Suggestion Backlog Slog, 10/9/2017

[RyanCavanaugh]

• Allow --strict to take a version number #18776 Allow --strict to take a verison number
• Nested Tagged Unions #18758 Functions are ArrayLike
• Constrain return statements in generator functions. #18726 Constrain return statements in generators
• Dangerous "name" (and potentially others) global  #18433 Global name, again
• Allow class inheritance in ambient contexts, even with a base private constructor #18283 Allow ambient classes to inherit from private constructors
• Merging constants with namespaces #18163 Merging constants with namespaces
• Infer extra parameter types from default values #17559 Infer types from parameter defaults
• Support non-null assertion for destructuring #17390 Support non-null assertion in destructuring
• Arrow function is not compatible with ordinary one #17113 Allow arrow function to override prototype method
• Refine type of "typeof" operator #16314 Refine results of typeof operator
• Noncrashing property access through non-null assertion operator should narrow that property to NonNullable #15655 A successful property access should narrow the accessed expression to non-null
• Call signatures of union types #7294 masaeedu has a proposed algorithm for union type calls

[RyanCavanaugh]

• Allow --strict to take a version number #18776 Allow --strict to take a version number
  • We have no way to make this guarantee going forward
  • What does strict: 2.5 even mean?
  • The options themselves are there
  • 👎
• Remove batching on unittest thread, use historical data to inform batching #18578 Nested discriminated unions
  • Similar root cause to Type guards on variable break refinements on its properties #18840
  • Trying to figure out what to do without synthesizing tons of new types
  • Bug for Future?
• Functions are compatible with ArrayLike<any> #18757 Functions are ArrayLike
  • Numeric index signatures should not imply this behavior?
  • Try it on RWC and see what happens
  • indexTypesRelatedTo
    • if (!targetInfo || targetInfo.type.flags & TypeFlags.Any && !sourceIsPrimitive && (kind === IndexKind.String)) {
    • ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
• Constrain return statements in generator functions. #18726 Constrain return statements in generators
  • Technically there should be three type parameters
  • Breaking change?
  • Graceful change: Enforce return types only when explicit second type parameter is specified in return type annotation
  • Generic defaulted second type parameter should be OK
  • There's also an input type which is not currently represented
  • AMF
• Dangerous "name" (and potentially others) global  #18433 Global name, again
  • Approximately 0 people read this as a global on purpose
  • length is also bad
  • How do we just poison this completely?
  • void would be better, but still allows comparison to undefined
• Allow class inheritance in ambient contexts, even with a base private constructor #18283 Allow ambient classes to inherit from private constructors
  • 👍
• Infer extra parameter types from default values #17559 Infer types from parameter defaults
  • How does this happen?
  • The only way this parameter might show up is through an unsound alias anyway
  • Duplicate Extra arguments with defaults get implicit any even with noImplicitAny #17904
  • NMI around scenarios
• Support non-null assertion for destructuring #17390 Support non-null assertion in destructuring
  • Destructuring syntax is already super hairy
  • 👎
• Arrow function is not compatible with ordinary one #17113 Allow arrow function to override prototype method
  • 👍
• Refine type of "typeof" operator #16314 Refine results of typeof operator
  • Inhibits defensive programming
• Noncrashing property access through non-null assertion operator should narrow that property to NonNullable #15655 A successful property access should narrow the accessed expression to non-null
  • Would have to specifically special case x!.y as being equivalent to x && x.y
  • Or synthesize tons of new types ala Type guards on variable break refinements on its properties #18840
  • AMF
• Call signatures of union types #7294 masaeedu has a proposed algorithm for union type calls
  • We believe the only truly correct way is to try every call signature and union the return types
  • But it's going to be potentially very slow
  • Also has problems with binding and caching relating to contextual types
  • Could we do this without allowing context-sensitive types?
    • Would suck
    • But we could detect an illegal rebinding
  • How would we report errors in this case?
  • For the future 🚀

[falsandtru]

#18433 Global name, again

See also #15424

[RyanCavanaugh]

Thanks @falsandtru - that one was particularly impossible to search for 😆