Provide safe way for implementing IScriptExtension::instance_has

[TitanNano]

After looking into #970 I noticed that there is currently no good way to implement IScriptExtension::instance_has. Godots own implementations of this function rely on some way to access the existing script instances inside the engine.

With the new gdextension function that has been introduced in 4.3 extensions now also have access to the existing script instances in the engine. While it's not safe to access these instance pointers, we can provide safe abstractions for certain use cases. I think this is such a use-case and so far, I haven't identified any other use-case.

---

There are a couple of necessary changes to enable the integration testing for this new function.

[GodotRust]

API docs are being generated and will be shortly available at: https://godot-rust.github.io/docs/gdext/pr-1013

[Bromeon]

Thanks a lot!

Maybe general question, is the typical use case to check whether an (obj, script, lang) combination to exist -- but not e.g. to query for a part of the tuple the other valid values?

I've glanced over it, might have a deeper look later 🙂

[TitanNano]

Maybe general question, is the typical use case to check whether an (obj, script, lang) combination to exist -- but not e.g. to query for a part of the tuple the other valid values?

Generally the engine creates a script instance inside the objects script setter function, so I don't really see a realistic use case for this yet. So far, the only technical use case I could identify was to easily implement IScriptExtension::instance_has.
For that use case, we only need (obj, script) but object_get_script_instance was implemented to take an object and a ScriptLanguage while the Script class at the same time does not expose the get_language() method.

Regardless of this, we sill should be able to use script_instance_exists in cases where we only have (obj, lang) by using obj.get_script() as the script.

[Bromeon]

Since 4.4, you can use Variant::object_id():

gdext/godot-core/src/builtin/variant/mod.rs

Lines 110 to 122 in 40186f3

	/// For variants holding an object, returns the object's instance ID.
	///
	/// If the variant is not an object, returns `None`.
	///
	/// If the object is dead, the instance ID is still returned. Use [`Variant::try_to::<Gd<T>>()`][Self::try_to]
	/// to retrieve only live objects.
	#[cfg(since_api = "4.4")]
	pub fn object_id(&self) -> Option<crate::obj::InstanceId> {
	// SAFETY: safe to call for non-object variants (returns 0).
	let raw_id: u64 = unsafe { interface_fn!(variant_get_object_instance_id)(self.var_sys()) };
	crate::obj::InstanceId::try_from_u64(raw_id)
	}

Since you only want to check for equality, this would save the Variant::to() marshalling, the clone() and the upcast() conversion.

Maybe something like:

Suggested change

	let object_script: Gd<Script> = object_script_variant.to();
	let instance = unsafe { get_instance_fn(object.obj_sys(), language.obj_sys()) };
	!instance.is_null() && object_script == script.clone().upcast()
	let instance = unsafe { get_instance_fn(object.obj_sys(), language.obj_sys()) };
	let script_id = object_script_variant.object_id().expect(...);
	!instance.is_null() && object_script.instance_id() == script_id

I tried backporting Variant::object_id() to <4.4 so you can use it universally, but encountered a bug while doing so. Will follow up on this 👍

[TitanNano]

So I was able to use upcast_ref() and __object_id() for the comparison, but the Variant::to() marshalling without Variant::object_id() right now.