Here is some of the feedback I received on the questions posed and the solutions provided in my Scala’s Alternatives to Exceptions presentation at the BFPG by Kristian Domagala:

It looks like Sanjiv raised some interesting questions at the end of his slides. As I wasn’t at the meeting, I don’t know to what extent the questions were answered, but I thought I would chime in with my 2 cents.

I’ve found the Either type to be useful for error conditions that can be resolved either through code (eg, using a fallback), or by re-directing back to the user to address. Often this means the Left and Right types are mapped to the same type, and the Either type is merged into a single value (ie, Either[A,A] => A). In the examples you present, you’re essentially doing that all together at the end with the fold function.

In terms of chaining exceptions, in Scala terms, this can be handled using flatMap function on the right-projection of Either, assuming you use a common type on the left. Sometimes you will need to map the left types to a common type, but I’ve found it’s usually a String or something representing an exception message, and little additional effort is required.

Going to the particular examples in the slides, I note that the PersistentOutcome type is very similar to Either; ie, Either[String, Unit]. In fact, you could remove the need for pattern matching on the type by aliasing it to Either and re-using the functions defined there:

type PersistentOutcome = Either[String, Unit]

Alternatively, you could achieve the same thing using an implicit def, so that the pattern matching is done in one place and whenever you see a PersistentOutcome, you can treat it as an Either[String, Unit]:

implicit def poToStringUnit(po:PersistentOutcome):Either[String,Unit] = po match {
  case Failure(x) => Left(x)
  case Success => Right(())
}

note that you’ve started to go down the latter path in the reloaded example. The above gives you something you can re-use anywhere. Once you’ve got Either[String,Unit], it’s only one left-map away from becoming Either[Exception,Unit]:

def stringUnitToExUnit(e:Either[String,Unit]):Either[Exception,Unit] = e.left.map(new Exception(_))

Now we’re dealing with more consistent types throughout the execution (Either[Exception,_]), and the addSpends function can be reduced to:

def addSpends(date:Sdate, f:(DailySpend) => Unit):Either[Exception, Unit] =
    Either.joinRight(
      (spender < date).right.flatMap(
        ds => { f(ds); spender > ds }
      ).right.map(stringUnitToExUnit)
    )

To see that I’m not cheating, here’s my working:

def addSpends(date:Sdate, f:(DailySpend) => Unit):Either[Exception, Unit] = {
    val eds:Either[Exception,DailySpend] = spender < date
    val epo:Either[Exception,PersistentOutcome] =
        eds.right.flatMap(ds => {f(ds); spender > ds})
    val esu:Either[Exception,Either[String,Unit]] = epo
    val eeu:Either[Exception,Either[Exception,Unit]] =
        esu.right.map(stringUnitToExUnit)
    val eu:Either[Exception,Unit] = Either.joinRight(eeu)
    eu
  }

Going to the main function, I notice that you’re folding twice on the same type (Either[Exception,Unit]). You can get rid of the first fold, and with a little bit of point-free style come down to:

def main(args: Array[String]) {
  import Function.const
  addSpends(yesterday, addItems1).right.flatMap(
  const(addSpends(today, addItems2))).fold(
  printError, const(printAllSpends))
}

Once again, the long-hand:

def main(args: Array[String]) {
  val eu1:Either[Exception,Unit] = addSpends(yesterday, addItems1)
  val eu2:Either[Exception,Unit] =
  eu1.right.flatMap(_ => addSpends(today, addItems2))
  val u:Unit = eu2.fold(e => printError(e), _ => printAllSpends)
  u
}

I hope this gives some food for thought and I haven’t misinterpreted what you are trying to achieve!

