Imported From: https://issues.scala-lang.org/browse/SI-9003?orig=1
Reporter: @DarkDimius
Affected Versions: 2.11.4
See #5158, #6070

@retronym said (edited on Nov 25, 2014 1:39:57 PM UTC):
Could you please provide self contained, minimal examples? While it isn't hard to fill in the blanks, the onus should be on the reporter.

See val mutableBinders = in MatchTranslation.scala for the spot where we force creation of temp vals for patterns bound to var fields of case classes. I guess this needs to be extended for name-based patmat.

Relates to #5158, #6070

@DarkDimius said:
While minimizing example(that is actually part of scala/scala#4136) down to

package scala.tools.nsc.backend.jvm

object test {
  object TestInt {
    type Tree = Product2[Any, Int]

    object Block {
      def _1: Any = field._1
      def _2: Int = field._2
      var field: Tree = null
      def get: this.type = this
      def isEmpty: Boolean = field == null
      def isDefined = !isEmpty
      def unapply(s: Tree): this.type ={
        field = s
        this
      }
    }
  }

  def genBlock(tree: Any): Unit = tree match{
    case TestInt.Block(a, b) =>
      genBlock(a)
      print(b)
    case _ =>
  }

  def main(args: Array[String]): Unit = {
    genBlock(((1,2), 3))
  }

}

if this code is compiled with ant build it will output 23, which is correct.
But if compiled with
ant -Dscalac.args.optimise=-optimise -Dextra.repo.url=http://private-repo.typesafe.com/typesafe/scala-pr-validation-snapshots/ -Dstarr.version=2.11.5-2d0104d-SNAPSHOT -Dlocker.skip=1 -Dstarr.use.released=1 it will output 33. Which is incorrect.

@gkossakowski said:
Have you considered using scala bisector https://gist.github.com/retronym/9160186?

@retronym said:
Without -optimize, the pattern matcher always create temporaries for pattern binders, other than ones it identifies as unstable.

Here's a minimal repro:

object NamedBasedPatMat {
  var i = 0
  def isEmpty = false
  def get = i
  def unapply(a: NamedBasedPatMat.type) = this
}

object Test {
  def main(args: Array[String]): Unit = {
    NamedBasedPatMat match {
      case NamedBasedPatMat(i) =>
        NamedBasedPatMat.i = 1
        assert(i == 0) // fails under -optimize
    }
  }
}

@retronym said:
Actually, we've got the same bug in regular sequence extractors:

scala-hash v2.11.4 -optimize
Welcome to Scala version 2.11.4-20141023-110636-d783face36 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_25).
Type in expressions to have them evaluated.
Type :help for more information.

scala> val b = Buffer(1, 2, 3)
b: scala.collection.mutable.Buffer[Int] = ArrayBuffer(1, 2, 3)

scala> b match { case Buffer(1, i, 3) => b(1) = 42; i }
res0: Int = 42

@retronym said:
scala/scala#4164

@DarkDimius said:
Thanks for fast reaction Jason!
Do I get it right that in order to get my PR compiled with fixed compiler I need to rebase?

@gkossakowski said:
Hi Dmitry,

We're building everything using stable version of the compiler. At the moment it's set to 2.11.2: https://github.com/scala/scala/blob/2.11.x/versions.properties#L7

I'll update it to point at 2.11.4.

@adriaanm said:
For your development, you could set starr.version to 2.11.5-SNAPSHOT until 2.11.5 is out.