Tuple.Head being unable to work with NonEmptyTuple in match types.

[Decel]

Compiler version

3.3.1-RC1

Minimized code

type second[X <: Tuple2[Any, Any]] = Tuple.Head[Tuple.Tail[X]] // error
type middle[X <: Tuple3[Any, Any, Any]] = Tuple.Head[Tuple.Tail[X]] // error

val a: Tuple.Head[Tuple.Tail[Tuple2[Int, String]]] = ??? // works
val b: Tuple.Head[Tuple.Tail[Tuple3[Int, String, Boolean]]] = ??? // works

Output

-- [E057] Type Mismatch Error: ---------------------------------------------------
1 |type second[X <: Tuple2[Any, Any]] = Tuple.Head[Tuple.Tail[X]]
  |                                                ^
  | Type argument Tuple.Tail[X] does not conform to upper bound NonEmptyTuple
  |
  | Note: a match type could not be fully reduced:
  |
  |   trying to reduce  Tuple.Tail[X]
  |   failed since selector X
  |   does not match  case _ *: xs => xs
  |   and cannot be shown to be disjoint from it either.
  |
  | longer explanation available when compiling with `-explain`

Expectation

Should compile.

[odersky]

I think this should not compile, but the error message should be better.

To resolve this match type, we need to widen X to Tuple2[Any, Any] that does match _ *: xs but does not give us a safe instantiation of xs. It's really the same as #17180.

The error message is misleading, however. It should say that the instantiation is not unique.

[Decel]

This kind of behavior can be recreated in a lot of instances:

type RemoveTwo[X <: Tuple3[Any, Any, Any]] = Tuple.Tail[Tuple.Tail[X]] // error

type AltMiddle[X <: Tuple3[Any, Any, Any]] = Tuple.Tail[Tuple.Init[X]] // error

type RemoveTwo[X <: Tuple3[Any, Any, Any]] = Tuple.Head[Tuple.Concat[X, Tuple]] // error

Maybe we shouldn't have NonEmptyTuple as part of the type boundary on any Tuple method and just add additional cases for EmptyTuple?

[Decel]

I think this should not compile, but the error message should be better.

To resolve this match type, we need to widen X to Tuple2[Any, Any] that does match _ *: xs but does not give us a safe instantiation of xs. It's really the same as #17180.

The error message is misleading, however. It should say that the instantiation is not unique.

I'd personally treat this as not being related to match types at all. If anything, this is correct behavior since NonEmptyTuple <: Tuple.

I think Tuple.Head and Tuple.Tail should just not guarantee to be non-empty, -- we can just either throw error or return EmptyTuple.