Use GDExtension to_string in Node

[raulsntos]

Before this PR, when a GDExtension class that derives from Node implements _to_string like this:

String _to_string() const {
    return "MyClass";
}

Godot will print @MyNode@2:MyClass instead of just MyClass like it does for other Object-derived classes.

Matches the Object::to_string implementation:

godot/core/object/object.cpp

Lines 925 to 940 in e96ad5a

	String Object::to_string() {
	if (script_instance) {
	bool valid;
	String ret = script_instance->to_string(&valid);
	if (valid) {
	return ret;
	}
	}
	if (_extension && _extension->to_string) {
	String ret;
	GDExtensionBool is_valid;
	_extension->to_string(_extension_instance, &is_valid, &ret);
	return ret;
	}
	return "<" + get_class() + "#" + itos(get_instance_id()) + ">";
	}

[dsnopek]

Looks good to me!

[Bromeon]

is_valid has neither a defined input value (uninitialized), nor is its output checked -- is that intended?

[dsnopek]

Those are good questions! However, this code exactly matches what we already have in Object::to_string(), so it isn't making things any worse. :-) If we do end up evaluating how is_valid should be used, then it should be fixed in both places.

[Bromeon]

Thanks for clarification!

Since we'll likely forget to update one of the two places in that case, should we add a comment in both? (Or directly reuse a function, not sure how easy it would be).

[raulsntos]

Yeah, I just copied Object::to_string but I also thought it was weird that is_valid is unused. We should probably check before returning, like we do with valid in scripts:

godot/core/object/object.cpp

Lines 926 to 932 in e96ad5a

	if (script_instance) {
	bool valid;
	String ret = script_instance->to_string(&valid);
	if (valid) {
	return ret;
	}
	}

Note that valid is also uninitialized. I guess this variable is always assigned by to_string so there's no need to initialize it, but I'm not too familiar with C++ so if we should always initialize it just let me know.

[akien-mga]

At least the default implementations of to_string() seem to properly set r_valid to false before errors, and true on success.

That's a common pattern for calling methods on ScriptInstance or GDExtension, though it's hard to guarantee that it's always safe. The tradeoff between avoiding one potentially unneeded initialization and risking using uninitialized memory needs to be assessed (especially whether not initializing these actually makes a difference to performance at all).

It's a much bigger scope than just this PR though. For this I'd suggest just putting a comment in both implementations that they should be kept in sync (not for what to do with valid specifically, but the whole method).

[akien-mga]

Thanks!