Perform update_mouse_cursor_state at most once per frame.

[Sauermann]

Set a flag to perform the update after input event processing during NOTIFICATION_PROCESS.

fix no longer #77773 (was solved in a different way)

[KoBeWi]

This enables processing of all Window nodes, which is a bit wasteful.
I think the queue_redraw() approach would be more suitable (just check how it's implemented; it's a flag + deferred call).

[Sauermann]

This enables processing of all Window nodes, which is a bit wasteful.

I was afraid of that and am open to other suggestions.

I think the queue_redraw() approach would be more suitable

queue_redraw() is defined in CanvasItem, so I am not sure how to use it from Window, which doesn't inherit from CanvasItem.

[KoBeWi]

I mean to use the same approach as that method, not the method itself.
Something like

void update_mouse_cursor_state()
    if (update_queued) return

    update_queued = true
    call_deferred("_update_mouse_cursor_state")

func _update_mouse_cursor_state()
    update here
    update_queued = false

[Sauermann]

Indeed I prefer this approach much more. Thanks for pointing me in the direction of call_deferred.

[Sauermann]

While I prefer the second solution, I was unable to implement it in a way, that could be merged in a timely manner.

The problem was the interaction of Viewport::update_mouse_cursor_state with Viewport::local_input_handled.
In my opinion the best solution for this would be, that Viewport::update_mouse_cursor_state should not utilize InputEvents.
An initial, not yet ready implementation of this solution is available at the temporary branch https://github.com/Sauermann/godot/tree/wip-fix-77773-better-solution-for-slider-stuck
That solution would make this PR obsolete. However that solution is based on PRs which are unlikely to make it into 4.1.

So I have reverted this PR to its previous state which enables processing on Window nodes, in order to provide a working solution to fix the linked bug.

[KoBeWi]

For lesser performance impact, you could enable process when the update is pending and disable it after update.
Not sure how costly is toggling processing though (for something that potentially can happen relatively often).

[Sauermann]

Right, I am not going to use processing, but the flag + deferred call method.

However this will have to wait until I have rewritten update_mouse_cursor_state so that it doesn't use InputEvents.
#67791 is the first step of this rewrite.

But since the referenced bug-report has been solved in a different way, it is fine for me, if this PR take more time.