Compilation error with parameterised case class in a non-root package named 'scala'

[scabug]

Attempting to compile the following file:

package a.b.scala
case class Test[T](t : T)

gives this error:

test.scala:2: error: type Nothing is not a member of package a.b.scala
case class Test[T](t : T)
                ^
one error found

Also occurs if the case class is in package a.b and package a.b.scala is defined.

Tested against rev 13855

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-430?orig=1
Reporter: Andrew Foggin (andyfoggin)

[scabug]

@richdougherty said:
More information:

// Compiles
package a {
  case class A[T]
}

// Compiles
package b.scala {
  class B[T]
}

// Doesn't compile: type Nothing is not a member of c.scala
package c.scala {
  case class C[T]
}

// Doesn't compile: type Nothing is not a member of d.scala 
package d.scala.d {
  case class D[T]
}

// Doesn't compile: type Any is not a member of e.scala 
package e.scala {
  case class E[T >: Nothing]
}

// Compiles
package f.scala {
  case class F[T >: Nothing <: Any]
}

[scabug]

@richdougherty said:
FYI, this bug occurs in 2.7.0 rc1 and rc2. However, there is a workaround: explicitly giving Any and Nothing as type bounds.

[scabug]

Michael (michid) said:
This might be related:

// Doesn't compile: type <byname> is not a member of package h.scala
package h.scala {
  case class H(s: String)(t: =>String)
}

[scabug]

@odersky said:
Milestone next_bugfix deleted

[scabug]

@paulp said:
Due to the less than consistent way package qualifying is handled inside the compiler, it's hard to be sure I got all the manifestations of this, nor to have much confidence another one won't be reintroduced down the road; but I think the current issues are fixed in r17172.

Barring an unlikely change in package resolution semantics I'd strongly recommend against using "scala" (or "java") as a package segment name.