between and sepBy return unknown type

[devurandom]

Following code has type errors with TypeScript 5.1.3:

import {
    between,
    char,
    regex,
    sepBy,
    sequenceOf,
    whitespace
} from "arcsecond";

const word = regex(/^\w+/);

const quotedString = between(char('"'))(char('"'))(regex(/^[^"]+/));

const attribute = sequenceOf([word, char("="), quotedString]).map(
    ([name, _, value]): [string, string] => [name, value]
);

const attributeList = sepBy(whitespace)(attribute).map(
    (values): [string, string][] => values
);

export const parse = (input: string) => attributeList.run(input);

❯ pnpm build

> arcsecond-repro@0.1.0 build [REDACTED]/arcsecond-repro
> tsc -p .

src/parser.ts:15:52 - error TS2322: Type 'unknown' is not assignable to type 'string'.

15     ([name, _, value]): [string, string] => [name, value]
                                                      ~~~~~

src/parser.ts:19:37 - error TS2322: Type 'unknown[]' is not assignable to type '[string, string][]'.
  Type 'unknown' is not assignable to type '[string, string]'.

19     (values): [string, string][] => values
                                       ~~~~~~


Found 2 errors in the same file, starting at: src/parser.ts:15

 ELIFECYCLE  Command failed with exit code 2.

I would expect:

• quotedString to contain type string, but TypeScript detects it as unknown.
  • Generally: I would expect between(_)(_)(T) to have type T instead of unknown.
• attributeList to contain type [string, string][], but TypeScript detects it as unknown[].
  • Generally: I would expect sepBy(_)(T) to have type T[] instead of unknown[].

Please find a complete example in https://github.com/devurandom/arcsecond-issue-109-repro.

[francisrstokes]

Hey @devurandom,
Thanks for reporting the issue, and thanks for taking the time to set up a proper repro environment! This is unfortunately a known issue with TypeScript, and how it resolves types in curried functions:

const f = (a: unknown) => (b: unknown) => (c: string): [typeof a, typeof b, typeof c] => {
    return [a, b, c];
}

const g = f(1)(true)("hello"); // -> const g: [unknown, unknown, string]

A workaround for this is to use as Parser<x, y, z> to ensure that your parsers are typed correctly.

[devurandom]

• So I always have to write sepBy(_)(t) as Parser<T[]> and between(_)(_)(t) as Parser<T>?
• Should this be added to the tutorial, for example ./tutorial/tutorial-part-4.md which gives examples for sepBy?
• Would it be possible to add a non-curried version of sepBy and between (and any other to which this applies), to avoid the TypeScript issue?
  • I have in the past often created functions like (x) => sepBy(_, x), but sepBy(_)(x) as Parser<X> wouldn't be the first thing I would try.
• Do you know which upstream issue in particular breaks this? I found several that might be related, but am not sure which or whether any of them relates to my problem:
  • Proposal: Partial Type Argument Inference microsoft/TypeScript#26242
  • Allow currying of generics microsoft/TypeScript#30123
  • Generic curried functions lose type information when partially applied within another generic type microsoft/TypeScript#46940
  • Type inference for higher-order, generic curried function breaks down when the function is applied to another generic function microsoft/TypeScript#49312

[devurandom]

Hey @devurandom, Thanks for reporting the issue, and thanks for taking the time to set up a proper repro environment! This is unfortunately a known issue with TypeScript, and how it resolves types in curried functions:

const f = (a: unknown) => (b: unknown) => (c: string): [typeof a, typeof b, typeof c] => {
    return [a, b, c];
}

const g = f(1)(true)("hello"); // -> const g: [unknown, unknown, string]

Could you explain this a bit more, please? The return type of g in your example seems quite expected to me.

What I have with between seems more like:

const f = (a: unknown) => (b: unknown) => (c: string): [typeof a, typeof b, typeof c] => {
    return [a, b, c];
}

const g = f(1)(true)("hello"); // -> const g: [unknown, unknown, unknown] //<< The third invocation / curry argument has the wrong type.

And confusingly the type seems semi-correct, getting the "array aspect" correctly:

const f = (a: unknown) => (b: unknown) => (c: string[]): [typeof a, typeof b, typeof c] => {
    return [a, b, c];
}

const g = f(1)(true)(["hello","world"]); // -> const g: [unknown, unknown, unknown[]] //<< TypeScript noticed the third invocation returns an array, but it got the type of the elements wrong.

[devurandom]

• Would it be possible to add a non-curried version of sepBy and between (and any other to which this applies), to avoid the TypeScript issue?
  • I have in the past often created functions like (x) => sepBy(_, x), but sepBy(_)(x) as Parser<X> wouldn't be the first thing I would try.

What I did as a temporary countermeasure:

import {
    between as _between,
    Parser,
    sepBy as _sepBy,
} from "arcsecond";

const between = <L, R, T>(
    left: Parser<L>,
    right: Parser<R>,
    value: Parser<T>
) => _between(left)(right)(value) as Parser<T>;

const sepBy = <S, T, E, D>(sep: Parser<S, E, D>, value: Parser<T, E, D>) =>
    _sepBy(sep)(value) as Parser<T[], E, D>;

A bit surprising was that between could not be defined as:

const between = <L, R, T, E, D>(
    left: Parser<L, E, D>,
    right: Parser<R, E, D>,
    value: Parser<T, E, D>
) => _between(left)(right)(value) as Parser<T, E, D>;

because TypeScript would report an error:

TS2345: Argument of type  Parser<L, E, D>  is not assignable to parameter of type  Parser<L, string, any> 
  Type  E  is not assignable to type  string 

probably because between is:

const between =
    <L, T, R>(leftParser: Parser<L>) =>
    (rightParser: Parser<R>) =>
    (parser: Parser<T>): Parser<T> =>
        sequenceOf([leftParser, parser, rightParser]).map(([_l, x, _r]) => x);

and not:

const _between =
    <L, T, R, E, D>(leftParser: Parser<L, E, D>) =>
    (rightParser: Parser<R, E, D>) =>
    (parser: Parser<T, E, D>): Parser<T, E, D> =>
        sequenceOf([leftParser, parser, rightParser]).map(([_l, x, _r]) => x);

which in turn is an effect of sequenceOf not taking E and D template arguments.

sepBy did not have this problem.