wlr-layer-shell: skip pre-commit validation for destroyed surfaces

[BVMiko]

Description

I hit an issue while using gtk4-layer-shell: the client gets disconnected the moment it hides a layer-shell window.

To hide a window, gtk4-layer-shell destroys the zwlr_layer_surface_v1 and then commits the wl_surface (with a nil buffer).
On destroy the cached layer state is reset to defaults (size 0, no anchors), but the pre-commit hook is still attached to the surface, so that commit runs it. The state now matches the pending.size.w == 0 && !pending.anchor.anchored_horizontally() check, which posts InvalidSize ("width 0 requested without setting left and right anchors") and takes down the whole client.

This PR just skips the pre-commit validation once the zwlr_layer_surface_v1 has been destroyed. It only short-circuits the hook; it doesn't affect how the surface is torn down.

Checklist

• I agree to smithay's Developer Certificate of Origin.

[Drakulix]

This PR just skips the pre-commit validation once the zwlr_layer_surface_v1 has been destroyed. It only short-circuits the hook; it doesn't affect how the surface is torn down.

Hmm, I am not sure this is fully correct. We obviously don't want to validate any state that would be specific to the layer-surface (like geometry or anchors), but attaching a null buffer actually has an effect further down the commit hook, which is resetting the unmapped surface for another initial configure.

Arguably gtk4-layer-shell should really send the unmap (null-buffer commit) before destroying the surface, but we can work-around it.