Use safe window handles and wgpu::Instance::create_surface

[ids1024]

Objective

I noticed that #12978 introduces a reference counted wrapper around windows, but still uses RawWindowHandle and RawDisplayHandle, with wgpu::Instance::create_surface_unsafe. This can be changed easily enough to use wgpu::Instance::create_surface, and not use the raw handles anywhere.

Solution

Instead of containing an Arc<dyn _> along with raw handles, RawHandleWrapper now contains just an Arc<dyn _>, and ThreadLockedRawWindowHandleWrapper implements HasWindowHandle/HasDisplayHandle by calling those those methods on the trait object.

bevy_render uses wgpu::Instance::create_surface with the ThreadLockedRawWindowHandleWrapper. So this increments the reference count, but guarantees the window will outlive the wgpu::Instance.

Apparently we can just cast a Arc<W> to an Arc<dyn Trait>, so that makes WindowWrapper a little simpler too.

Testing

I'm not that familiar with Bevy, but examples seem to work, and this shouldn't break anything.

Migration Guide

RawHandleWrapper no longer has window_handle and display_handle fields. Instead, the get_handle() method has to be used.

[github-actions]

Welcome, new contributor!

Please make sure you've read our contributing guide and we look forward to reviewing your pull request shortly ✨

[BD103]

Thanks for stopping by, we appreciate it! :)

This change looks promising, though I'm less familiar with the windowing code. Do you think it will hurt performance?

[BD103]

Can this be automatically derived by #[derive(Debug)], or does it complain that dyn WindowTrait does not implement Debug?

[ids1024]

Yeah, it's needed since dyn WindowTrait doesn't implement Debug. (The standard library has an implementation for dyn Any.) It seems simplest to just implement Debug for RawHandleWrapper.

[BD103]

Ok, sounds good!

[ids1024]

Do you think it will hurt performance?

This should have no non-trivial impact on performance. It will just add a reference count to the Arc and store a copy in the wgpu::Instance to ensure the window isn't destroyed before the Instance.

Otherwise things should be basically unchanged.

[BD103]

Thank you!

[Friz64]

Taking it a step further, I like it!

Sadly though, this conflicts with the workaround to #13208 I suggested (implemented in #13221).

[alice-i-cecile]

Does this resolve #13208? If so, we should merge this instead of #13221.

[mockersf]

no it doesn't

[ids1024]

Hopefully a better solution can be found for that behavior on macOS. But if merging that workaround is the best for now, I guess the changes here can be merged later when that is fixed.

Come to think of it, I think WindowWrapper is unneeded now, with the Arc upcasting. RawHandleWrapper::new could just take an Arc<W>, and the Arc<W> could be used anywhere else WindowWrapper is used.

Another thing that seems a bit odd is that WindowWrapper::new requires Send + Sync, while the ThreadLockedRawWindowHandleWrapper stuff seems to be meant to deal with windows that aren't thread safe?

Wgpu in comparison has pub trait WindowHandle: HasWindowHandle + HasDisplayHandle + WasmNotSendSync {}, where WasmNotSendSync is Send + Sync except on Wasm.

[Friz64]

Come to think of it, I think WindowWrapper is unneeded now, with the Arc upcasting. RawHandleWrapper::new could just take an Arc<W>, and the Arc<W> could be used anywhere else WindowWrapper is used.

Another thing that seems a bit odd is that WindowWrapper::new requires Send + Sync, while the ThreadLockedRawWindowHandleWrapper stuff seems to be meant to deal with windows that aren't thread safe?

Yeah, these are just remnants from my PR at this point. I think this PR should address this.

[Friz64]

Sadly though, this conflicts with the workaround to #13208 I suggested (implemented in #13221).

Now, because we didn't go through with my quick & dirty workaround (and instead went with #13236), I think this PR is unblocked.

[mockersf]

this crashes on android:

05-12 19:24:03.414 E/log event( 1386): Cannot get the native window, it's null and will always be null before Event::Resumed and after Event::Suspended. Make sure you only call this function between those events.log.target = "winit::platform_impl::platform"; log.module_path = "winit::platform_impl::platform"; log.file = "/home/runner/.cargo/registry/src/index.crates.io-6f17d22bba15001f/winit-0.29.15/src/platform_impl/android/mod.rs"; log.line = 987;
05-12 19:24:03.414 I/RustStdoutStderr( 1386): thread 'Compute Task Pool (0)' panicked at crates/bevy_render/src/view/window/mod.rs:465:22:
05-12 19:24:03.414 I/RustStdoutStderr( 1386): Failed to create wgpu surface: CreateSurfaceError { inner: RawHandle(Unavailable) }
05-12 19:24:03.414 I/RustStdoutStderr( 1386): note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
05-12 19:24:03.415 I/RustStdoutStderr( 1386): Encountered a panic in system `bevy_render::view::window::create_surfaces`!

[ids1024]

Hm, so when RawHandleWrapper::new is run, AsWindowHandle works (it's between resumed and suspended?), but when create_surfaces is run, its suspended? I guess?

Not sure why that would be the case, but I'm not that familiar with how this works on Android.