Minimalist YAML parser with support for borrowed string slices

[dtolnay]

In dtolnay/serde-yaml#94 I would like to move serde-yaml off of yaml-rust as a YAML backend and to a much simpler backend with support for a minimum viable subset of YAML syntax.

I think the minimal API for this would be:

pub enum Yaml<'a> {
    // strings, booleans, numbers, nulls, all treated the same
    Scalar(&'a str),

    // flow style like `[x, x, x]`
    // or block style like:
    //     - x
    //     - x
    Sequence(Vec<Yaml<'a>>),

    // flow style like `{x: X, x: X}`
    // or block style like:
    //     x: X
    //     x: X
    Mapping(Vec<Entry<'a>>),
}

pub struct Entry<'a> {
    key: Yaml<'a>,
    value: Yaml<'a>,
}

pub fn parse<'a>(input: &'a str) -> Result<Yaml<'a>>;

The long tail of seldom used YAML features need not be supported. That means no tags, anchors, aliases, countless distinct string types, ...

In python world there's strictyaml with similar goal, if you want to check what elements are needed

[dtolnay]

Thanks! That looks like a super nicely done library with similar motivations as this project.

[jakeschurch]

@dtolnay I think I would like to take a crack at this

[dtolnay]

Just checking in -- any progress on this @jakeschurch? Is there anything you are stuck on?

[jakeschurch]

Thanks for checking in @dtolnay -- so far things are going well - will def. Check in if I have any questions or problems

[dtolnay]

Sounds good. I noticed that you forked serde-yaml and made some big changes to its Value type. I would recommend sticking to a 100% self contained fn parse<'a>(input: &'a str) -> Result<Yaml<'a>> instead -- serde_yaml::Value is a frontend type and shouldn't need to change at all when we later swap backends.

[jakeschurch]

OK. Sounds like I need to read a little more into the code.

[dtolnay]

Less into the code. 😉

Nothing in serde_yaml is relevant to implementing this function.

[jakeschurch]

Sorry! Just meant what parts of the code now hook up to yaml-rust so I can have the signatures in mind implementing the back-end

[jakeschurch]

hey @dtolnay just wanted to let you know that I haven't gotten around to implementing this yet cause I'm super busy with work - whatever you want to do, let me know -- sorry to keep you waiting

[pickfire]

I believe it would be nice to support the additional feature using feature gates. I will be trying this as it is a blocker for me to implement yaml2json.

[pickfire]

I added my implementation in https://github.com/pickfire/mini-yaml, very experimental. Currently can parse certain flow scalar, flow/block sequence, flow mapping and block implicit mapping. Still figuring out how to parse implicit flow mapping and maybe later need to check the spec again, maybe later can add serialization and deserialization.