Self-type annotations make abstract type members refinements be silently ignored

[scabug]

trait Base {
  trait BaseInner {
    val base = 0
  }
  type Inner <: BaseInner
}

trait T1 extends Base {
  trait Inner1 extends BaseInner {
    val t1 = 0
  }
  type Inner <: Inner1
  def f1(i: Inner) { i.base + i.t1 }
}

trait T2 extends Base { self: T1 =>
  trait Inner2 extends Inner1 {
    val t2 = 0
  }
  type Inner <: Inner2
  def f2(i: Inner) { i.base + i.t1 + i.t2 }
}

yields

$$ scala I.scala
/home/mikhail.vorozhtsov/I.scala:22: error: value t2 is not a member of T2.this.Inner
  def f2(i: Inner) { i.base + i.t1 + i.t2 }
                                       ^
one error found

Strangely type Inner = Inner2 works, but then changing type Inner <: Inner1 to type Inner = Inner1 breaks it again.

Tested on 2.8.1 and 2.9.0.r24471-b20110317020039

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-4349?orig=1
Reporter: Mikhail Vorozhtsov (mikhail.vorozhtsov)

[scabug]

@ingoem said:
This has to do with linearization order. The self type implicitly becomes T2 with T1 in which case the abstract type declaration type Inner <: Inner1 takes precedence. A workaround for now is to force the right order by writing self: T1 with T2 =>.