SystemSet with PhantomData, or ordering system by data dependencies

[hyranno]

Hello.
I'm trying to determine system ordering by data dependencies. What I want is like this.
The system produce the subjects by the dependencies.

pub fn add_data_dependency<Marker>(
    app: &mut App,
    system: impl IntoSystemConfigs<Marker>,
    subjects: Vec<Box<dyn SystemSet>>,
    dependencies: Vec<Box<dyn SystemSet>>,
) {
    /* ... */
}

So I wrote the code below, to represent SystemSet which produces data T.
But this doesn't compile, saying T needs to be Clone, Eq, Send, ... and so on.

use std::marker::PhantomData;

use bevy::prelude::*;

#[derive(SystemSet, Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub struct DataLabel<T: 'static> {
    _phantom: PhantomData<T>,
}
impl<T: 'static> DataLabel<T> {
    pub fn set() -> Box<dyn SystemSet> {
        Box::new( Self{_phantom: PhantomData } )
    }
}

I found Rust issue which may relate (rust-lang/rust#26925), but not understand well.

Is there a way to make it compile?
Or, existing way to ordering system by data dependencies?

more explanation

For example, consider system1 write ComponentA and system2 read ComponentA.
I say system2 depends ComponentA.
In standard form, we can system2.after(system1).
But I want to write system2.after(DataLabel<ComponentA>), where DataLabel<ComponentA> is SystemSet which may write to ComponentA. I mean this for ordering system by data dependencies.
In this form, we don't need to list all systems which may write to ComponentA. We can just think about the system2.
I'm trying to write function add_data_dependencies which add system to DataLabel<Subject> and order them after DataLabel<Dependency>.
Then the problem, I can't write DataLabel<T>, the SystemSet with PhantomData<T>.

[nicopap]

I don't understand what you mean by "ordering by data dependency". You seem to want to have just a way to order systems? Check out the following example: https://github.com/bevyengine/bevy/blob/v0.11.3/examples/ecs/ecs_guide.rs#L293-L315

You have two ways to order systems:

1. Using .after(foo) and .before(bar) combinators

app.add_systems(Update, (first_system.before(second_system), second_system));

2. Using .chain()

app.add_systems(Update, (first_system, second_system).chain());

Within a single .chain(), you can also run systems in parallel as follow

app.add_systems(Update, (
    (system1a, system1b, system1c),
    system2a,
    (system3a, system3b),
).chain());

Maybe you don't mean that? But I really don't understand what "Ordering by data dependency" means. Fundamentally, even if two systems access the same data (for example, a resource or component), you still need to specify an order. It's not enough to know that two systems access the same data, hence, it's not possible to "order by data dependency". But I don't think that's what you mean.

[hyranno]

Thank you for reply, and sorry for lack of explanation.
I added explanation for what I'm trying to do.

[jancespivo]

Hi, it might be completely unrelated but I struggled with PhantomData as well until I found this recipe:

#[derive(Default)]
pub struct MyStruct<T> {
    marker: PhantomData<fn() -> T>
}

And also I would start with the minimal solution so when you want to derive SystemSet and set bound T to Component

#[derive(Default, SystemSet)]
pub struct DataLabel<T: Component> {
    _phantom: PhantomData<fn() -> T>,
}

I hope I've helped at least a bit ;) I'm quite curious what you want to achieve.

[hyranno]

Unfortunately, that doesn't work.
impl thing doesn't compile yet, saying trait bounds on T are not satisfied.

[hyranno]

Well, I decided to implement traits by hand without derive macro.
And what I wanted to do (ordering system by data dependencies) is now become a tiny crate : https://github.com/hyranno/bevy_tiny_data_flow .

Then, now I found #8595 .
Looks like better.

Any way, thank you!