Improvement notification export for GDExtensions

[snatvb]

Tested versions

• 4.2 stable

System information

Windows 11 - 4.2

Issue description

If user tries convert @export_file("*.txt") to Array:

extends Node

class_name NewScript

@export_file("*.txt") var files: Array[String]

LSP shows error:

And in console:

But if user does it in rust:

#[derive(GodotClass)]
#[class(init, base=Node)]
struct Foo {
    base: Base<Node>,
    #[export(file)]
    resources: Array<GString>,
}

#[godot_api]
impl INode for Foo {}

It leads to this:

For user it's unexpected and it would be rather to notify about incorrect annotation.

Steps to reproduce

• Add extenstion
• Describe incorrect export

Minimal reproduction project (MRP)

godot-rust-repro.zip

[lilizoey]

i dont think this is a godot issue? isn't this just an issue with the rust bindings? im not sure what godot could do here.

[Bromeon]

As I understood, there are two things:

1. A feature request/discussion whether @export_file("*.filter") could potentially be allowed on Array -- visualized as a list in UI.
2. If not: the question whether the error message "@export_file" requires a variable of type "String" but type "XY" was given instead is exclusive to the GDScript parser, or it can/could also be used for GDExtension properties.

On Rust side, we should in parallel make sure that the property is registered properly with Godot.

[dsnopek]

2. If not: the question whether the error message "@export_file" requires a variable of type "String" but type "XY" was given instead is exclusive to the GDScript parser, or it can/could also be used for GDExtension properties.

Currently, this message comes from gdscript_parser.cpp.

I suppose we could put similar validation in ClassDB::add_property(), however, the error message couldn't really be as good.

From Godot's perspective, it's looking at data passed on the PropertyInfo struct, and the error message could only refer to that (for example, saying the PropertyInfo::hint won't work with the given PropertyInfo::type). However, the binding probably doesn't directly expose PropertyInfo, but instead some binding-specific way of doing it, and so it'd be able to provide a better error message that refers to that.

So, I personally think giving a nice error message should be up to the GDExtension binding (in this case, the Rust binding).

[dsnopek]

Discussed at the GDExtension meeting:

• It would be nice to have an API that that could check if a property is valid, that could potentially be used by GDScript, ClassDB and GDExtension bindings, and still provide nice error messages.
• However, some bindings may be able to check for property validity at compile time (making this unnecessary)
• The property hints have decent documentation but we'd need to make sure it's complete and stays that way.
• We'd also need to ensure that the potential future validation API would remain in sync.
• For now, we recommend that GDExtension bindings attempt to handle this validation themselves, based on the documentation linked above.

[snatvb]

Now this works in 4.4.1:

extends Node

class_name NewScript

@export_file("*.txt") var files: Array[String]

Interesting, could we get it for native bindings?

[dsnopek]

could we get it for native bindings?

It is possible for native bindings, so long as the language binding exposes it somehow. In godot-cpp, this would be like:

ADD_PROPERTY(PropertyInfo(Variant::STRING, "file", PROPERTY_HINT_FILE, "*.txt"), "set_file", "get_file");