"Lookup Symbol" or "Ctrl+LMB" not working for functions defined in script of another node

[nagidev]

---

Bugsquad note: This issue has been confirmed several times already. No need to confirm it further.

---

Godot version

v4.2.beta1.official [b137180]

System information

Windows 10 - Godot v4.1.1.stable.official

Issue description

Including a reference to a child node, and accessing a function on the child node works in game.
However, trying to use the "Lookup symbol" option or CTRL+LMB on the function from the script editor does not work at all.

This works fine in Godot 3.5.

Additionally, the functions defined in the child node are not suggested in auto-complete in the parent node's script.

Steps to reproduce

1. Create a scene with a parent node and a child node (type does not matter).
   ParentNode
   ChildNode
2. Attach scripts to both nodes.
3. Define a function (say, foo) in the child node's script.
4. Reference the child node in the parent node's script (say, @onready var child = $ChildNode) Note: the issue only happens when onready is used.
5. Call the function foo from the child anywhere in the parent node's script.
6. Firstly, the function "foo" is not suggested in the auto-complete. Second, "lookup symbol" or "CTRL+LMB" don't open the function definition.

Minimal reproduction project

lookupBug.zip

[eterlan]

As far as I know, if you don't declare class_name, you can't get any intellisense for your custom function.
The workaround is to declared class_name, and field with type.
I make some test and find some quite interesting behaviour.
case 1: child node has declared class_name, it has intellisense from parent if I declare the field as var child : Child
case 2: child node has declared class_name, it has no intellisense from parent if I declared the field as @onready var child: Child = $child

[nagidev]

Getting the same results. This code seems to be working as expected and Lookup symbol is working.

extends Node

var child_node : ChildNode

func _ready():
	child_node = $ChildNode

func parent_function():
	child_node.child_function()

Although, this means we need to write extra code and define a class_name for each script. It works just fine in 3.5.

[API-Beast]

It seems like 4.1 has trouble deducing the type of $ expressions, while in 3.x it did that just fine.

[nosdregamon]

Isn't this because $ChildNode is defined @onready, and unless we run this as tool-script, onready is simply never called?

[nagidev]

Turning the script into a tool script doesn't change the issue. In fact, I originally noticed the issue while working on tool scripts.

[willnationsdev]

I would guess that it's because $Name can return anything extending Node and it cannot ascertain definitively whether or not the returned node will be a valid reference for the hinted type. Is it also the same (with @onready) if you do $ChildNode as ChildNode or perhaps = ($ChildNode as ChildNode) (should both be equivalent, but just in case the parser is making a distinction)?

[dalexeev]

Generally speaking, autocompletion with dynamic typing is not always possible. But in this case, we could infer a weak type using the editor's information about the current scene, just as we autocomplete node paths. (The same script can be attached to multiple scenes, so autocompletion is not possible without an open scene.)

[nagidev]

Godot 3 managed to pull this off quite reliably. Is there a way to bring back that feature?

[API-Beast]

This is a general issue regarding $ syntax that also affects auto-complete. Here for example in 3.x it would show me the "aim_at" function, but in 4.x it will only show me generic Node functions.

I would really argue this is a regression from 3.x to 4.x rather than a "discussion" as it tagged right now. It is a feature that was core in 3.x and was lost during the rewrites.

[KoltPenny]

I'm also having this issue! We'll see how it goes.