Guard conditions in Java

So, my impetus for writing this post is the ‘require’ method in scala. It works a bit like the failed ‘assert’ that came to Java a while ago but never really took off. The ‘require’ in scala is used to write guard conditions, similar to assertions you may make in JUnit tests – here is an example:

object Shuffler {
  
  def shuffle(deck:Deck) = {
    require(deck.isNotEmpty, "deck cannot be empty")
    // code omitted...
  }

}

In Java I’ve seen the following:

public void shuffle(Deck deck) {
  // guard condition
  if (deck.isEmpty()) {  
    throw new IllegalArgumentException("Cannot shuffle an empty deck");
  }
  // code omitted...
}

and also

public void shuffle(Deck deck) {
  // guard condition
  if (deck.isNotEmpty()) {  
    // code omitted...
  } else {
    throw new IllegalArgumentException("Cannot shuffle an empty deck");
  }
}

This style of code can quickly create multiple nested ifs and make the code less readable. An old adage said “no more than one return point in a method” but this rule has not stood the test of time. As Kent Beck writes, this “was to prevent the confusion possible when jumping into and out of many locations in the same routine. It made good sense when applied to FORTRAN or assembly language programs written with lots of global data where even understanding which statements were executed was hard work … with small methods and mostly local data, it is needlessly conservative.”

Option 1: Roll-your-own Guard Condition

We fail fast – error if the condition we cannot handle is true. Note – now our happy path code (omitted) is not inside an if or else. We can write multiple guard conditions but can recongnise the guard conditions for what they are above the actual part of the method that affects change.

public void shuffle(Deck deck) {
  // guard condition
  if (deck.isEmpty()) {  
    throw new IllegalArgumentException("Cannot shuffle an empty deck");
  } 
  // code omitted...
}

Option 2: External Libaries

I think we can improve on this using a static method similar to the Assert class we use in JUnit, and classes to do this already exist. Our options include Validate (commons lang), Assert (Spring framework) and Preconditions (google-guava).

Here is an example using Preconditions:

import com.google.common.base.Preconditions;

public void shuffle(Deck deck) {
  // guard condition
  Preconditions.checkArgument(deck.isNotEmpty(), "Cannot shuffle an empty deck");
  
  // code omitted...
}

This is much simpler and easier to read for me. I believe that, as a rule of thumb, the code doing the primary purpose of a function should generally be equal to the amount of code doing the following tasks: Logging, Auditing, Error Handling, Guard Conditions.

A good comparison of the options is available here: http://www.sw-engineering-candies.com/blog-1/comparison-of-ways-to-check-preconditions-in-java

Testing methods – visibility / testability problem.

So, a big argument I’ve encountered in Java OOP is a design problem about testing methods that aren’t public. I shall give an example.

I am writing a virtual BlackJack game. I have a class CardDealer which deals cards. It has a method which works out if the deck needs reshuffling, with the following logic:

public class CardDealer implements ICardDealer {
  
  @Override
  public void dealCards(Collection<Player> players, Deck deck)  {
    shuffleIfRequired(players, deck);
  }

  private void shuffleIfRequired(Collection<Player> players, Deck deck) {
    // calculate if enough cards left undealt in deck and if not shuffle...
  }
}

Now the problem is whether to test this method in isolation, and there are conflicting concerns:

1. Only test the Class against its public interface. Oh, well we can’t test the shuffleIfRequired method in isolation then. In my experience this is good and bad. I might have 5 ‘shuffleIfRequired’ scenarios I want to test and for each one I have to concern myself with the whole of the ‘dealCards’ method.

The answer – extract class (Refactoring by Martin Fowler). Using the logic given above should take this method out of this class, removing its context and then we can provide the functionality in another class, e.g.

public class Shuffle {
  public static void ifRequired(int players, Deck deck) {
    // context-free functionality here.
  }
}

Arguments against include ‘moving the responsibility out of the most appropriate class’, and ‘yet another utility class/method’.

2. Put testing above method visibility. This is typically done by making private methods protected or package-private / default access – to allow calling them from Unit tests (JUnit classes are typically in the same package). This argument can sound very bad to some people, their argument being that this waters down the design and leave the class open to misuse. There are 2 arguments for this as follows:

1. the code is more at risk from untested code than misused class, assuming code reviews.
2. the non-private, non-public method will not be part of the public interface of the class – being neither public nor in the interface.

