Fix preload cyclic references

[adamscott]

Edit: This PR has been replaced with #67714
Edit: Currently not working as intended. Will post updates.

This simple PR makes it possible to use cyclic references with preloaded scripts.

After trying to load completely a script using ResourceLoader::load() (that calls down the line ResourceFormatLoaderGDScript::load() then right after GDScriptCache::get_full_script()), if the reference is null (which means that the script is not yet compiled), then we use instead the GDScriptCache::get_shallow_script().

The beauty of GDScriptCache::get_shallow_script() is that it gives back the result of GDScriptCache::get_full_script() if it has been fully loaded once.

So, this code works with this PR, which returns this error on master: Constant value uses script from "res://b.gd" which is loaded but not compiled.:

# main.gd
const A = preload("res://a.gd")

func _ready() -> void:
	A.hello()  # The console will print: "A.HELLO_WORLD: hello, world!"

# a.gd
extends Node

const B = preload("res://b.gd")

const HELLO_WORLD: = "hello, world!"

static func test() -> void:
	B.test()

# b.gd
extends Node

const A = preload("res://a.gd")

static func test() -> void:
	prints("A.HELLO_WORLD:", A.HELLO_WORLD)

Minimal reproduction project:
cyclic3.zip

Fixes #58181, fixes #61043, fixes #32165

[adamscott]

I know that I need to add a check to see if the preloaded path is a script or not before using GDScriptCache.

[adamscott]

This is where it gets weird. There's two behaviors, depending on the code on this line.

1. With GDScriptCache::get_shallow_script() as is:
   • my minimal reproduction project ✔️ ;
   • the new test cases ❌ ;
   • modules/gdscript/tests/scripts/analyzer/features/preload_constant_types_are_inferred.gd ❌ ;
   • build does not fail due to memory leaks ✔️ .
2. With ResourceLoader::load() and then GDScriptCache::get_shallow_script() if p_preload->resource.is_null():
   • my minimal reproduction project ❌ ;
   • the new test cases ✔️ ;
   • modules/gdscript/tests/scripts/analyzer/features/preload_constant_types_are_inferred.gd ✔️ ;
   • build fails due to memory leaks ⚠️

[adamscott]

@Calinou Could you reopen my PR? I messed up my rebases and erased my code, even force pushed it. I recovered it, but I cannot reopen my PR myself.

[akien-mga]

It can't be reopened because at some time the contents of your branch were identical to the master branch, so GitHub force closed it, and it can't be reopened in such case.

That's actually a bug because I see you did fix the branch already and there's now a commit, but GitHub doesn't notice. So you'll have to open a new PR.

[MewPurPur]

#67714 This is the current PR with this.