regression: assertion error in Mixin.transformInfo, with $$anon script wrapper

[scabug]

=== What steps will reproduce the problem ===

def x[N[_]] = ()

trait ExistT[M[_]] {
  x[M]
}

=== What is the expected behavior? ===

Compiles okay, as per 2.8.1

=== What do you see instead? ===

java.lang.AssertionError: assertion failed
	at scala.Predef$$.assert(Predef.scala:85)
	at scala.tools.nsc.transform.Mixin.transformInfo(Mixin.scala:375)
	at scala.tools.nsc.transform.InfoTransform$$Phase$$$$anon$$1.transform(InfoTransform.scala:38)
	at scala.tools.nsc.symtab.Symbols$$Symbol.rawInfo(Symbols.scala:774)
	at scala.tools.nsc.symtab.Symbols$$Symbol.info(Symbols.scala:717)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer.scala$$tools$$nsc$$transform$$Mixin$$MixinTransformer$$$$postTransform(Mixin.scala:1113)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer$$$$anonfun$$23.apply(Mixin.scala:1213)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer$$$$anonfun$$23.apply(Mixin.scala:1213)
	at scala.tools.nsc.symtab.SymbolTable.atPhase(SymbolTable.scala:96)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer.transform(Mixin.scala:1213)
	at scala.tools.nsc.ast.Trees$$Transformer.transformTemplate(Trees.scala:860)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transform$$2.apply(Trees.scala:752)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transform$$2.apply(Trees.scala:751)
	at scala.tools.nsc.ast.Trees$$Transformer.atOwner(Trees.scala:884)
	at scala.tools.nsc.ast.Trees$$Transformer.transform(Trees.scala:750)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer.transform(Mixin.scala:1212)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transformStats$$1.apply(Trees.scala:876)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transformStats$$1.apply(Trees.scala:874)
	at scala.collection.immutable.List.loop$$1(List.scala:117)
	at scala.collection.immutable.List.mapConserve(List.scala:133)
	at scala.tools.nsc.ast.Trees$$Transformer.transformStats(Trees.scala:874)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transform$$1.apply(Trees.scala:746)
	at scala.tools.nsc.ast.Trees$$Transformer$$$$anonfun$$transform$$1.apply(Trees.scala:746)
	at scala.tools.nsc.ast.Trees$$Transformer.atOwner(Trees.scala:884)
	at scala.tools.nsc.ast.Trees$$Transformer.transform(Trees.scala:745)
	at scala.tools.nsc.transform.Mixin$$MixinTransformer.transform(Mixin.scala:1212)
	at scala.tools.nsc.ast.Trees$$Transformer.transformUnit(Trees.scala:877)
	at scala.tools.nsc.transform.Transform$$Phase.apply(Transform.scala:30)

=== Additional information ===

No crash after s/trait/class/.

I was actually trying to reproduce the following spurious compile error from the scalaz branch for Scala 2.9.x, from this tree: https://github.com/scalaz/scalaz/tree/3a58739500416eeda09d5150a2f3f7d1db2ad86e

I plan to isolate it further and lodge a separate ticket.

[error] /home/scalatools/hudson/.hudson/jobs/scalaz/workspace/http/src/main/scala/scalaz/http/response/Response.scala:50: kinds of the type arguments (OUT) do not conform to the expected kinds of the type parameters (type OUT).
[error] OUT's type parameters do not match type OUT's expected parameters: <none>'s bounds <notype> are stricter than type _'s declared bounds >: Nothing <: Any
[error]   def apply(l: StatusLine) = response[OUT](l, headers, body)
[error]                                      ^
[error] /home/scalatools/hudson/.hudson/jobs/scalaz/workspace/http/src/main/scala/scalaz/http/request/Request.scala:36: kinds of the type arguments (IN) do not conform to the expected kinds of the type parameters (type IN).
[error] IN's type parameters do not match type IN's expected parameters: <none>'s bounds <notype> are stricter than type _'s declared bounds >: Nothing <: Any
[error]   def apply(h: RequestHeader, v: NonEmptyList[Char]) = request[IN](line, (h, v) :: headers, body)

=== What versions of the following are you using? ===

• Scala: 2.9.0.r24073-b20110124174250

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-4202?orig=1
Reporter: @retronym

[scabug]

@paulp said:
Oddly this compiles (and runs) fine for me with versions of scala both before and after yours. Are you giving it a command line option or something?

Welcome to Scala version 2.9.0.r24136-b20110128174513 (Java HotSpot(TM) 64-Bit Server VM, Java 1.6.0_22).

scala> object o {
     |   def x[N[_]] = ()
     | 
     |   trait ExistT[M[_]] {
     |     x[M]
     |   }
     | }
defined module o

scala> o.x

Welcome to Scala version 2.9.0.r24068-b20110123233401 (Java HotSpot(TM) 64-Bit Server VM, Java 1.6.0_22).

scala> object o {
     |   def x[N[_]] = ()
     | 
     |   trait ExistT[M[_]] {
     |     x[M]
     |   }
     | }
defined module o

scala> o.x

[scabug]

@retronym said:
Oh darn. I was running it as a script. Working that back to something I can demonstrate in vanilla scalac:

object Main {
  def x[N[_]] = ()

  def main = {
    class $$anon { // rename to 'anon' to get rid of crash
      object scratch {

        trait ExistT[M[_]] { // s/trait/class/ to get rid of crash
          x[M]
        }
      }
    }
  }
}

Gives:

 ~/code/scratch: scalac -version
Scala compiler version 2.8.1.final -- Copyright 2002-2010, LAMP/EPFL
 ~/code/scratch: scalac /Users/jason/code/scalaz/core/scratch.scala ~/code/scratch: ~/usr/scala-2.9.0.r24145-b20110130020134/bin/scalac   /Users/jason/code/scalaz/core/scratch.scala
Exception in thread "main" java.lang.AssertionError: assertion failed
	at scala.Predef$$.assert(Predef.scala:85)
	at scala.tools.nsc.transform.Mixin.transformInfo(Mixin.scala:375)

[scabug]

@adriaanm said:

object Main {
  def x = 1
  class $$anon { // rename to 'anon' to get rid of crash
    object scratch {
      trait ExistT { // s/trait/class/ to get rid of crash
        x
      }
    }
  }
}

look ma, no type constructors!

