Imported From: https://issues.scala-lang.org/browse/SI-7515?orig=1
Reporter: Erik Bruchez (ebruchez)
Affected Versions: 2.10.0, 2.10.1

@retronym said:
Another workaround:

scala> for ((k, v) <- Map("a" -> "b"); (a, b) = (k.length, v.length)) yield v -> k
res1: scala.collection.immutable.Map[String,String] = Map(b -> a)

not fixed in Scala 3: scala/scala3#16812

When you consider that intermediate variables in for-comprehensions desugar to map calls with a tuple result (I don't see that changing in Scala 2), then it makes more sense. We can see the desugared code with reify:

import scala.reflect.runtime.universe._
println(reify(for ((k, v) <- Map("a" -> "b"); a = k.length; b = v.length) yield v -> k))

Expr[scala.collection.immutable.Iterable[(String, String)]](Predef.Map.apply(Predef.ArrowAssoc("a").$minus$greater("b")).withFilter(((check$ifrefutable$1) => check$ifrefutable$1: @unchecked match {
  case Tuple2((k @ _), (v @ _)) => true
  case _ => false
})).map(((x$2) => x$2: @unchecked match {
  case (x$1 @ Tuple2((k @ _), (v @ _))) => {
    val a = k.length();
    val b = v.length();
    Tuple3.apply(x$1, a, b)
  }
})).map(((x$3) => x$3: @unchecked match {
  case Tuple3(Tuple2((k @ _), (v @ _)), (a @ _), (b @ _)) => Predef.ArrowAssoc(v).$minus$greater(k)
})))

Of course the intermediate Tuple3 can't preserve the Map type. But even if it could I would argue that this is probably not what one intended. Map and Set for-comprehensions are tricky because preserving the collection type applies not only to the final result, but also intermediate results. But this will clump together for example in this case keys with the same length and you get a different number of elements out. Think of it like an hour glass where the sand can only go through the tight spot.