Use kind rather than type as discriminator in MF2 JSON data model

[lucacasonato]

Many languages use type as a keyword, which makes naming fields of (deserialised) structs type difficult. An example of this is Rust, where naming struct properties type is not possible without doing a fun dance of prefixing the field with r#.

Right now MF2 uses "type" as the discriminator in the JSON data model.

Can I suggest you use "kind" as the name for this discriminator to avoid this inconvenience? kind is generally not used as a keyword in programming languages.

[eemeli]

One issue with kind is that we're already using it, for markup.

Regarding type, my expectation would be that in languages like Rust the JSON representation of each node would be marshaled into some struct, which would not need a "type" field. Is that not a reasonable presumption?

type is also intentionally chosen to match the common practices in JavaScript and other ASTs, though that's not universal.

[lucacasonato]

Yes, it is possible that in Rust you would marshal into an enum for example. In this case the point becomes moot.

However, it's not so simple in other languages. In Go for example, type Foo struct { type string } is not valid.

I think it may also be confusing to have both a "kind" and "type" on the same object. Maybe a different name can be found for the markup "kind"? Maybe "position" ("position": "open", "position": "close" etc)?

I think "kind" is also not necessarily less common than "type" in other ASTs. A difference here from for example the JS ASTs expressed in JS, is that this AST format is designed to be used in many different languages, not just JS. As such, I think it's quite important that the format is easially portable to other languages. Part of making it feel "native" to more than just JS, is that you don't have to use weird incantations like r#type, or type_ in non JS languages.