are a bunch of hanging references in App which rely on the (Glazier) WindowHandle, which can't really access it because of the ownership boundary between them. This PR breaks accessibility in its current form, and there's not really any clean way to restore it without changing the App/AppLauncher API.

[Philipp-M]

not really any clean way to restore it without changing the App/AppLauncher API

What do you mean exactly?

[xorgy]

@Philipp-M

What do you mean exactly?

I mean that the AppLauncher has to own the window, but the App accesses it directly to implement accessibility integration, and there is no clean way to let both of them access it. Right now there is no mechanism for opening subwindows anyway, which was the purpose of separating App and AppLauncher, so collapsing these interfaces together seems to be the quickest way to get it working.

[raphlinus]

I skimmed the PR and tested it. It works for me (tested on mac), and further work blocks on it. My preference is to merge soon, and document all the regressions and rough edges to those can be separate PRs. Thanks for the work!

[xorgy]

Yeah. I'll add the removals for the the features that are broken by this PR and pull it out of draft; we can go-no-go at the Office Hours tomorrow.

[Philipp-M]

I'm running into a weird issue at this line (so far I could pinpoint it at least):

xilem/src/app_main.rs

Line 231 in 623f18a

let surface_texture = surface

This seems to deadlock for me with this PR, after a few frames painted (slightly under 10 frames).
I'm not sure it's directly related to this PR though, as I have sometimes a crash when resizing on current main (at the same line with an failed to acquire next swapchain texture: Outdated error) but it doesn't crash or deadlock at this line otherwise.

I'm not sure where the exact issue could be. I'm testing on Hyprland (wayland + nvidia beta driver) currently. I can try on X11 if it works there for me.

I just tried the wgpu raytracing triangle example which works for me (maybe relevant).

[Philipp-M]

I just tested it on X11, there it works. I can still provoke the Outdated swapchain texture issue though when resizing...

[Philipp-M]

Ok, I'm pretty sure it's a wgpu issue, I could reproduce the issue with the examples on wgpu 0.18 (which is what xilem still uses). But not on wgpu 0.19

[Philipp-M]

Ok, no unfortunately it wasn't as easy as that...
I have quickly hacked in wgpu 0.19 and updated to current vello (I guess I can upstream that on your/this branch, when it's already done anyway?).

Same issue though, deadlock at that line after some frames...

[Philipp-M]

I could "fix" it by commenting out these two lines: https://github.com/gfx-rs/wgpu/blob/b8a8ff6e44162a41b8d54184a88613794dd3d2e4/wgpu-hal/src/vulkan/instance.rs#L967

And since these lines don't exist on the trunk branch on wgpu, it does work indeed with

[patch.crates-io]
wgpu = { git = 'https://github.com/gfx-rs/wgpu.git' }

in Cargo.toml (after updating/adjusting wgpu and vello in xilem).

When no one else has these issues (seems to happen on wayland + vulkan + maybe nvidia), I'd be ok to merge with this regression, as it otherwise looks good to me.

Maybe it also makes sense to add the lines above to Cargo.toml (but I guess this can be its own PR on top of this, including the update to wgpu and vello).

[Philipp-M]

FWIW, here's my changes to get that working for me with wayland: Philipp-M@9ab5c83

[xorgy]

FWIW, here's my changes to get that working for me with wayland: Philipp-M@9ab5c83

When this is merged, please put that up as a PR. :+ )