redo-signals: Breaking cycles

Consider a slider/text-field:

When the user edits one, the other should update to match:

Can we express this in FRP-style?

val textField = new JTextField("50", 6)
val slider = new JSlider

val win = new JFrame("Test")
win.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE)
win.setLayout(new FlowLayout)
win.add(textField)
win.add(slider)
win.pack()
win.setLocationRelativeTo(null)
win.setVisible(true)

textField.text.foreachDelayed { text => implicit u =>
  slider.value.updateLater(text map { text =>
    Try(text.toInt) match {
      case Failure(_) => slider.value.now
      case Success(value) => value
    }
  })
} (slider.value.redoObserving)

slider.value.foreachDelayed { value => implicit u =>
  textField.text.updateLater(value map (_.toString))
} (textField.text.redoObserving)

It turns out we can. But the code looks cyclical... how does it not produce an infinite loop?

foreachDelayed comes to the rescue. It runs on the invalidation sweep. A signal can only be invalidated once before it is recalculated. So, the invalidation sweep can never produce an infinite loop; it simply traverses the graph of signals and invalidates each one it hits.