Design Meeting Notes, 11/19/2021

[DanielRosenwasser]

Strict Class Initialization with Computed Properties

#45974

• PR closes a bunch of bugs.
• In strictClassInitialization, we check for "regular" properties, but not properties declared as comuted properties.
• Why is this different from other narrowings?
  • Well we have to specifically create a "reference" to understand how a computed
• Still adding to control flow graph.
• Need some eyes before we merge.

Issues with Parameters and ReturnType

#46855

• MethodKeys extends Methods<T> -> T with only its methods

  • T[Method] took us too long to expand, and said "maybe!" and let it pass through.
  • Now we can tell that this is really an invalid option.

• We "narrow" Methods based on some criteria, but then try to index back into T with it, and don't have any way to surface that criteria.

• Can use... a type helper called Cast 😬

  type Cast<T, U> = T extends U ? T : T & U;

• How often does this come up?

  • Mongoose types got hit, their conclusion was the type was wrong.

• Would be nice if we had a first-class primitive to handle this.

  • Higher order encoding?
  • Maybe just a new primitive.

• Aside: all of this is very hard to reason about!

• The Maybe behavior seems like the wrong thing.

  • Most upsetting thing about this is not the change in behavior, it's that we didn't do this before and people accidentally took a dependency.

satisfies Operator (Contextually Typed-by and Assignable-to Operator)

#7481

• Need to write down the 2-3 use-cases.

• Details like freshness, excess property checking, etc.

• Until then, hold off on the PR, but we want to use it to play with different behavior.

• Had a lot of syntax discussions, but the semantics are going to drive the syntax.

• Maybe use a different keyword to ensure we're not tying any preconceived ideas?

• What if you could write const anywhere in an as and it inferred the const-y version of the expression type? Then you could write

  type Validate<C, T extends C> = T;
  
  { x: 0 } as Validate<Point2D, const>;