Add type definitions

[ivanross]

This adds type definitions to traph function.

It is possible to have type checking for the object passed to traph function by passing two types: the expected input object type and the expected output object type.

type In = { values: number[] }
type Out = { count: number , mean: number }

traph<In,Out>({
	count: (i, o) => i.values.length
	mean: (i, o) => i.values.reduce(sum) / o.count
})

In each callback i type is In and o type is Out.

Also, the argument passed to traph must have all and only the keys of Out, where every key must have proper return type:

traph<In,Out>({
	square: (i, o) => ... // Error: square in not assignable 
	count: (i, o) => String(i.values) // Error: count type must be 'number'
})

This PR adds tsd test too.

[caesarsol]

Wow thanks!!!

Is it impossible to infer output types though?
We would lose the o typing, right?

[caesarsol]

Oh my god, I missed this library completely... I'm losing my edge

[ivanross]

🤯

[caesarsol]

How could we call both ava and tsd with the yarn test and yarn test:watch scripts?

[ivanross]

Not sure if I get what you mean. Are you wondering how to call both ava and tsd in watch mode at the same time with one call? However, a "watch" option seems to be missing in tsd

[ivanross]

I'm afraid so. I think that circular reference is unavoidable. I've tried to infer Output type from traph argument, but as you say we lose o typing.

With these types, the output is inferred when getters use only i parameter. For instance, if you look at the first test:

const transform = traph({
	fullName: i => `${i.first} ${i.last}`,
    initials: i => `${i.first[0].toUpperCase()}.${i.last[0].toUpperCase()}.`,
})

const output = transform({ first: 'Richard', last: 'Feynman' })

output is inferred to be { fullName: string, initials: string }. But as soon as you add o to one getter, even if it is not used, typescript fallbacks to Record<string,any>.