Suggestion Backlog Slog, 10/8/2018

[RyanCavanaugh]

• Type-mutating functions #27568 Imperative type guards
• Assignability rule for conditional types needs to require check types and distributivity to be identical #27118 Assignability of conditional types (co/contra/invariant?)
• Proposal: Add an "exclusive or" (^) operator #14094 ^ type operator
• Object spreads should retain shallow type narrowing #26645 Apply narrowings to top-level spread results
• Proposal: strict flag to prevent uninferrable generic types from being inferred as {} #27288 Option to error when generic is inferred as {}
• Report errors on all plausible overloads #27249 Report errors on all plausible overloads
• Covariant assignability of type guard functions is unsound #26981 Covariant assignability of type guards
• Error output: Don't deep compare to native objects #26986 Customize error messages for built-in types

[RyanCavanaugh]

• Type-mutating functions #27568 Imperative type guards
  • No control flow nodes to attach this to
  • 👎 for same reasons
• Assignability rule for conditional types needs to require check types and distributivity to be identical #27118 Assignability of conditional types (co/contra/invariant?)
  • Should test if check type is identical, not subtype
  • The code in checker.ts has good aesthetics, but is wrong
  • Anders to investigate
• Proposal: Add an "exclusive or" (^) operator #14094 ^ type operator
  • New mapped types workarounds seem good enough
  • 👎
  • Long sidecar discussion of what it means to declare an optional never property (TL;DR don't)
• Object spreads should retain shallow type narrowing #26645 Apply narrowings to top-level spread results
  • Duplicate to the other issue
• Proposal: strict flag to prevent uninferrable generic types from being inferred as {} #27288 Option to error when generic is inferred as {}
  • Rule?: Error when zero candidates are present
    • What does this do to "unused" generics?
  • Some instantiations occur outside of situations where we could issue an error
  • Is inferring unknown better?
  • Daniel to investigate what the world looks like today
• Report errors on all plausible overloads #27249 Report errors on all plausible overloads
  • Proposed algorithm:
    • Filter by arity
    • Pick the one(s) that have the "most" matches (non-erroring arguments)
    • "Closeness" algorithm
  • Add a related span pointing to the overload we picked
• Covariant assignability of type guard functions is unsound #26981 Covariant assignability of type guards
  • Breaking change
  • What's the value?
  • Argument of whether people understand that "correct" type guards must return true for all values in their domain
  • See how bad switching to invariant would be
• Error output: Don't deep compare to native objects #26986 Customize error messages for built-in types
  • Stop elaboration in certain cases where the target type is "knowably" not a thing you would substitute in
  • Be smarter about certain relational errors
    • T | U assigned to T
    • T[] assigned to T or vice versa
    • Promise<T> to T or vice versa
    • T would be assignable to U but not vice versa
    • ...
    • What else?

[jack-williams]

Re: #27118

Should test if check type is identical, not subtype

Has the team looked at the impact of switching from bivariance to invariance? I don't really have an intuition for what is used in RWC, but I could imagine something like Exclude<A, null> <: Exclude<B, null> if A <: B being useful.

[inoyakaigor]

#14094 ^ type operator
New mapped types workarounds seem good enough

Is it mean what ^ operator will not be added?

[mattmccutchen]

@jack-williams Let's take the discussion of #27118 to that issue.