Mandatory eta expansion of higher-kinded types

Language Design
1

Are there any plans to stop requiring eta expansion of higher-kinded type bounds?

trait A
trait B[T <: A]

// Error: type B takes type parameters
def foo[C <: B]: C[A] = ???

// No error, but lots of boilerplate
def bar[C <: [T <: A] =>> B[T]]: C[A] = ???
2

This sort of thing used to be supported but was explicitly disallowed before 3.0: Disallow unapplied types in aliases and bounds by odersky · Pull Request #9090 · lampepfl/dotty · GitHub

For the record, it’s possible to write this slightly more concisely:

def bar[C[T <: A] <: B[T]]: C[A] = ???
2 Likes
3

Thanks! I would never have thought of adding a seemingly unbound T like that.

Is there anywhere in the documentation where I can add such an example?

1 Like
4

Right now this is documented both in the spec, section 4.3.1 and in the Scala 3 “Type Lambdas - More Details” page. It seems we’re missing a high-level doc page dedicated to higher-kinded types where this sort of things could also be mentioned, so I’m not sure there’s any good place to add such an example right now.

1 Like
5

I cannot actually see it documented in either of the links you shared. I’m not implying your links are bad; I’m probably just not looking properly, but below I describe what I see.

The page from the first link says

Scala has several other advanced types that are not shown in this book, including:
…

  • Higher-kinded types

, which seems to imply it is not documented in the book. I also tried to follow the link at the bottom of that page, to the Scala 3 Reference (Is that the spec?), but I don’t see any section numbers on that page, and higher-kinded types are only mentioned once, without any actual documentation that I can see.

Concerning the second link, the part most closely resembling documentation I could find is

A partially applied type constructor such as List is assumed to be equivalent to its eta expansion. I.e, List = [X] =>> List[X].

, which seems to almost imply the opposite, namely that there should be no distinction between a higher-kinded type and its eta expansion.

Sorry if I’m being dense! In that case, feel free to ignore my message.

6

The spec link was wrong. It was meant to point to
https://www.scala-lang.org/files/archive/spec/3.4/04-basic-definitions.html#desugaring-of-parameterized-type-definitions

2 Likes
7

And I see now that I did in fact not look properly at the second link. I just found this part:

The same expansions apply to type parameters. For instance,

[F[X] <: Coll[X]]

is treated as a shorthand for

[F >: Nothing <: [X] =>> Coll[X]]