No member completions for existential types

[tgodzik]

Reproduction steps

Scala version: 2.12.16 (though with latest kind projector it fails also on earlier versions)

With kind projector plugin https://github.com/typelevel/kind-projector

object a {
 def baz[F[_], A]: F[A] = ???
 baz[Either[Int, ?], String].fold@@
}

Problem

When trying to get the type members of baz[Either[Int, ?], String] are getting empty results from Global.completionsAt(pos) with pos at @@

It seems we get an existential type here:

Select(
  TypeApply(
    Ident(baz),
    List(
      ExistentialTypeTree(
        AppliedTypeTree(Ident(Either), List(Ident(Int), Ident($qmark$1))),
        List(
          TypeDef(Modifiers(2097168L, , List()), $qmark$1, List(), TypeBoundsTree(<empty>, <empty>))
        )
      ),
      Ident(String)
    )
  ),
  fold
)

Is this something that might be easily fixable, or would be quite involved?

[SethTisue]

with latest kind projector it fails also on earlier versions

I'm confused. Something changed recently in kind-projector?

[tgodzik]

I realised we were testing 0.10.3, which up until 2.12.16 was not returning a exitential type. That changed in 2.12.16 somehow and I tried to update to the newest kind-projector, but that turned out to also return the same result.

[SethTisue]

Okay.

My impression is that the issue isn't severe enough to warrant withdrawing 2.12.16? Do you agree?

We already have one regression that warrants building 2.12.17 on a short-ish timetable (2 months, maybe), so we already know that missing the 2.12.16 train doesn't mean an epically long delay.

[smarter]

baz[Either[Int, ?], String].fold

This syntax does not exist anymore in kind-projector, ? was replaced by * (the old syntax was deprecated a long time ago and fully removed one year ago: typelevel/kind-projector#186). This allowed us to repurpose ? by default in Scala 2: scala/scala#9990 / scala/scala#10005, so this change is expected.

[tgodzik]

Well that explains it 😅 Thanks @smarter ! I went a totally different way delving into the presentation compiler 😅

My impression is that the issue isn't severe enough to warrant withdrawing 2.12.16? Do you agree?

I think this was a false alarm, sorry about that. It seems to work properly.