Suggestion Backlog Slog, 4/16/2018

[RyanCavanaugh]

The "Ryan is optimistic about our pace" edition

• Support a @nonnull/@nonnullable JSDoc assertion comment #23405 Support non-null assertion operator in JSDoc
• Consider collapsing multiple repeated type-related error messages when in --pretty #23393 Collapse multiple error reports in --pretty
• Warn on unused property in object literal with spread #23340 Warn for excess properties in spreaded object literal
• Uninitialized variables work around strictNullChecks (follow-up to #13884) #23305 Enforce never-initialized varibles / strictVariableInitialization
• Feature Request / Proposal: constructor type guard #23274 Type guards by expr.constructor (sameish as instanceof)
• Inconsistent type compatibility for a type with a call signature and and index signature #23226 isObjectTypeWithInferableIndex returns false for intersection type with index signature
• Support locally scoped type alias nodes #23188 Locally-scoped type aliases
• Comparing constrained generic types/substitution types to conditional types #23132 Improve behavior of conditional types instantiated with constrained generics
• T[K] should be never when T is never #23005 T[K] should be never when K is never
• Using string enum as key in {[key]: Type} throws not an index signature #22892 string enum object should be indexable type
• esModuleInterop should work even when compiling to esnext modules #22851 esModuleInterop should work when target = esnext
• Abstract classes that implement interfaces shouldn't require method signatures #22815 Abstract classes that implement interfaces shouldn't require method signatures
• Casting to Partial<x> seems to be errorful #22806 Partial<T> should be a weak type
• parameter initializer is bound to the wrong variable #22769 Allow unshadowed parameter initializer references in ES6+
• Typing number of children in TSX #22759 Let JSX children be subject to tupleness
• Bitwise enum initializer: compile-time constant but not allowed as a type reference? #22709 Allow references to literal-ish-initialized enums in type positions
• Proposal: Overload Function.bind for when no argArray is provided. #22669 Add a new overload to Function#bind
• Proposal: Easier Migration with Loose Mode in TypeScript Files #22665 Add a mode to make TS act like JS
• Proposal: allow type arguments of generic being named #22631 Allow named type parameter arguments
• tsconfig cannot be specified along with an explicit list of files to compile - TS5042 #22649 Combine tsconfig with file list
• Inferred type symbol in conditional type should be visible-but-banned in falsy arm #22610 Keep infer variable in scope in false arm of conditional type
• Proposal: better type narrowing in overloaded functions through overload set pruning #22609 Do some type guards based on overload when a good discriminant is present
• Provide a back-compat mechanism for bundled .d.ts files in package.json #22605 Provide a back-compat mechanism for bundled .d.ts files in package.json
• typeof union.membername errors. #22598 Allow speculative union type property access in typeof (expression) operands
• Mapped types enumerating keys in string behave poorly #22509 Something something mapped types
• Any number can be assigned to enum member type #22311 Error on out-of-scope literal assigned to enum
• Allow bitwise test on undefined #22230 Allow implicit undefined->0 coercion under strictNullChecks for bitwise ops
• foo.ts is resolved before foo.d.ts even if the latter is in files[] #22208 Prefer .d.ts over .ts if it's in files array

[RyanCavanaugh]

