Imported From: https://issues.scala-lang.org/browse/SI-6541?orig=1
Reporter: @lrytz
Affected Versions: 2.11.0-M1

@szeiger said:
To clarify, the correct type for the extractor should be

  Option[(A, B[_$1 forSome { type _$1 }])]

aka

  Option[(A, B[_])]

right?

@Blaisorblade said:
I think that Option[(A, B[])] is right, but that's defined by the SLS (3.2.10) to be equivalent to Option[(A, B[X] forSome {type X})], although that was counterintuitive also for me.
That makes a different when B is contravariant, since Option[(A, B[X] forSome {type X})] is then equivalent to Option[(A, B[Nothing])] while Option[(A, B[$1 forSome { type _$1 }])] is equivalent to Option[(A, B[Any])] (if I get simplificaton rule 4 in SLS 3.2.10 right).

@adriaanm said:
not a regression, so not up for fixing this during the RC cycle
affects binary compatibility, so can't do it in 2.10.x
ergo, 2.11.x

@Blaisorblade said:
To clarify yet again (since I'm confused), the compiler should in fact infer: Option[(A, B[X] forSome {type X})], that is, Option[(A, B[_])].

It seems that in the body of the method, _$1 should be bound by matching on B's type: x$0.b match {case b: B[t] => Some((x$0.a, b))}. But maybe _$1 is in fact bound in some equivalent way, which can only be seen in the AST (trying to take a look now). Alternatively, the AST is just inconsistent and this is not detected, but that doesn't sound likely.

But I don't get what the impact of the bug is from the bug report. Looking at the thread, I see that:

1. We get a warning on generated code.
2. The generated type is wrong, and that comes from type inference, so it's somewhat confusing - is there a type inference bug?
3. The generated code might be wrong, and maybe the error is swallowed later by erasure.
   We don't know if the generated code works, but I'd guess so.

@retronym said:
Here's a sketch of how we could address this. It's surprising tricky to get at all the information do to this during unapply synthesis (at least, it seems that way to me.)

https://github.com/retronym/scala/compare/ticket/6541

@paulp said:
I hadn't realized how easy it is to hit this:

scala> case class Foo(clazz: Class[_])
<console>:7: warning: inferred existential type Option[Class[_$1]] forSome { type _$1 }, which cannot be expressed by wildcards,  should be enabled
by making the implicit value scala.language.existentials visible.
This can be achieved by adding the import clause 'import scala.language.existentials'
or by setting the compiler option -language:existentials.
See the Scala docs for value scala.language.existentials for a discussion
why the feature should be explicitly enabled.
       case class Foo(clazz: Class[_])
                  ^
defined class Foo

@gkossakowski said:
Unassigning and rescheduling to M7 as previous deadline was missed.

@retronym said:
Batter up! scala/scala#4017

Eyal Roth (errr) said (edited on Nov 1, 2016 2:08:39 PM UTC):
Scala 2.11.8 yields the inferred existential type warning yet again because of this.

@SethTisue said:
I didn't look too deeply, but I think this is fixed in 2.12.0? Paul's example doesn't warn, nor does Lukas's original code.

@lrytz said:
yeah, forgot to close this: scala/scala#4017

@lrytz said:
oh wait, i see now it was reopened recently?

@lrytz said:
i will take a look.

@lrytz said:
The fix works as expected, since 2.11.5. Note that in 2.11, -Xsource:2.12 is required.