Design Meeting Notes, 12/13/2023

[DanielRosenwasser]

Preventing inference to a type parameter (NoInfer)

#14829

#52968

• Old PR adds a new type kind (29th TypeFlag!)
• There is a tail of places where NoInfer needs to be handled/sanitized.
  • Getting apparent types, etc.
• But the work required to do this is identical to what we do with substitution types.
  • So we have a prototype of NoInfer that is built atop substitution types.
• Some get unhappy in some cases when two candidates fail - but just giving people the ability to control this can alleviate the issue.
• You can now declare a type parameter that never is inferred to. You'll need to provide a type argument. Is that understood and considered okay?
  • Yes.
  • Technically it's not the case that you'll need a type argument, you'll just end up with the default type argument.
• Are there places where you as a user would want to disable return type inference?
  • const foo: annotation = getSomeFoo()
  • Hard to remember where these came up.
  • Mostly fixed const { x, y, z } = genericFunc(), but should review those.
• Going to experiment here.
• [[Later addition: Add NoInfer<T> intrinsic represented as special substitution type #56794 ]]

Making Arrays into Tuples When Assigning to Homomorphic Mapped Types

#56555

• We know homomorphic mapped types preserve tuple inputs.
• Preserving tuple-shapeness means you can infer better to the mapped type.
• We think that sounds right!

Comparing Against Apparent Keys of Target Type Source Has Deferred Index Type

#56742

• Scenario is keyof SourceMappedType -> keyof TargetMappedType
• Generally speaking, this reflects the same logic we take around how we construct types, just the reverse as we pick things apart.

Modules!

• Been working on finer-grained module options.
• A few scenarios we have been trying to fix:
  • Accurately modeling bundler behavior (some which work closer to Node, some which always work with ESM)
  • Shipping CJS/ESM
• CJS/ESM
  • Could try to do some shenanigans
  • tshy and bundlers like tsup help with these scenarios.
  • Does it make sense to try to "address" CJS/ESM?
    • Maybe as far as type-checking goes?
    • But there's some subtlety because there is pre-bundler and post-bundler.
    • The inputs may be assumed to all "just work" within bundler, but the outputs will be running in nodenext.
• Declaration emit
  • The files should really reflect what the bundler does.

[fatcerberus]

Heads up: The "later addition" issue link is broken.

[DanielRosenwasser]

Thanks!