Imported From: https://issues.scala-lang.org/browse/SI-5900?orig=1
Reporter: @jrudolph

@adriaanm said (edited on Jun 13, 2012 8:53:00 AM UTC):
as Johannes says, this is due to adapt's (5.1)

(5) Convert constructors in a pattern as follows: 
(5.1) If constructor refers to a case class factory, set tree's type to the unique 
      instance of its primary constructor that is a subtype of  the expected type. 
(5.2) If constructor refers to an extractor, convert to application of 
      unapply or unapplySeq method. 

@jrudolph said:
Discussion here:

https://groups.google.com/d/topic/scala-language/zihVwcwJR2Q/discussion

@adriaanm said:
note to self:

The unapply and case class scenarios should behave in the same way.
Technically, I think we should retract "failed" inference results (such as Any) using adjustTypeArgs,
and replace them by skolems, as we do for unapply's.

@adriaanm said:
another workaround is to add a type ascription to the scrutinee (??? : C[_] ) match ...

@adriaanm said (edited on Nov 22, 2013 10:14:00 PM UTC):
a variation:

  scala> case class Transition[S](x: S) ;  object C ; (??? : Any)   match  { case Transition(C) => }
    <console>:11: error: pattern type is incompatible with expected type;
     found   : C.type
     required: S
            (??? : Any) match  { case Transition(C) => }
                                                 ^

@adriaanm said:
Jason bisected it:

scala-hash is missing a build or two around the critical area, but is in:
*   acd7780 Merge pull request #3014 from ceedubs/pr/implicitNotFound-scaladoc
|\
| * 9835d33 Describe type parameter interpolation in @implicitNotFound documentation
* |   90a3126 Merge pull request #3005 from paulp/pr/7886
|\ \
| * | 5708e9d SI-6680 unsoundness in gadt typing.
| * | 95d5554 SI-7886 unsoundness in pattern matcher.

@retronym said:
Here's my take on it (the variation in Adriaan's comment which is a regression).

https://github.com/retronym/scala/compare/ticket/5900?expand=1

@retronym said:
I haven't really figured out the right way to fix this one. My branch works, but I feel like its not the most direct fix.

@adriaan: once you've delivered M7 to the world, lets put our heads together and seek out something better.

@retronym said:
My notes on the difference in typechecking / soundness between case classes and extractors: https://gist.github.com/retronym/7704153

@adriaanm said:
Sorry, I don't know how to fix this in time for 2.11

@retronym said:
I think we can revert just enough of Paul's soundness fix to avoid the regression.

I'd rather not ship a change to pattern inference only to change it again when we fix this for real. I'm marking as 2.11.0 again and will followup tomorrow with a patch.

@adriaanm said:
Great!

@retronym said (edited on Feb 11, 2014 10:35:32 PM UTC):
Still working on this one, will try again for the elusive tomorrow.

@retronym said:
The regression from 2.10 is addressed in scala/scala#3514 by reverting part of #7886.

@adriaanm said:
rebased/tests fixed at: scala/scala#3523

@adriaanm said:
Regression fixed but this ticket is about the real fix, which can't happen in 2.11.x

@milessabin said:
It'd be great to get this updated for 2.12.x and merged.