Design Meeting Notes, 6/13/2018

[DanielRosenwasser]

Rest parameters and tuple types

#24897

• In a nutshell, the type of a rest parameter can now be a tuple.

  • Previously we said rest parameters could only be arrays.
  • What this means is that an n-tuple-trailed rest parameter (i.e. (arg0: boolean, ...args: [number, string])) is the same as having n trailing parameters (i.e. (arg0: boolean, arg1: number, arg2: string)).

• Because we need to encode everything that parameter lists have, we need a few things: empty length, optionality and open-endedness parameters

  • Now have an empty tuple types.
  • Also now allowing tuple elements to be optional using ?.
    • Means that the length becomes a union of
    • Is the ? part of the type?
      • No, it's a modifier for tuple elements
    • We said last week we'd be okay giving the syntax up for this case.
      • In that context.
      • And plus, we already recognize this in JSDoc anyway, so we've already effectively given it up.
  • Rest args mean we need open-ended tuples.

• Currently can't operate on generic type argument lists.

• All methods are generic on the called function (they use a this parameter to infer the parameter types of the functions themselves.)

  • Can Function itself just be made generic instead of doing this?
    • Well, doesn't work with overloads.
    • Well there are just going to be things that don't work.

• What doesn't work?

  • Inferring from a generic argument list:

    function call<T extends unknown[], U>(f: (...args: T) => U, ...args: T) {
        f()
    }

Conditional type relationships

#24821

• Today, a type isn't assignable to a conditional type unless it's another conditional type.
  • Makes it hard to actually implement something that returns a conditional type.
• We also have what are called "substitution types": constraints that are built up in every conditional type's extends check.
  • If you're in the false branch, this PR builds up negated constraints.
• To perform assignability checks, we first assume that
• But a conditional type can give you a false positive for when the runtime type is more derived.
• What about T extends T ? ... : ...
  • Should still work.
  • Will have to think about this.
• Mixed assuredness

More conditional type relationships

#22735
#24929
#23541

• Use conditional type to narrow return type of function with a union-typed argument #24929
  • Seems okay, but only because they're disjoint.

Also, how far do you want to go?

function 

• We want to take baby steps.
  • One good baby step - if you're assignable to both branches, it should just work.

[ajafff]

How do open ended tuples work w.r.t. their length property. Is this an infinite union? Is this something range types (#15480) would be useful for?

[mhegazy]

How do open ended tuples work w.r.t. their length property.

it is a number.