Support collecting with custom allocators about chumsky

[safasofuoglu]

Initial attempt:

fn f32vec<'a, E: Error<'a, &'a [u8]> + 'a, T: FromLexical>(
) -> impl Parser<'a, &'a [u8], (), extra::Full<E, bumpalo::collections::Vec<'a, T>, ()>> {
    number::<{ lexical::format::STANDARD }, _, _, _>()
        .padded()
        .map_with_state(|number, _, vec: &mut bumpalo::collections::Vec<T>| vec.push(number))
        .separated_by(just(b","))
        .collect::<()>()
        .delimited_by(just(b'['), just(b']'))
}

let str = b"[2, 5]";
let bump = Bump::new();
let mut vec = bumpalo::collections::Vec::new_in(&bump);
let res = f32vec::<Rich<u8>, f32>().parse_with_state(str, &mut vec);
dbg!(res);
dbg!(vec);

While it's technically possible to invoke f32vec manually, it's unclear how to use it in combinators, especially when and how the Vec gets created. Too eager creation would defeat the purpose. Ideally, f32vec would take a reference to bump and return an owned bumpalo::collections::Vec.

[Zij-IT]

Here is the best I got. It's not ideal, as it would be nice to have foldr/foldl where the initial value is constructed with access to the state with something like: Fn(&mut E::State) -> A

use bumpalo::collections;
use bumpalo::Bump;
use chumsky::{error::Error, prelude::*};
use lexical::FromLexical;

fn f32vec<'a, 'b: 'a, E: Error<'a, &'a [u8]> + 'a, T: FromLexical>(
) -> impl Parser<'a, &'a [u8], collections::Vec<'a, T>, extra::Full<E, &'b bumpalo::Bump, ()>> {
    let number = number::<{ lexical::format::STANDARD }, _, _, _>().padded();

    empty()
        .map_with_state(|_, _, bump| bumpalo::vec![in *bump])
        .foldl(number.separated_by(just(b",")), |mut vec, i| {
            vec.push(i);
            vec
        })
        .delimited_by(just(b'['), just(b']'))
}


#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn empty() {
        let str = b"[]";
        let bump = Bump::new();
        let res = f32vec::<Rich<u8>, f32>()
            .parse_with_state(str, &mut &bump)
            .into_result();

        assert_eq!(res, Ok(bumpalo::vec![in &bump;]));
    }

    #[test]
    fn one_elem() {
        let str = b"[2]";
        let bump = Bump::new();
        let res = f32vec::<Rich<u8>, f32>()
            .parse_with_state(str, &mut &bump)
            .into_result();

        assert_eq!(res, Ok(bumpalo::vec![in &bump; 2.0]));
    }

    #[test]
    fn two_elem() {
        let str = b"[2, 5]";
        let bump = Bump::new();
        let res = f32vec::<Rich<u8>, f32>()
            .parse_with_state(str, &mut &bump)
            .into_result();

        assert_eq!(res, Ok(bumpalo::vec![in &bump; 2.0, 5.0]));
    }

    #[test]
    fn five_elem() {
        let str = b"[1, 2, 3, 4, 5]";
        let bump = Bump::new();
        let res = f32vec::<Rich<u8>, f32>()
            .parse_with_state(str, &mut &bump)
            .into_result();

        assert_eq!(res, Ok(bumpalo::vec![in &bump; 1.0, 2.0, 3.0, 4.0, 5.0]));
    }
}

While it's technically possible to invoke f32vec manually, it's unclear how to use it in combinators, especially when and how the Vec gets created. Too eager creation would defeat the purpose. Ideally, f32vec would take a reference to bump and return an owned bumpalo::collections::Vec.

If you are referring to the creation of the Vec, you should know that just like std::collections::Vec, bumpalo::collections::Vec don't allocate until elements are pushed into them

[zesterer]

It's not ideal, as it would be nice to have foldr/foldl where the initial value is constructed with access to the state

And foldl/r_with_state isn't suitable?

[Zij-IT]

I'm using foldl but I have to lead the parser with an awkard empty().map_with_state(...) and drop the two first arguments @zesterer

Perhaps something like this:

    number::<{ lexical::format::STANDARD }, _, _, _>()
        .padded()
        .separated_by(just(b","))
        // Assume `bumpalo::Vec` impl `Collection`
        .collect_from_state(|bump| bumpalo::vec![in *bump])
        .delimited_by(just(b'['), just(b']'))

[zesterer]

Oh, I see. Yes, that's a bit annoying. I wonder whether it would be feasible to write some sort of Container impl for bumpalo's vector (extending the Container trait to have access to the parser state).

[zesterer]

Ah, you got there first 😀

[safasofuoglu]

@Zij-IT wonderful example!

I guess things can be made more ergonomic, but nice to know there's a way with the current facilities and it's not so bad.