Feedback on oneof

[gcanti]

💡 Idea

Currently oneof;

1. forces the user to add a type annotation at the call site
2. doesn't statically enforce at least one arb in input

Repros:

import * as fc from 'fast-check'

fc.oneof(fc.string(), fc.boolean()) // error
fc.oneof<string | boolean>(fc.string(), fc.boolean()) // ok

import * as fc from 'fast-check'

fc.oneof() // no error
fc.oneof(...[]) // no error

Not sure whether you may be interested (type level tricks are nice when they work, however they could add some maintenance burden) but here's a possible solution:

import * as fc from 'fast-check'

export declare function oneof<A extends [unknown, ...Array<unknown>]>(
  ...arbs: { [K in keyof A]: fc.Arbitrary<A[K]> }
): fc.Arbitrary<A[number]>

// 1)

// const arb: fc.Arbitrary<string | boolean>
const arb = oneof(fc.string(), fc.boolean()) // ok

// 2)

oneof() // error
oneof(...[]) // error

[dubzzz]

Having better typings for oneof would definitely be a great thing.

The fact to avoid users to explicitly type <string, number> on fc.oneof(fc.nat(), fc. string()) would be pretty cool.

If you want to I can let you open a PR for that purpose. Otherwise I can handle it in the future.

---

Concerning the idea of forcing at least one parameter at call site, it was originally the case but the constraint has been removed by 1833a4b

I'll need to dig a little bit more to understand why. But I believe that it caused issues when combined with other arbitraries or chain:

fc.set(fc.uuid(), 1, 10)
  .chain(allUuids => fc.record({
    oneRecord: fc.oneof(...allUuids),
    records: fc.constant(allUuids),
  }))

Enforcing at least one parameter with typings might break existing code (that is working today) 🙄

[gcanti]

Enforcing at least one parameter with typings might break existing code (that is working today)

Agree, technically is a breaking change

[dubzzz]

As I'm planning a next major of fast-check (mostly to add a full support for esm), I'll maybe take advantage of it to include your request into it 👍

[jasikpark]

I believe this issue has been resolved, right? I've had no problems w/ needing to specify .oneof generic types manually nowadays 👍

[dubzzz]

It should not be that far indeed 🙂 But I think we still support the empty case from a typing point of view. If so it should not be that hard to forbid and finally close this old issue.

[jasikpark]

woot