Where the ‘test against contract’ enforces each test of shuffleIfRequired be a call to the ‘dealCards’ method, this approach means dealCards can be ignored – which some people accept and some people think is not so good.

3. Test private methods using reflection to invoke the method. This means hard-coding the method name into the test case and losing the power of refactoring tools and compile-time checking. There aren’t many arguments for this, but I think generally this is:

the test will break the first time the code is built locally and can be fixed easily enough depending on how much the method is called and how much it is changed.

Vinny’s Verdict:

I won’t tell people which of these to go with – I doubt I’ll change anyone’s opinion. But… I think it’s worth at least discussing here and hopefully this post can help focus discussions on it.

#8 of 99 Problems – Eliminate consecutive duplicates of list elements.

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

This problem again relies on the functionality built into Scala collections and I don’t add much to the original solution.

Problem:
If a list contains repeated elements they should be replaced with a single copy of the element. The order of the elements should not be changed.
Example:

var result = compress(List('a, 'a, 'a, 'a, 'b, 'c, 'c, 'a, 'a, 'd, 'e, 'e, 'e, 'e))
println("Result = " + result)

Output: Result = List(‘a, ‘b, ‘c, ‘a, ‘d, ‘e)

Answer:

object Class08 extends App {

  def compress[A](ls: List[A]): List[A] = ls match {
    case Nil       => Nil
    case h :: tail => h :: compress(tail.dropWhile(_ == h))
  }

  var result = compress(List('a, 'a, 'a, 'a, 'b, 'c, 'c, 'a, 'a, 'd, 'e, 'e, 'e, 'e))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

This is a good example of using dropWhile, classified as a ‘subiterator’ here – http://www.scala-lang.org/docu/files/collections-api/collections_43.html

#7 of 99 Problems – flatten a list

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

This problem is deeply rooted in the functional programming side of scala, which is not my strongest subject as I come from an OO background. I cannot add to the original solution given and merely post the solution here for completeness – there is no point posting 90 or so of 99 problems 🙂

Anyway, here it is…

Question:
Flatten a nested list structure.
Example:

var result = flatten(List(List(1, 1), 2, List(3, List(5, 8))))
println("Result = " + result)

Output: Result = List(1, 1, 2, 3, 5, 8)

Answer:

object Class07 extends App {
  def flatten(ls: List[Any]): List[Any] = ls flatMap {
    /* this is an example of pattern matching on type, here
     * the code says 'if this item is itself a list'
     */
    case ms: List[_] => flatten(ms)
    // this matches any other item
    case e => List(e)
  }
  var result = flatten(List(List(1, 1), 2, List(3, List(5, 8))))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

#6 of 99 Problems – is a list a palindrome in scala

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

This is more involved than previous problems. We are checking if a list is a palindrome and we will do this, as in previous problems, using pattern matching on the List. I use 3 scenarios which would be ifs in Java.

1. If the list is less than 2 items it is empty or has 1 item. I count these as being palindromes.
2. If the head (first item) and the last (last item) are the same then it might be – let’s recurse 🙂
3. If 2 items or more and first and last differ then return false.

This means using an if in our case to check first and last items in the 2nd scenario.

A note on pattern matching: The line of code below will create 2 values from the list – head and tail – where head is the first item and tail is everything else. It makes matching and recursion simpler.

case head::tail => // do something...

Anyway, here’s the code:

Question:
Find out whether a list is a palindrome.
Example:

var result = isPalindrome(List(1, 2, 3, 2, 1))
println("Result = " + result)

Output: Result = true

Answer:

import scala.annotation.tailrec

object Class06 extends App {

  var result = isPalindrome(List(1, 2, 3, 2, 1))
 
  @tailrec // we expect tail-recursive optimization
  def isPalindrome(l:List[Any]):Boolean = l match {
    case yes if(l.length < 2) => true  // true if small enough
    case head::tail if (head==tail.last) => isPalindrome(tail.init) // recurse without first and last
    case _ =>  false // this is like an else
  }   

  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

This shows some of the features of pattern matching in scala. The word yes is just a name I’ve given to the first case which runs off the if condition. There is a convention to use an underscore for catch-all cases, a convention I follow here. The original list of questions doesn’t show anything beyond list==list.reverse but I think this answer is a good example of recursion and pattern matching.

