BADHACK: gles/egl: Require surface before creating instance / query adapter on Wayland

[MarijnS95]

Connections
Fixes #5505
Could address #3176 if we make Inner an Option
Fixes #7965

Description

In egl, Inner has been made Clone in #2350, and prior to that some EGL fields were copied into AdapterContext since #1729.

At the same time wgpu to this day doesn't seem to accept a RawDisplayHandle when creating an Instance, meaning that on Wayland at least we end up creating a bogus EGLDisplay (and respective GL context) that isn't compatible with the window/surface that the user later wants wgpu to present to (not to mention, the user/caller has zero control over the Wayland or other connection to use, this is all hardcoded based on library detection and connecting to the default Wayland socket...).

Each time wl_display_connect() is called, which is likely also what happens internally when EGL_DEFAULT_DISPLAY is passed for EGL_PLATFORM_WAYLAND_KHR, a different and incompatible display connection is used. Even if wgpu's test_wayland_display() would have connected to the same Wayland server as eglGetPlatformDisplay(WAYLAND, DEFAULT_DISPLAY) and the one the caller uses for its surface, these cannot be used interchangeably.

To patch up this bogus display, gfx-rs/gfx#3545 inserted some logic to replace the internal EGLDisplay and GL context whenever a surface is created which does receive a RawDisplayHandle with the correct wl_display handle to use to be compatible with the given wl_surface:

wgpu/wgpu-hal/src/gles/egl.rs

Lines 991 to 1034 in 96ef5ec

	(Rwh::Wayland(_), raw_window_handle::RawDisplayHandle::Wayland(display_handle)) => {
	if inner
	.wl_display
	.map(|ptr| ptr != display_handle.display.as_ptr())
	.unwrap_or(true)
	{
	/* Wayland displays are not sharable between surfaces so if the
	* surface we receive from this handle is from a different
	* display, we must re-initialize the context.
	*
	* See gfx-rs/gfx#3545
	*/
	log::warn!("Re-initializing Gles context due to Wayland window");
	use core::ops::DerefMut;
	let display_attributes = [khronos_egl::ATTRIB_NONE];
	let display = unsafe {
	inner
	.egl
	.instance
	.upcast::<khronos_egl::EGL1_5>()
	.unwrap()
	.get_platform_display(
	EGL_PLATFORM_WAYLAND_KHR,
	display_handle.display.as_ptr(),
	&display_attributes,
	)
	}
	.unwrap();
	let new_inner = Inner::create(
	self.flags,
	Arc::clone(&inner.egl.instance),
	display,
	inner.force_gles_minor_version,
	)?;
	let old_inner = core::mem::replace(inner.deref_mut(), new_inner);
	inner.wl_display = Some(display_handle.display.as_ptr());
	drop(old_inner);
	}
	}

Now all the way back to the first sentence, about cloning a bunch of EGL handles. The problem mainly occurs when the user either enumerated Adapters before this operation, and worse if they started creating resources on it. Because of this clone:

wgpu/wgpu-hal/src/gles/egl.rs

Lines 1100 to 1103 in 96ef5ec

	AdapterContext {
	glow: Mutex::new(gl),
	egl: Some(inner.egl.clone()),
	},

Those adapters have the old EGLDisplay and old GL context. Which we just closed when replacing Inner with the "new wl_display-based" EGLDisplay and context:

wgpu/wgpu-hal/src/gles/egl.rs

Lines 707 to 721 in 96ef5ec

	impl Drop for Inner {
	fn drop(&mut self) {
	if let Err(e) = self
	.egl
	.instance
	.destroy_context(self.egl.display, self.egl.raw)
	{
	log::warn!("Error in destroy_context: {e:?}");
	}
	if let Err(e) = terminate_display(&self.egl.instance, self.egl.display) {
	log::warn!("Error in terminate: {e:?}");
	}
	}
	}

Whenever the user now uses the adapter to make the context current, a BadDisplay / BadContext error is generated because the context and display handles are both destroyed.

To solve that, one of the proposed solutions was to follow what Web does: require a surface when enumerating Adapters so that the right EGLDisplay is copied over. This is however overly restrictive:

• After creating a surface, the internals are updated and a surface no longer needs to be passed when enumerating surfaces.
• A surface (typically from a window) shouldn't be needed at all, if merely the RawDisplayHandle could be passed.

For now, see if this unblocks folks while we work on a more likable solution. Stay tuned.

Testing
amdgpu with Wayland compositor, with e.g.;

WGPU_BACKEND=gl cargo r -p wgpu-examples -- shadow

Squash or Rebase?

rebase

Checklist

• Run cargo fmt.
• Run taplo format.
• Run cargo clippy --tests. If applicable, add:
  • --target wasm32-unknown-unknown
• Run cargo xtask test to run tests.
• If this contains user-facing changes, add a CHANGELOG.md entry.