[#66] Add mutable metadata trait

[ctron]

I had to create two new methods (but maybe there is a better way):

    /// Gets the mutable metadata of this resource value.
    fn metadata_mut(&mut self) -> Option<&mut<Self as Metadata>::Ty>;

    /// Sets the metadata of this resource value.
    fn set_metadata(&mut self, metadata: <Self as Metadata>::Ty);

The first allows to get an Option, holding a mutable reference to the metadata. However when using Default::default(), you get empty metadata and need a way to set it. I would have used fn metadata_mut(&mut self) -> &mut Option<<Self as Metadata>::Ty> instead, but in some cases the value is not backed by an Option, but rather by the actual value.

So I added a second set_metedata function, which allows to set the value. As in some cases the value may be backed by the actual value, I don't think a "clear" method is possible.

[Arnavion]

Do you happen to know if metadata needs to be optional in the first place?

ObjectMeta was added in https://github.com/kubernetes/kubernetes/commit/55163a7df1b50824155560216b1edd9f4aec5365#diff-9ff06ad723720f9428a65da3710cf436R423 and it was already json:"omitempty" at the time. Then every json:"omitempty" field gained the +optional annotation in https://github.com/kubernetes/kubernetes/commit/7414cafbeb378dcc4f6fe6ef402c2d54c0ee5633#diff-9ff06ad723720f9428a65da3710cf436R1926 So on the face of it having it be optional is intentional by upstream, but I'm not sure there is ever any situation where either the client's request or the server's response can omit it.

[ctron]

I don't think metadata is optional. Reading it back, you will always have at least things like the selflink or the uid. When creating initially, I think in every case you need to put the name field.

Also I think that omitempty doesn't necessarily translate to Option. e.g. a string in go, which is "" (empty) would not be serialized with omitempty. Having a *string( pointer to string) would serialize "" with omitempty. See: https://play.golang.org/p/Ip17iRwOudA

This works only with primitive types. Types which can be "empty". e.g. false' for boolean,""for string,0for integers, ornilfor pointers. However a struct cannot be "empty" (unless it is a pointer). So even if a struct has only "empty" fields, it is serialized: https://play.golang.org/p/HP9A86s_e-C (threeis missing, buttwo` is not).

However, when deserializing, and not using pointers, you get the full tree deserialized, without the case of having nils.

Translating this to Rust, I would propose to use e.g. #[serde(skip_serializing_if = "Map::is_empty")] for things that can be empty and are marked omitempty. And only use Option when a pointer value is used: string -> String and *string -> Option<String>.

[Arnavion]

Also I think that omitempty doesn't necessarily translate to Option.

That annotation was removed from the golang code later and only +optional remains. But anyway it doesn't matter; the codegen is generated from the swagger spec, which says that the metadata field is not required, so representing it as Option is correct.

[ctron]

Also I think that omitempty doesn't necessarily translate to Option.

That annotation was removed from the golang code later and only +optional remains. But anyway it doesn't matter; the codegen is generated from the swagger spec, which says that the metadata field is not required, so representing it as Option is correct.

Sounds reasonable. So do you think this change is ok? Or do you want any changes?

[Arnavion]

Well, the question still stands about whether metadata needs to be optional or not. It would be nice to resolve that so that we don't need both metadata_mut and set_metadata in the trait. But I'm not sure we can resolve it between the two of us. (I'd previously asked this of the folks at kube too but they also did not opine on the matter.)

Anyway, can you review https://github.com/Arnavion/k8s-openapi/compare/fix-66 ? It's your PR + some whitespace changes (added spaces around the = in set_metadata's expr, removed one extra trailing newline from the impl Metadata generated code) + rebased on latest master.

[ctron]

Anyway, can you review https://github.com/Arnavion/k8s-openapi/compare/fix-66 ? It's your PR + some whitespace changes (added spaces around the = in set_metadata's expr, removed one extra trailing newline from the impl Metadata generated code) + rebased on latest master.

:+1: looks good to me, thanks!