#5 of 99 Problems – Reverse a List in scala

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

This example reverses the list purely as an example of writing a recursive function in scala. Like the other examples I’ve written, the aim is simple code, not necessarily the most performant and not using built in List functionality per se.

Question:
Reverse a List
Example:

var result = reverse(List(1, 1, 2, 3, 5, 8))
println("Result = " + result)

Output: Result = List(8, 5, 3, 2, 1, 1)

Answer:

object Class05 extends App {

  // l is Parameter of type List[Any] and return is of type List[Any]
  def reverse(l:List[Any]): List[Any] = l match {
    case Nil => l // if empty list passed return it and stop recursion
    case head :: tail => reverse(tail) ::: List(head)     // recurse with all but first items, appending first item.
  }   

  val result = reverse(List(1, 1, 2, 3, 5, 8))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

This is an example of a pattern of writing recursive functions but breaking out of the recursion when our list is exhausted. I must admit – the original 99 problem post provides more elegant solutions than I am capable of and is worth a look to see other ways of achieving the same thing,

#4 of 99 Problems – count items in a list

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

The easiest way to get the item count for a List in scala is to use the size or length functions provided by List. This answer is an example of writing a function in scala, more specifically a recursive function without side-effects.

Question:
Find the number of elements of a list.
Example:

var result = length(List(1, 1, 2, 3, 5, 8))
println("Result = " + result)

Output: Result = 6

Answer:

object Class04 extends App {

  // Outer function:  
  def length(l:List[Any]):Int = {

    // Inner function:
    @tailrec
    def length(count:Int, l:List[Any]):Int = l match {
      case head::tail => length(count + 1, tail) // recurse
      case Nil        => count                   // empty list so return count
    }

    // call that recursive inner function with initial state:
    length(0, l)
  }

  val result = length(List(1, 1, 2, 3, 5, 8))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

This is an example of a pattern of writing recursive functions – an outer and an inner function. The outer function creates the starting point and calls the inner recursive function which exhaustively iterates, in this case through a list.

#3 of 99 Problems

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

Question:
Find the Kth element of a list.
By convention, the first element in the list is element 0.
Example:

var result = nth(2, List(1, 1, 2, 3, 5, 8))
println("Result = " + result)

Output: Result = 2

Answer:

object Class03 extends App {

  @tailrec
  def nth(n:Int, l:List[Any]):Any = (n,l) match {
    case (0, head::tail) => head // return top item when counter reaches zero
    case (_, head::tail) => nth(n-1, tail) // recurse with counter-1 and list minus top element.
  }

  val result = nth(2, List(1, 1, 2, 3, 5, 8))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

This pattern for decrementing a counter without a temporary variable is common in functional programming.

#2 of 99 Problems

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

Question:
Find the last but one element of a list.
Example:

val result = penultimate(List(1, 1, 2, 3, 5, 8))
println("Result = " + result)

Output: Result = 5

Answer:

object Class02 extends App {

  @tailrec
  def penultimate(l:List[Any]):Any = l match {
    case head::tail if (tail.length == 1) => head // return top item if only 1 item left
    case head::tail                       => penultimate(tail) // otherwise recurse
  }

  val result = penultimate(List(1, 1, 2, 3, 5, 8))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

The @tailrec tells the compiler that we are expecting tail-recursion optimisation and to error if it is not able to perform this.

#1 of 99 Problems

This post and the original question are licensed under a Creative Commons Attribution-Share Alike license and all code is released to the public domain.

Question:
Find the last element of a list.
Example:

val result = last(List(1, 1, 2, 3, 5, 8))
println("Result = " + result)

Output: Result = 8

Answer:

object Class01 extends App {
  def last(l:List[Any]) = l.last // this is the method - pretty simple
  val result = last(List(1, 1, 2, 3, 5, 8))
  println("Result = " + result)
}

Original Question here: http://aperiodic.net/phil/scala/s-99/

Note the definition of the List parameter has to be provided with a type – here I use ‘Any’. This is not optional as it is in Java.