[GraphEdit] Correctly disconnect signal to connection_layer

[maidopi-usagi]

Fix #94811

[akien-mga]

Thanks! And congrats for your first merged Godot contribution 🎉

[Naros]

@akien-mga and @Geometror are we sure this code change is correct?

I've built this locally against master, and I'm am noticing that this leads to this error.

ERROR: Attempt to disconnect a nonexistent connection from '[Wrapped:0]'. Signal: 'item_rect_changed', callable: 'CanvasItem::queue_redraw'.
   at: (core\object\object.cpp:1529)

In our extended GraphEdit implementation, we have two notification handlers, one for NOTIFICATION_READY and another for NOTIFICATION_THEME_CHANGED. In both of these cases, we have a handler that fires that adds/removes nodes from GraphEdit.

The theme changed event is what seems to create the issue where the remove_child_notify indicates that the item_rect_changed signal isn't connected yet to the CanvasItem::queue_redraw handler.

The handler that both READY and THEME_CHANGED call basically removes nodes like this:

for (GraphNode* node : node_list) {
  remove_child(node);
  node->queue_free();
}

It's this that creates an issue during theme changed which wasn't an issue in Godot 4.3.stable.

What's even more bizarre is this only manifests if the UI has two GraphEdit instances open with two different graphs. If there is only one GraphEdit open, there's no issue.

[Naros]

If I wrap the call in the theme notification as:

case NOTIFICATION_THEME_CHANGED: 
  if (is_node_ready()) {
    // call my custom synchronize met
    _synchronize_graph_with_script();
  }
  break;

Then I don't see the issue, its only if the implementation adds/removes nodes before the node is ready that the calls to the add/remove child notify handlers lead to the signal not yet being connected.

[maidopi-usagi]

It's this that creates an issue during theme changed which wasn't an issue in Godot 4.3.stable.

@Naros Thanks for reporting! I'm wondering if the problem you're encountering might be related to this specific commit, as this PR has already been contained in 4.3.stable build. Would you mind opening a separate issue and provide additional details about your testing environment and reproduction steps?

[Naros]

Hmm @maidopi-usagi in that case, I wonder if the issue is some change in Godot 4.4 that breaks this commit. I nor none of our user community saw this on Godot 4.3.stable 🤔. Let me continue to check and once I know for sure, I'll open another issue.