Create bevy_label crate

[hukasu]

Objective

Closes #17713

Solution

Create a crate bevy_label and update re-exports of DynEq on bevy_ecs, bevy_app, and bevy_render.

Testing

cargo run -p ci
cargo check -p bevy_ecs --no-default-features --features edge_executor,critical-section,bevy_debug_stepping,bevy_reflect --target x86_64-unknown-none

Migration Guide

Moved bevy_ecs::label to bevy_label and bevy_ecs::intern to bevy_label::intern

[mockersf]

#17713 lacks motivation apart from "it can be done". What advantages to you expect from this?

[bushrat011899]

Documentation and use of std need to be changed IMO.

Regarding whether we should do this: the most popular internment crate, internment, currently has zero support for no_std, but is otherwise mature and widely used in the Rust ecosystem already. Either:

1. We should polish this crate further, potentially trying to surpass internment and make bevy_label a fully standalone crate (e.g., disqualified). Or;
2. We should abandon Bevy's interning solution and improve internment based on our needs (e.g., no_std support).

I do think there is merit in trying to remove the labelling system from bevy_ecs; it's an already very large crate with a lot of responsibilities across a wide range of domains. Anything which can be split out without sacrificing DX should be pursued by virtue of making bevy_ecs itself easier to maintain. The sheer quantity of items in bevy_ecs which are pub(crate) alone makes it extremely daunting to make breaking changes unless you're well versed in how the whole crate works.

[bushrat011899]

If we're going to pull this out, this documentation should be far more fleshed out with some examples as to how people are supposed to use this crate.

[hukasu]

that i will leave open until a decision is reached on whether we will continue with this or change to internment, so maybe add the Needs docs label so that we don't forget

[alice-i-cecile]

I think my preference here is to spin out the functionality into its own Bevy-org crate, or improve an existing one enough that we can pull a dependency on it. This is useful to the broader ecosystem, loosely coupled and pretty stable.

[hukasu]

#17713 lacks motivation apart from "it can be done". What advantages to you expect from this?

@ItsDoot can you answer that?

[hukasu]

Also, discursion moves faster as a PR that as an issue, the issue existed for 3 weeks with no discursion whatsoever

[ItsDoot]

I do think there is merit in trying to remove the labelling system from bevy_ecs; it's an already very large crate with a lot of responsibilities across a wide range of domains. Anything which can be split out without sacrificing DX should be pursued by virtue of making bevy_ecs itself easier to maintain. The sheer quantity of items in bevy_ecs which are pub(crate) alone makes it extremely daunting to make breaking changes unless you're well versed in how the whole crate works.

FWIW, this^ is the motivation I was thinking of when making the issue. Whether that is worthy-enough motivation to split it is up to someone else 👍

[alice-i-cecile]

Whether that is worthy-enough motivation to split it is up to someone else 👍

Someone else here: that's worthy enough motivation to me ;) We're also reusing similar stuff across rendering, and may see this pattern crop-up elsewhere in the engine.