• Support a @nonnull/@nonnullable JSDoc assertion comment #23405 Support non-null assertion operator in JSDoc
  • Report from Webpack checkJs conversion
  • Needed for Array#pop in strictNullChecks
  • Some lurking issues for SNC+checkJs, e.g. function fn(x = null) { gives a bad type due to lack of widening
  • 👍 for (/* @nonnull */ expr)
• Consider collapsing multiple repeated type-related error messages when in --pretty #23393 Collapse multiple error reports in --pretty
  • Certain categories of errors we know will show up many many many times
  • But other categories of error should not be grouped
  • No examples of ideal output?
  • The same error in multiple files may have different meanings
  • Are we talking about what the diagnostics API does, or what --pretty does?
  • What about just stopping at e.g. the first 3 "Cannot find name x" errors per file?
  • Related to multi-location diagnostics issues which lists potentially-collapsible errors
  • Smoosh
  • We should do M-LDs first
    • Does this imply a multi-multi-location diagnostic?
  • We should put effort into our diagnostics like Rust does
  • Revisit at that point
• Warn on unused property in object literal with spread #23340 Warn for known-overwritten properties in spreaded object literal
  • Seems tempting?
  • 👍
• Uninitialized variables work around strictNullChecks (follow-up to #13884) #23305 Enforce never-initialized varibles / strictVariableInitialization
  • There is a lint rule for this
  • Error doesn't apply for global variables or for wrong-time assignments
  • Weird that it's not enforced?
  • We have a walk for checking whether parameters are assigned to
  • Sameish as "unused" checks already present
  • No flag for this - breaking change
  • 👍 reuse existing "unused" logic
• Feature Request / Proposal: constructor type guard #23274 Type guards by expr.constructor (sameish as instanceof)
  • New kind of type guard, not a refinement based on definitions
  • Behaves like instanceof but only does anything in the true branch
  • What about primitives?
    • No narrowing of primitives
  • 👍
• Inconsistent type compatibility for a type with a call signature and and index signature #23226 isObjectTypeWithInferableIndex returns false for intersection type with index signature
  • Current rule is that T is assignable to S & U if T is assignable to S and T is assignable to U
  • Do we want these lines to be OK or an error?
    • OK
  • Changing intersection type behavior just makes both an error, which is not what we want
  • Why do we exclude objects with call/construct signatures from inferring an index signature?
    • Otherwise function expressions would be StringMaps
  • Kinda want a rule that says call signatures are ok if you have at least one declared property
  • Is this an "I noticed" or a bug? What's the unreduced repro look like?
• Support locally scoped type alias nodes #23188 Locally-scoped type aliases
  • Discussion of how to read this syntax
  • Sidebar complaining about let expressions
  • People (who?) want a way to introduce aliases in large declarations
  • Can these all be eagerly expanded (to make this easy)?
    • No, because it defeats typeToString
  • 👎
• Comparing constrained generic types/substitution types to conditional types #23132 Improve behavior of conditional types instantiated with constrained generics
  • Ryan: "This is simple" (Narrator: it wasn't.)
  • This is only for the true case because the false case is not provable due to intersection types
  • Who writes this?
    • Overloads with collapsed parameters using conditional types
    • See example
  • "I'm not seeing why this is that useful" "It isn't"
  • Cautioning: Don't abuse conditional types to write what are really distinct overloads
    • Conditional types should really only appear in the return type position
  • AMF; concrete examples please

Out of time

• T[K] should be never when T is never #23005 T[K] should be never when K is never
  *
• Using string enum as key in {[key]: Type} throws not an index signature #22892 string enum object should be indexable type
  *
• esModuleInterop should work even when compiling to esnext modules #22851 esModuleInterop should work when target = esnext
  *
• Abstract classes that implement interfaces shouldn't require method signatures #22815 Abstract classes that implement interfaces shouldn't require method signatures
  *
• Casting to Partial<x> seems to be errorful #22806 Partial<T> should be a weak type
  *
• parameter initializer is bound to the wrong variable #22769 Allow unshadowed parameter initializer references in ES6+
  *
• Typing number of children in TSX #22759 Let JSX children be subject to tupleness
  *
• Bitwise enum initializer: compile-time constant but not allowed as a type reference? #22709 Allow references to literal-ish-initialized enums in type positions
  *
• Proposal: Overload Function.bind for when no argArray is provided. #22669 Add a new overload to Function#bind
  *
• Proposal: Easier Migration with Loose Mode in TypeScript Files #22665 Add a mode to make TS act like JS
  *
• Proposal: allow type arguments of generic being named #22631 Allow named type parameter arguments
  *
• tsconfig cannot be specified along with an explicit list of files to compile - TS5042 #22649 Combine tsconfig with file list
  *
• Inferred type symbol in conditional type should be visible-but-banned in falsy arm #22610 Keep infer variable in scope in false arm of conditional type
  *
• Proposal: better type narrowing in overloaded functions through overload set pruning #22609 Do some type guards based on overload when a good discriminant is present
  *
• Provide a back-compat mechanism for bundled .d.ts files in package.json #22605 Provide a back-compat mechanism for bundled .d.ts files in package.json
  *
• typeof union.membername errors. #22598 Allow speculative union type property access in typeof (expression) operands
  *
• Mapped types enumerating keys in string behave poorly #22509 Something something mapped types
  *
• Any number can be assigned to enum member type #22311 Error on out-of-scope literal assigned to enum
  *
• Allow bitwise test on undefined #22230 Allow implicit undefined->0 coercion under strictNullChecks for bitwise ops
  *
• foo.ts is resolved before foo.d.ts even if the latter is in files[] #22208 Prefer .d.ts over .ts if it's in files array
  *