Surface::configure can panic with "map callback was leaked"

[DJMcNab]

Description
When Surface::configure is called, it's possible for the panic at:

wgpu/wgpu-core/src/resource.rs

Lines 204 to 210 in c85cbea

	impl Drop for BufferMapCallback {
	fn drop(&mut self) {
	if self.inner.is_some() {
	panic!("Map callback was leaked");
	}
	}
	}

to trigger, specifically when:

wgpu/wgpu-core/src/device/global.rs

Lines 2343 to 2345 in c85cbea

	if let Err(e) = device.maintain(hub, wgt::Maintain::Wait, &mut token) {
	break e.into();
	}

returned a non-empty Ok

Repro steps
Running vello's examples (since linebender/vello#350, at the time of writing linebender/vello@34483eb).
Pressing v to toggle vsync (and force a reconfigure) may also be required, although sometimes for reasons unknown this isn't needed

Expected vs observed behavior
I expected for a panic not to occur

I observed that a panic occured

Platform
Wayland Gnome on Fedora 38 on an intel i5-1240p
Reproduction also confirmed on "on Sway and with an amdgpu"

[benjamin-sieffert]

Running into the same issue after upgrading wgpu-core from 0.17.0 to 0.17.1.
Since supposedly it is related to VSync, here is what happens for me, using amdgpu kernel driver on linux, no amdvlk installed, i.e. vulkan via mesa is used (as far as I understand all this).
Automatically choosing presentation mode by rule AutoNoVsync. Chose Immediate
My monitor can do FreeSync, though. Maybe first Immediate is used, but then immediately switched to Adaptive-Sync?

[dhardy]

#4220 does not fix this for me: I still get "Map callback was leaked" when resizing a window using 152c239.

Reproduce:

git clone https://github.com/kas-gui/kas.git --single-branch --branch wgpu
cd kas
cargo run --example layout
# resize the window

[DJMcNab]

It looks to me like the callbacks are only called if there's an error

Not checked the code on a big screen though

[nical]

Gosh, you are right, that fix wasn't very good.