Expose resource names

[MikailBag]

Closes #73

TODO:

• [x] rollback temporary changes
• [ ] resolve "TODOs" connected with empty resource name

[Arnavion]

Please call it PLURAL or PLURAL_NAME rather than NAME.

[MikailBag]

I have renamed it to be a PLURAL_NAME

[MikailBag]

Last two commits additionally expose resource scope (namespace-scoped or cluster-scoped)

[MikailBag]

The most technically challenging part here (except parsing URLs :smile: ) is to provide enough information for type-safe cluster-scoped and namespaces resources distinction (e.g. kube-rs would like to statically prevent Api<Node>::namespaced).

In this PR, I propose roughly the following:

trait Scope {}
impl Scope for String {}
impl Scope for () {}

trait Resource {
    // ...
   type Scope: Scope;
}

// generated: 
// cluster-scoped resource example
impl Resource for Node {
   // ...
   type Scope = ();
}
// namespace resource example
impl Resource for Pod {
    // ...
   type Scope = String;
}

These type-level markers can then be used as type bounds in generic implementations.

However, the current state is not ideal, because the following code is rejected by the compiler:

struct Api<R>(std::marker::PhantomData<R>);

impl<R: Resource<Scope=String>> Api<R> {
    fn new(ns: &str) -> Self {
        todo!()
    }
}

impl<R: Resource<Scope=()>> Api<R> {
    fn new() -> Self {
        todo!()
    }
}

with an error duplicate definitions with name "new". It seems like a compiler limitation, and I can't think of a workaround. However, I think that even in the current state this PR exposes enough additional information to make generic API clients more correct and safe.

cc @clux

[Arnavion]

Yes, I went over this in https://github.com/Arnavion/k8s-openapi/issues/65#issuecomment-601839119

[Arnavion]

Just checking, this isn't ready for review yet, yes? I say this based on the unchecked TODO in the OP.

[MikailBag]

I think it can be reviewed.

I'd like to hear your opinion regarding those empty strings (e.g. this PR assigns empty resource name for every List, which is not really invalid, but can be confusing)

[Arnavion]

Is there a reason to use String and () ? If not, it should create new ClusterScopedResource and NamespaceScopedResource types and use those, like what I wrote in the second bullet-point in https://github.com/Arnavion/k8s-openapi/issues/65#issuecomment-601839119

• instead adding const SCOPE: ResourceScope to Resource with trait ResourceScope { } struct ClusterScopedResource; impl ResourceScope for ClusterScopedResource { } struct NamespaceScopedResource; impl ResourceScope for NamespaceScopedResource { } struct NotScopedResource; impl ResourceScope for NotScopedResource { }

Though since consts can't be used for trait bounds, it should be type Scope: ResourceScope; like your current implementation.

[Arnavion]

As for List's PLURAL_NAME it depends on what PLURAL_NAME would be used for (by you or anyone else)? If it's just for constructing CRUD URLs dynamically, it should be fine because such URLs don't exist for lists.