[scabug]

@retronym said:
Looks like I need to work on my minimization skills. 'tcpoly' tag duly removed.

For the record, my use of '$$' was just to replicate the inner machinations of the script runner. The following crashes, when run as a script.

script.scala: def x{};trait X{x}

[scabug]

@retronym said:
Here's another example, distilled from Specs:

object t4202 {
  () => {
    trait T {
      def t = ()
    }
    object T2 extends T {
      t
    }
  }
}

[scabug]

@paulp said:
Replying to [comment:7 retronym]:

Here's another example, distilled from Specs:

That looks an awful lot like #2897, which is:

 def f1(t: String) = {
    trait T {
      def xs = Nil map (_ => t)
    }
    ()
  }

Not exactly the same bug I don't think, just brothers.

[scabug]

@paulp said:
Actually they are the same bug I think. By the way "object T2" in your comment doesn't have any bearing on the bug, it crashes fine without it.

Putting r24206 to work, here it is after lazyvals:

// I'm using: () => { trait T { def t = () } }
  object t4202 final
      constructor t4202 
      value <local t4202> 
          anonymous class $$anonfun final <synthetic>
              constructor $$anonfun 
              method apply final
              method apply final <bridge>
              value <local $$anonfun> 

And then here after lambdalift. Notice the trait implementation class is being renamed to "T$$class$$1" which is pretty clearly wrong. I'm able to push the error downfield by excluding .isImplClass but the problem is deeper.

  object t4202 final
      anonymous class $$anonfun$$1 final <synthetic>
          constructor $$anonfun$$1 
          method apply final
          method apply final <bridge>
*         trait T$$1 abstract
*             method t 
*             value <local T> 
*             value t4202$$$$anonfun$$T$$$$$$outer <synthetic> <stable>
*         trait T$$class$$1 abstract
*             method $$init$$ 
*             method t 
*             value <local T$$class> 
          value <local $$anonfun> 

[scabug]

@paulp said:
Well, I thought it was clearly wrong based on the fact that 1406/1406 implementation classes in trunk have a name ending in "$$class". But 2.8.1 does compile it with that name, so maybe it's not wrong.

-rw-r--r--  1 paulp  staff    309 Feb  6 01:08 t4202$$$$anonfun$$1$$T$$1.class
-rw-r--r--  1 paulp  staff    442 Feb  6 01:08 t4202$$$$anonfun$$1$$T$$class$$1.class
-rw-r--r--  1 paulp  staff    754 Feb  6 01:08 t4202$$$$anonfun$$1.class
-rw-r--r--  1 paulp  staff    407 Feb  6 01:08 t4202$$.class
-rw-r--r--  1 paulp  staff    435 Feb  6 01:08 t4202.class

Drives me nuts trying to infer everything from the implementation. Is this documented somewhere?

[scabug]

@adriaanm said:
as I understand it, the fix for this is simple: don't give interpreter wrappers blingy names (might I suggest replacing $$ by CHF?)

we don't officially deal in dollars anyway (at least not in user-supplied names)

Paul, I'm sorry I can't provide more insight on the use of the dollar in generated names, but I think we can settle for a pragmatic solution here. Could the deeper problem simply be that calling this class $$anon triggers a conversion that's not meant to be done in here?

    val ANON_CLASS_NAME: NameType    = "$$anon"
[...]
    final def isAnonymousClass    = isClass && (name containsName tpnme.ANON_CLASS_NAME)

[scabug]

@retronym said:
As described in the comments above, the same assertion trips with the following, but maybe that's a different bug.

object t4202 {
  () => {
    trait T {
      def t = ()
    } 
  }
}

[scabug]

@adriaanm said:
Hubert, wild guess: could this be related to r24225?

In any case, it does not seem to be a REPL issue, based on Jason's latest comment.

[scabug]

@hubertp said:
Well it is definitely not caused by it, given that my commit is from 6 days ago but I will try to have a look at it.

[scabug]

@hubertp said:
(In r24326) Closes #4202 and removes empty TreeHashMap (superseded by HashMap implementation). No review

[scabug]

@paulp said:
This still crashes as of r24432.

object t4202 {
  () => {
    trait T {
      def t = ()
    }
    object T2 extends T {
      t
    }
  }
}

[scabug]

@hubertp said:
(In r24527) Closes #4202 again, closes #4363. if someone can enlighten me why we actually had that exclusion for companion objects in the first place I would be grateful. review by odersky

[scabug]

@paulp said:
Replying to [comment:21 plocinic]:

(In r24527) Closes #4202 again, closes #4363. if someone can enlighten me why we actually had that exclusion for companion objects in the first place I would be grateful.

r8575
"fixed gilles problem with spurious recursive types"

For whatever you may be able to make of that.

[scabug]

@hubertp said:
Replying to [comment:22 extempore]:

r8575
"fixed gilles problem with spurious recursive types"

For whatever you may be able to make of that.

Yeah, I used my svn skills to find that too, though the actual comment isn't very informative. Also I assumed that it might be no longer valid given that it was 6 years ago ;)