iced_layershell: Mouse Interaction is reset when new layer shell window is opened

[Ind-E]

When a new layer shell window is opened, it seems to reset the current mouse interaction. This can be seen in-action with the multi_window example - after you press a button to open a new window, the cursor shape changes from pointer to default.

I think it's because of this line:

exwlshelleventloop/iced_layershell/src/multi_window/window_manager.rs

Line 29 in 9af9811

pub mouse_interaction: mouse::Interaction,

[Decodetalkers]

the real problem is here https://github.com/waycrate/exwlshelleventloop/blob/master/iced_layershell/src/multi_window.rs#L998-L1007

because when a new window is open, it will always update the screen. Here should check if it is the focused screen. Can you help me fix it? or I will fix it sooner

[Decodetalkers]

Ok, I have make a pr, can you try it to see if it is fixed?

[Ind-E]

I tried it with the pr, and nothing seems to have changed

I'm going to sleep now, you should be able to test it yourself with the multi_window example

[Decodetalkers]

ok, I found out the problem

first is here

https://github.com/waycrate/exwlshelleventloop/blob/master/layershellev/src/lib.rs#L953-L958

We set the current_window when a new window appear,

and then when https://github.com/waycrate/exwlshelleventloop/blob/master/layershellev/src/lib.rs#L1644 keyboard enter, we set it again. When a new window is created, the keyboard will sure enter it.

so maybe we should add another function to get the focused id, which just focus on the enter event that does not include the keyboard event. or do not accept the surface in keyboard enter event, as the current surface. And sure, should not set the current_surface when a new window is created.. but maybe keep the logic will be a better choice? I do not know

[Decodetalkers]

Ok. I will fix it tonight

[Decodetalkers]

After reconsidering about it, I think we would better not fix it. First, layershell has an option to get the whole focus, so that time, the focus needs to be changed. And who has the focus of the keyboard should be shown as focused status. This is what I think

[httpsworldview]

This is also the cause of broken keyboard focus tracking in XDG toplevel child windows.

// layershellev/src/lib.rs ~960
fn push_window(&mut self, window_state_unit: WindowStateUnit<T>) {
    let surface = window_state_unit.wl_surface.clone();
    self.units.push(window_state_unit);
    self.update_current_surface(Some(surface)); // <--
}

update_current_surface here causes two problems:

• It queues a Focused event before iced_layershell's window_manager has registered the new window, so handle_window_event can't look it up and the event is dropped.
• When the compositor later sends the real wl_keyboard::Enter for that surface, update_current_surface sees surface == self.current_surface and early-returns, swallowing the legitimate focus event.

The result is that the first window created after startup never appears focused, so both mouse interaction and keyboard event routing are broken for it.

I think the fix is simply removing the update_current_surface call:

fn push_window(&mut self, window_state_unit: WindowStateUnit<T>) {
    self.units.push(window_state_unit);
}

This should be safe even for KeyboardInteractivity::Exclusive layer shells because the compositor is required to send wl_keyboard::Enter for exclusive surfaces, so the focus will be set correctly through the normal path. No keyboard events can arrive between push_window and the next roundtrip(), so there's no gap where focus state would be stale.

I'm also pretty sure this would resolve #317, but I haven't tested it.

[Decodetalkers]

This is also the cause of broken keyboard focus tracking in XDG toplevel child windows.

// layershellev/src/lib.rs ~960
fn push_window(&mut self, window_state_unit: WindowStateUnit) {
let surface = window_state_unit.wl_surface.clone();
self.units.push(window_state_unit);
self.update_current_surface(Some(surface)); // <--
}

update_current_surface here causes two problems:

* It queues a `Focused` event before iced_layershell's `window_manager` has registered the new window, so `handle_window_event` can't look it up and the event is dropped.

* When the compositor later sends the real `wl_keyboard::Enter` for that surface, `update_current_surface` sees `surface == self.current_surface` and early-returns, swallowing the legitimate focus event.

The result is that the first window created after startup never appears focused, so both mouse interaction and keyboard event routing are broken for it.

I think the fix is simply removing the update_current_surface call:

fn push_window(&mut self, window_state_unit: WindowStateUnit) {
self.units.push(window_state_unit);
}

This should be safe even for KeyboardInteractivity::Exclusive layer shells because the compositor is required to send wl_keyboard::Enter for exclusive surfaces, so the focus will be set correctly through the normal path. No keyboard events can arrive between push_window and the next roundtrip(), so there's no gap where focus state would be stale.

I'm also pretty sure this would resolve #317, but I haven't tested it.

Thank you for comment this. I also tried to fix it in this way, but when new windows open, the enter event of keyboard always come, so I did not fix it. You can also take a try.