Surface::drop panics after device loss

[GeneralGDA]

Surface::drop panics "Trying to destroy a SwapchainAcquireSemaphore that is still in use by a SurfaceTexture" after device loss

---

After a GPU device loss, dropping a wgpu::Surface that had a frame in progress panics with:

thread 'main' panicked at wgpu-hal-29.0.0/src/vulkan/swapchain/native.rs:359:
Trying to destroy a SwapchainAcquireSemaphore that is still in use by a SurfaceTexture

Steps to Reproduce

1. Call Surface::get_current_texture() successfully — this sets Presentation::acquired_texture = Some(...) inside wgpu-core and clones the Arc<SwapchainAcquireSemaphore> into the HAL SurfaceTexture (refcount goes 1 → 2).
2. Submit GPU work with Queue::submit() — device loss is detected here; wgpu-core calls Device::lose(), marking the device invalid, and fires the set_device_lost_callback callback.
3. Call SurfaceTexture::present() — this calls wgpu_core::Surface::present(), which calls device.check_is_valid(), gets Err(DeviceError::Lost), and returns early without taking acquired_texture. The field remains Some(...).
4. Drop the wgpu::Surface (e.g. while recreating graphics resources after device loss) — Drop for wgpu_core::Surface runs:

// wgpu-core/src/instance.rs
impl Drop for Surface {
    fn drop(&mut self) {
        if let Some(present) = self.presentation.lock().take() {
            for (&backend, surface) in &self.surface_per_backend {
                if backend == present.device.backend() {
                    unsafe { surface.unconfigure(present.device.raw()) }; // ← called here
                }
            }
        } // ← `present` (with non-None acquired_texture) dropped HERE
    }
}

surface.unconfigure() is called while present is still in scope, so acquired_texture is still alive. It holds an Arc<SwapchainAcquireSemaphore> clone (refcount = 2). Inside unconfigure():

// wgpu-hal/src/vulkan/swapchain/native.rs
unsafe fn release_resources(&mut self, device: &super::Device) {
    // ...
    for semaphore in self.acquire_semaphores.drain(..) {
        let arc_removed = Arc::into_inner(semaphore).expect(
            "Trying to destroy a SwapchainAcquireSemaphore that is still in use by a SurfaceTexture",
        ); // ← PANIC: Arc refcount is 2, not 1
    }
}

Arc::into_inner() returns None because refcount = 2, causing the panic.

Root Cause

Drop for wgpu_core::Surface violates unconfigure()'s own documented contract ("no resources derived from the swapchain in use") because it does not drain acquired_texture before calling unconfigure().

The underlying cause is that wgpu_core::Surface::present() (and discard()) guard themselves with device.check_is_valid()? and return early on device loss, leaving acquired_texture populated. The Drop impl doesn't account for this case.

Expected Behavior

Dropping a wgpu::Surface should not panic, regardless of device state.

Environment

• wgpu version: 29.0.0
• wgpu-hal version: 29.0.0
• wgpu-core version: 29.0.0
• Backend: Vulkan
• OS: Windows 11
• GPU: GeForce 5090 mobile