Design Meeting Notes, 10/2/2017

[DanielRosenwasser]

throw expressions

TC39 proposal: https://github.com/tc39/proposal-throw-expressions
PR: #18798

• Already have a PR.
• Current open problems:
  • Control flow analysis additions and reachability checks at the expression level.
• Should we put this under experimental?
  • Like the idea of being able to actually experiment with behavior more openly before shipping, especially with a stable feature like this.
  • However, doing this and calling everything "experimental" may not give people a clear picture of how experimental any one feature is.
• What about other statement level constructs?
  • Committee seems 50/50 on whether they want anything beyond control-flow constructs.
• Conclusion: let's just wait on stage 3, it seems like we may reach consensus relatively soon.

ECMAScript private fields

TC39 Proposal: https://github.com/tc39/proposal-private-fields
Issue: #9950

• Statics?
  • Ready but apparently issues?
• Are declarations required outside the constructor?
  • Yes.
• What about structural compatibility?
  • External to the class, these types are ostensibly the same.
  • What about methods that take other instances and use their privates?
  • Too complex; run-time private should give the same compatibility rules as design-time private.
  • Also,
• Downlevel emit?
  • WeakMap down to ES2015?
    • But performance!
  • Name mangling?
    • But not really private.
    • Object.defineProperty to avoid enumerability?
  • The driving reason for this feature was hard private. Anything we do here should have hide privacy.
• Conclusion: Explore performance of WeakMaps and WeakMap polyfills first?

Strict Function Types

#18654

• Most recent change: only methods and constructor types are exempty from the check.
  • Causes more breaks, but makes more consistent & predictable behavior.
  • Hack to get back to bivariant mode (x: Foo) => Bar becomes { " bivariant method"(x: Foo): Bar }.
• Also implementation resets errors after measurement checks when any type parameter was invariant.
  • Experimented with assertion-level syntax.
• One thing we've seen is that finding out why types have become invariant becomes extremely difficult!
• Problem: diff mapped type construct doesn't have appropriate variance measurements.
• Let's get that in this afternoon.

Allow dynamic names

• Let's try to get this into the RC.