Pattern matching introduces unbounded existential type, losing type information

[scabug]

I'm confused by the following compile-time error:

private def safeMax[A](xs: List[A])(implicit ev: Ordering[A]): Option[A] = 
  xs match {
    case ys@(_ :: _) => Some(ys.min)
    case Nil               => None
  }


// Exiting paste mode, now interpreting.

<console>:12: error: No implicit Ordering defined for ?A1.
           case ys@(_ :: _) => Some(ys.min)
                                       ^

I'm not sure if it's a bug, but it seemed suspect to me.

EDIT - I first posted this question on StackOverflow - http://stackoverflow.com/questions/38775071/no-implicit-ordering-defined-on-pattern-matched-list.

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-9879?orig=1
Reporter: Kevin Meredith (kevinmeredith)
Affected Versions: 2.11.8
See #1786, #6169, #127, #5195

[scabug]

@SethTisue said (edited on Aug 9, 2016 5:45:17 AM UTC):
Note that in general we prefer that an "I'm not sure if it's a bug" type report begin life as a mailing list thread or SO question rather than a JIRA ticket (since the number of tickets in the system is already overwhelming). (thanks for adding SO link)

Your code compiles and runs just fine in Scala 2.9. So this seems to be a regression in the new pattern matcher.

In the error message, the question mark in ?A1 indicates that the inferred type of ys is an existential type rather than simply List[A] as you were expecting.

:: is a case class, so this is a constructor pattern. SLS 8.1.6 (Constructor Patterns) says "If the case class is polymorphic, then its type parameters are instantiated so that the instantiation of c conforms to the expected type of the pattern". As far as I can tell, this doesn't fully specify the behavior; there might be multiple types which "conform" and therefore satisfy the spec language. So the door seems to be open for the compiler to introduce an existential type.

But it seems the existential doesn't even have a bound; at least, I would expect compiling with -Xprint:typer to show a bound if there was one, and it doesn't. The absence of a bound seems like a violation of 8.1.6 to me, since without the bound, there's no guarantee that List[?A1] conforms to List[A].

And the missing bound causes the error; I conclude that because e.g. this compiles if and only if the bound is included:

scala> def foo[A : Ordering](xs: List[A]) = { val ys: List[A1] forSome { type A1 <: A } = xs; ys.min }

There are a number of existing tickets involving various combinations of pattern matching, existential, and bounds. Some of these tickets (e.g. #1786, #6169) have complicated histories. So, leaving this for Adriaan and/or Jason to decide whether this a duplicate or not.

[scabug]

@SethTisue said:
As an aside, I'm baffled that it seems to matter here whether we use match or a val pattern:

scala> def foo[A : Ordering](xs: List[A]) = { val ys @ (_ :: _) = xs; ys.min }
foo: [A](xs: List[A])(implicit evidence$1: Ordering[A])A

scala> def foo[A : Ordering](xs: List[A]) = xs match { case ys @ (_ :: _) => ys.min }
<console>:11: error: No implicit Ordering defined for ?A1.
       def foo[A : Ordering](xs: List[A]) = xs match { case ys @ (_ :: _) => ys.min }
                                                                                ^

Perhaps it's a clue.

[scabug]

@SethTisue said:
at least vaguely related, hard to tell if underlying cause is same: #127 and #5195