Input event order for Viewports, Areas doesn't match documentation

[paulmiller]

Godot version

3.5.1.stable, 4.0.beta

System information

Arch Linux

Issue description

I think this is a bug, but it's also possible I'm misunderstanding the documentation.

The Using InputEvent tutorial describes this sequence:

1. Call Node._input() on every Node in reverse order.
2. Call Control._gui_input().
3. Call Node._unhandled_input() on every Node in reverse order.
4. If object picking is on, call Area._input_event() on any picked object.
5. Repeat for the next Viewport in the tree.

If I understand that correctly, then when clicking the CollisionShape in this scene tree:

• Spatial1
  • Spatial2
  • ViewportContainer
    • Viewport
      • Spatial3
  • Area
    • CollisionShape

The order of Spatial._input() and Area._input_event() calls should be:

1. Spatial2._input()
2. Spatial1._input()
3. Area._input_event()
4. Spatial3._input()

But It seems like Viewports receive input events in order with every other Node in the tree, rather than at the end. So the order is actually:

1. Spatial3._input()
2. Spatial2._input()
3. Spatial1._input()
4. Area._input_event()

I use Godot 3 node names here but the same happens in Godot 4.

The relevant Godot 4 code seems to be SceneTree::_call_input_pause(); it loops over the Nodes, and if a given node happens to be a SubViewportContainer, then Node::_call_input() > SubViewportContainer::input() > Viewport::push_input() > SceneTree::_call_input_pause(), which then loops over the Nodes under the SubViewportContainer.

Steps to reproduce

The attached Godot 3 project has the scene tree I described. Click the sphere and see the order logged in the console.

Minimal reproduction project

input-test.zip

[Sauermann]

Yes, the "Using InputEvent tutorial" is outdated. I intend to update it, however there are a few bugfix pull-requests waiting for review (e.g. #58334), that will have an impact on that tutorial, so I would like to update the tutorial after they get merged.

I created this infographic to showcase how event-processing works, which matches your findings:

[Sauermann]

I had a look at this issue and have updated the parts of the tutorial, that will be unaffected by the open PRs.