Fix newtype variant unwrapping inside enum, seq and map

[juntyr]

This is a follow-up to #319 which added the unwrap_variant_newtypes extension.

Unfortunately, it seems like I missed a few interactions with parsing enums, sequences and maps nested inside a newtype variant (i.e. not inside a containing non-newtype struct). I have fixed all three and added extra test cases to cover their intended functionality, i.e. NewtypeVariant( [ Struct(...) ] ), NewtypeVariant( { "key": Struct(...) } ), and NewtypeVariant( StructVariant(...) ) all parse again.

Discussion

Option<T> is another fun case. Here I think carrying through the unwrapping actually is quite neat, but I want to get your input if this should be the intended behaviour:

• NewtypeVariant( None )
• NewtypeVariant( Some(a: 4, bool: false) ) for Option<Struct> and struct Struct { a: u32, b: bool }
• NewtypeVariant( a: 4, bool: false ) is the implicit_some extension is switched on as well

I am also a bit unsure about how to handle deserialize_any. I think I may need your help with that one.

TODO

• I've included my change in CHANGELOG.md

[torkleyy]

Hey, thanks for the follow-up! I'm not sure when I will be able to review this.

I am also a bit unsure about how to handle deserialize_any. I think I may need your help with that one.

I don't think there's anything special to do for deserialize_any; if there's no type information given to us we don't need to implicitly generate anything. That does mean that self-described and statically typed data will not deserialize into the same thing, but that is also the case with other self-describing serde formats (and already the case with RON as well).

[juntyr]

I guess it might be easiest then to just turn off the newtype variant unwrapping if we enter deserialize_any? So then the semantics would be that unless type information is given, the unwrapping cannot be performed. Ideally there wouldn’t be a special case, of course, but some types (e.g. wrapped unit and struct) can only be parsed if they are described in the RON, which is exactly what the extension avoids.

[juntyr]

@torkleyy I've disabled newtype variant unwrapping inside deserialize_any as discussed and added the fix to the changelog

[torkleyy]

Thanks, looks good to me!