Design Meeting Notes, 4/3/2019

[RyanCavanaugh]

• Should we put Omit in lib.d.ts? Add Omit as a known interface #30455

  • In principle, no objections
  • What are the breakages?
    • 1 package on DT was impacted
  • BigTsQuery found 3 programs
  • The fix is trivial
  • Should we constrain K in the official definition?
    • Nah
    • Extract is not constrained either
    • The type can resolve either way
  • 👍

• Changing the default type constraint from {} to unknown Change the default type parameter constraints and defaults to unknown from {} #30637

  • Zero breaks on DT because Wesley fixed them
  • Breaks in RWC were in places where things actually could be null/undefined
    • 2-3 files where this happened
    • Typically e.g. foo<T>(x: T) { x.hasOwnProperty(...)}
  • Should we issue custom error messages?
    • Nah
  • 👍

• JSDoc feature parity [Design Policy] Consider JSDoc feature parity with Typescript #30624

  • object: Fine to treat this as object in input positions, but in output positions this is fatal because JS is often under-specified
  • String/string has a similar problem
  • interface - can't extend something
  • Key question: Should everything you can write in TS have a JSDoc equivalent?
    • You want something like "Let me write TS in this JSDoc comment"
    • We don't want to be inventing JSDoc syntax for all of our stuff
    • "Inline declaration file"
      • Why not make a .d.ts file?
        • Scoping alignment
      • Comparison with flow's :: syntax
        • Is /* :: widely-used amongst flow programmers?
      • /** @ts ...
        • What is the lexical scoping of this?
        • What are the legal positions of this sort of comment?
          • Legal at positions where a declaration would have been permitted
      • Type arguments (?)
      • Is this for annotation of functions, or for top-level declarations?
        • Both
      • How do you write module-scoped types?
        • Merge module scopes .js and .d.ts files of the same name?
      • What about the straightforward parser limitations?
      • What are the use cases?
        • "We exist on a line, not a square"
        • But this complicates the transitional story
          • It's important that we not add "false paths" of syntax that can't be transitioned
      • We started out to understand JS Doc, not invent new syntax
        • We're not going to support everything
        • Need to stop adding new JSDoc-specific behavior
      • Stake in the ground: Full feature parity is not a goal
      • We should be supporting extant JSDoc
  • as const should work in JSDoc cast syntax (bug)