R.merge and type intersection

[pzavolinsky]

Hi, I'm having problems with R.merge(a,b) where b is a subset of a:

interface Person {
  first: string,
  last: string
};
function setLast(last: string, person: Person) : Person {
  return R.merge(person, { last });
}

The code above complains with:

error TS2322: Type '<T1>(a: T1) => T1 & { last: string; }' is not assignable to type 'Person'.
  Property 'first' is missing in type '<T1>(a: T1) => T1 & { last: string; }'.

If I explicitly state T1 and T2, the code works but it gets very ugly:

function setLast(last: string, person: Person) : Person {
  return R.merge<Person, { last:string }>(person, { last });
}

Most of the time I use R.merge to update existing properties of immutable objects (like last in the example above). I think support for this use-case should be straight-forward. Worst case, this might be OKish:

R.merge<Person>(person, { last }) : Person & { last:string } // == Person

Having to specify the type of { last } feels like an unnecessary pain.

On a side note, for the use-case when I'm augmenting a type (i.e. A & B != A) I have no problem defining the new type.

[pzavolinsky]

More info on this.

It looks like the problem is a conflict in these two definitions:

merge<T2>(a: R.placeholder, b: T2): <T1>(a: T1) => T1&T2;
merge<T1, T2>(a: T1, b: T2): T1 & T2;

Removing the first one fixes the issue.

I recommend reviewing (and possibly removing) the R.placeholder variation in favor simplifying the most common use-case (at least in my experience) of using R.merge with one or two objects.

Incidentally, any R.placeholder scenario can be replaced with a R.flip(R.merge).

[donnut]

I just came up with the same finding after 8 days ;)
This works indeed.

The placeholder thing is something I'm not happy with at all. It doesn't fit typescript well, although I do see its use-case. It is probably better to make a clear choice and just not support them at all.

[pzavolinsky]

The placeholder thing is something I'm not happy with at all. (...) It is probably better to make a clear choice and just not support them at all.

Agreed, we decided not to use R.__ at all in our codebase (since it felt more obscure than the R.flip alternative).

[donnut]

I'll shoot a PR with the placeholders removed. If there are no serious objections, we remove them definitely.

[pzavolinsky]

Cool @donnut , thanks!