Going to your chaining solution:

val spender = new Spender with LedgeredMemoryPersister
type Result = Either[Exception, Unit]

  def main(args: Array[String]) {
    failFast(List(()=> addSpends(yesterday, addItems1),
                  ()=> addSpends(today, addItems2),
                  ()=> addSpends(SomeDay(18, February(), 2010), addItems2),
                  ()=> addSpends(SomeDay(19, February(), 2010), addItems1),
                  ()=> addSpends(SomeDay(20, February(), 2010), addItems1))).
            fold(ex => printError(ex), r => printAllSpends)
  }

  def failFast(funcs:List[Function0[Result]]) : Result = {
    if (funcs.isEmpty) Right({})
    else (funcs.head.apply()).fold(l => Left(l), r => failFast(funcs.tail))
  }

  def addSpends(date:Sdate, f:(DailySpend) => Result) : Result = {
    (spender < date).fold(l => Left(l), ds => f(ds))
  }

I think it is more complicated than it needs to be. Obviously, if you need to support a variable-length list of functions to execute, then you need something similar to what you wrote. Actually, probably what you want is more along the lines of:

val params = List((yesterday, addItems1), (today, addItems2). ...)
  params.foldLeft(Right(())((r, p) => r.right.flatMap(addSpends(p))

But if you are just using it as a language feature, then there is something already there for you. First I will write out the long-hand version:

addSpends(yesterday, addItems1).right.flatMap(
    r1 => addSpends(today, addItems2)).right.flatMap(
    r2 => addSpends(SomeDay(18, February(), 2010), addItems2)).right.flatMap(
    r3 => addSpends(SomeDay(19, February(), 2010), addItems1)).right.flatMap(
    r4 => addSpends(SomeDay(20, February(), 2010), addItems1)))).
        fold(ex => printError(ex), r5 => printAllSpends)
}

Using Scala’s for comprehensions, this is equivalent to:

def main(args: Array[String]) {
   for (r1 <- addSpends(yesterday, addItems1).right;
        r2 <- addSpends(today, addItems2)).right;
        r3 <- addSpends(SomeDay(18, February(), 2010), addItems2).right;
        r4 <- addSpends(SomeDay(19, February(), 2010), addItems1).right;
        r5 <- addSpends(SomeDay(20, February(), 2010), addItems1).right)
   yield ()).fold(ex => printError(ex), r => printAllSpends)
}

With an implicit conversion from Either to RightProjection, using ’_’ for the ignored success return values, moving the success case into the yield statement, and mapping the Left type to Unit, that comes down to:

def main(args: Array[String]) {
implicit def EtoRP[A,B](e:Either[A,B]) = e.right
Either.merge(
  for (_ <- addSpends(yesterday, addItems1);
       _ <- addSpends(today, addItems2);
       _ <- addSpends(SomeDay(18, February(), 2010), addItems2);
       _ <- addSpends(SomeDay(19, February(), 2010), addItems1);
       _ <- addSpends(SomeDay(20, February(), 2010), addItems1))
  yield (printAllSpends)).left.map(printError))
}

The Either[A,A] type I mentioned is what you need to use Either.merge. In the previous case, it is obtained by calling left.map to convert Either[Exception,Unit] to Either[Unit,Unit].

Another thing to note is that the examples I’ve provided are a way to show how to better use Either; not necessarily the best way to solve your problem. I would strongly argue that you have way too much in the way of side-effects in the original code. By isolating the side-effects from the purely functional (read: referentially transparent) parts of the code, you could potentially end up with a much more elegant and (de)composable solution. Learning a language that forces you to deal with side-effects will greatly help with working out the isolation.

Finally, I haven’t run any of the above through a compiler, so don’t be surprised if I’ve left something out/stuffed something up!

Given all the great solutions given by Kristian above I decided to go with the following:

type Result = Either[Exception, Unit]
def addSpends(date:Sdate, f:(DailySpend) => Result) : Result

  ...

  type AddByDate = Tuple2[Sdate, Function1[DailySpend, Result]]
  val params = List[AddByDate](
                    (SomeDay(16, February(), 2010), addItems1),
                    (SomeDay(17, February(), 2010), addItems2),
                    (SomeDay(18, February(), 2010), addItems2),
                    (SomeDay(19, February(), 2010), addItems1),
                    (SomeDay(20, February(), 2010), addItems1))
   params.foldLeft(Right():Result)((r, p) => r.right.flatMap(r1 => addSpends(p._1, p._2))).
           fold(ex => printError(ex), r => printAllSpends)

Can the syntax be made cleaner I wonder? Anyway, it’s been a great learning experience!