Repeated calls to Map#apply results in crash

Here is a minification. In def wtf() I create an immutable map and then call .apply 4 times. The first 3 times work but the 4th results in a NoSuchElementException exception.

import utest._

final case class X(a: X.Y, b: X.Y)

object X {

  final case class Y(value: Int)

  def fromInts(i: Int, j: Int): X =
    X(Y(i), Y(j))

  new scala.collection.mutable.HashMap[Int, Int]()
}

object Bug extends TestSuite {

  def v(i: Int, j: Int) = X.fromInts(i, j)

  val v10 = v(1, 0)
  val v11 = v(1, 1)
  val v12 = v(1, 2)
  val v20 = v(2, 0)
  val v21 = v(2, 1)

  def wtf() = {
    val a = List(v10, v11, v12, v20, v21).iterator.zipWithIndex.toMap

    println(a(v10)) // Prints 0
    println(a(v10)) // Prints 0
    println(a(v10)) // Prints 0
    println(a(v10)) // Throws java.util.NoSuchElementException: key not found: X(Y(1),Y(0))
  }

  override def tests = Tests {
    "wtf" - wtf()
  }
}

There are some bizarre things to notice about this minification:

• removing the new scala.collection.mutable.HashMap[Int, Int]() line avoids the bug
• similarly, prefixing it with lazy val x = avoids the bug
• changing def v(i: Int, j: Int) = X.fromInts(i, j) to use X(X.Y(i), X.Y(j)) directly, avoids the bug
• moving the val v10 lines from object Bug to inside wtf() avoids the bug
• removing the X.b field so that there’s only one field avoids the bug
• unboxing the X fields to just be two Ints avoids the bug
• creating a map of less than 5 items avoids the bug (makes sense cos they use different implementations)
• moving everything out of Bug into a different object Z, then changing the test line to "wtf" - Z.wtf() avoids the bug

[japgolly]

Btw, if anyone else comes across this issue in future, the higher-level problem I had was that I was using airbnb-js-shims in my tests because some other dependency was using Map and Set.

A workaround for this is to switch from airbnb-js-shims to core-js. In my case I simply imported 'core-js/features/map' and 'core-js/features/set' and problem avoided.

[sjrd]

Thanks for the repro. Due to the giant SIP meeting, I won’t have time to look at it until next week.

Just from the set of things needed to reproduce, my hunch is that the shim for imul implemented in es6-shims is either fundamentally buggy, or somehow triggers a bug in PhantomJS. The shim that is built in Scala.js seems to be correct, since not having es6-shims causes the bug not to surface, following your comment.