Do you want to request a _feature_ or report a _bug_?
A feature
With typescript it's possible to have
interface Foo {
a: string;
b: string;
}
type Bar = Pick<Foo, 'a'>;
// identical to
/*
interface Bar {
a: string;
}
*/
Is it technically possible to have something similar in io-ts that looks like
const FooT = t.interface({
a: t.string,
b: t.string,
});
const BarT = t.pick(FooT, t.literal('a'));
?
You can use a "standard" pick function
declare function pick<O, K extends keyof O>(o: O, keys: Array<K>): Pick<O, K>
const BarT = t.type(pick(FooT.props, ['a']))
Indeed.
2 points though:
pick should be implemented somewherePick<Foo, 'a'> second argument type checks that it makes sense. Pick<Foo, 'something'> wouldn't pass a type check. pick(FooT.props, ['a']) would accept anything as a second argument.Hence why I started this discussion: given io-ts already implements some base generic TS types like Record<> and Partial<>, would it make sense to provide an implementation for the Pick<> as well that carries its semantics?
You can use a "standard"
pickfunction
I also don't think that that solution works for intersection codecs, as far as I can tell. t.intersection([t.type({foo: t.string}), t.partial({bar: t.number})]) doesn't seem to have a props field, but i still want to pick from it as i can with merged TypeScript interfaces.
intersection has a types array were you can access the inner types. Then you can use props again.
Ah, makes sense. That said, it seems kinda silly that I'd have to know the inner implementation of io-ts to achieve this
That said, it seems kinda silly that I'd have to know the inner implementation of io-ts to achieve this
You don't have to, but you need to understand what an intersection type is. Then you can better reason about why it is the way it is. An intersection does not have to consist only of object like types.
const A = t.intersection([t.string, t.type({ bar: t.number })])
so without the extra layer of types you can not work with an intersection type like this.
except to execute a pick wouldn't i have to know about the inner types array and props field for type? that seems like a significant amount of undocumented implementation details of the particular library, and while i'm all for knowing it inside and out for your personal edification/power use, i don't see why I'd have to educate the rest of my team on this kind of stuff just so they can implement a pick on their own.
As for many open source projects the documentation is suboptimal, but types is document here https://gcanti.github.io/io-ts/modules/index.ts.html#intersectiontype-class. Providing additional documentation is always welcome so is new functionality too.
@mlegenhausen is it my original post still discussed? I'm not sure I understand how intersection has anything to do with Pick<T, K>?
@zerkms it is related because you can define new interface like types with an intersection, that combine multiple t.interface and t.partial definitions to a new type. For this new type you could also define a pick function as it would work without io-ts.
@mlegenhausen I'm not sure I'm following: Pick<T, K> where K is a keyof T.
With intersection-based solution you must declare both types of the keys and the values. That's the significant difference: I'd rather infer the value than have to declare it manually.
@zerkms this is valid typescript
type X = Pick<{ foo: number } & { bar?: string }, “bar”>
With a pick function, I assume this should work:
const x = pick(intersection([type({ foo: number }), partial({ bar: string })]), [“bar”]);
because it’s analogous TypeScript to io-ts code.
@osdiab would it accept ['any rubbish'] as its second argument?
Ah I see what you’re saying, but I feel fairly confident it’s possible for it to be inferred with a recursive type.
EDIT: working version at https://github.com/gcanti/io-ts/issues/300#issuecomment-554845184
type KeyOfCodec<Codec extends Any> =
Codec extends Intersection ?
KeyOfCodecs<Codec[“types”]>
: Codec extends Type ?
keyof Codec[“props”]
: something // other cases
// not sure if this is constructed correctly,
// if a fully recursive variadic type can’t
// work properly then it can at least be
// manually specified for a reasonable
// number of array lengths
type KeyOfCodecs<Array extends Any[]> =
Array extends [Head, ...Tail] ?
KeyOfCodec<infer Head> | KeyOfCodecs<infer Tail>
: never
Ah for the array this would be relevant?
https://github.com/microsoft/TypeScript/issues/25947#issuecomment-446916897
https://github.com/microsoft/TypeScript/issues/5453#issuecomment-419680547
This seems to work when I tried it out on my machine for inferring the keys properly, Typescript 3.7.2:
type Head<T extends any[]> = T extends [any, ...any[]] ? T[0] : never;
type Tail<T extends any[]> =
((...args: T) => never) extends ((a: any, ...args: infer R) => never)
? R
: never
export type KeysOfCodecs<Codecs extends Mixed[]> = {
recurse: KeyOfCodec<Head<Codecs>> | KeysOfCodecs<Tail<Codecs>>
end: never
}[Codecs extends [Mixed, ...Mixed[]] ? "recurse" : "end"];
export type KeyOfCodec<Codec extends Mixed> =
Codec extends IntersectionC<infer Codecs>
? KeysOfCodecs<Codecs>
: Codec extends TypeC<infer TypeProps>
? keyof TypeProps
: Codec extends PartialC<infer PartialProps>
? keyof PartialProps
: never;
export function pick<Codec extends Mixed, Keys extends KeyOfCodec<Codec>>(
codec: Codec,
keys: Keys[]
): PickCodec<Codec, Keys> { // PickCodec not implemented yet
throw new Error("not yet implemented")
}
Not sure how stable that is for TypeScript versions, as the inference of tuples is definitely a feature in flux in TypeScript, wouldn't work for old TypeScript versions for sure
@gcanti Is the canonical solution still to use a generic pick, or is io-ts going to provide an implementation at some point? Thanks!
is io-ts going to provide an implementation at some point?
@VanTanev No, I don't think so
is io-ts going to provide an implementation at some point?
@VanTanev No, I don't think so
@gcanti How about adding it to the non-core package io-ts-types?
How about adding it to the non-core package io-ts-types?
@ivawzh I'm not against that, however if the solution is not implementable using the new experimental modules, it's not going to last
For folks looking to make this work only for a simple t.type (like me), this seems to work reasonably well:
export function pick<P extends t.Props, K extends keyof P>(
Model: t.TypeC<P>,
keys: K[],
): t.TypeC<Pick<P, K>> {
const pickedProps = {} as Pick<P, K>;
keys.forEach(key => {
pickedProps[key] = Model.props[key];
});
return t.type(pickedProps);
}
Usage:
const PickedModel = pick(Model, ["id", "name"]);
type PickedModel = t.TypeOf<typeof PickedModel>;
And to complement's @mDibyo pick, this works for omit
export function omit<P extends t.Props, K extends keyof P>(
Model: t.TypeC<P>,
keys: K[],
): t.TypeC<Pick<P, Exclude<keyof P, K>>> {
const allKeys = Object.keys(Model) as K[];
const keesToKeep = allKeys.filter((x) => !keys.includes(x)) as Exclude<
typeof allKeys,
typeof keys
>;
return pick(Model, keesToKeep);
}
First, I really appreciate the work on this library. It's pretty epic. I've been evaluating io-ts, and I saw this compatibility chart which made me happy that this was tracking feature parity with TypeScript. omit and pick seem to be a point of divergence, so I'm curious what the philosophy of the project is going forward on maintaining parity.
Most helpful comment
Indeed.
2 points though:
pickshould be implemented somewherePick<Foo, 'a'>second argument type checks that it makes sense.Pick<Foo, 'something'>wouldn't pass a type check.pick(FooT.props, ['a'])would accept anything as a second argument.Hence why I started this discussion: given
io-tsalready implements some base generic TS types likeRecord<>andPartial<>, would it make sense to provide an implementation for thePick<>as well that carries its semantics?