Skip Top Navigation Bar
Previous in the Series
Current Tutorial
Writing and Combining Comparators
That's the end of the series!

Previous in the Series: Combining Lambda Expressions

Writing and Combining Comparators

 

Implementing a Comparator with a Lambda Expression

Thanks to the definition of functional interfaces, the good old Comparator<T> interface introduced in JDK 2 became functional. So, implementing a comparator can be done using a lambda expression.

Here is the only abstract method of the Comparator<T> interface:

@FunctionalInterface
public interface Comparator<T> {

    int compare(T o1, T o2);
}

The contract of a comparator is the following:

It is not strictly required that in case o1.equals(o2) is true, the comparison of o1 and o2 returns 0.

How can you create a comparator of integers, that would implement the natural order? Well, you can just use the method you saw at the beginning of this tutorial:

Comparator<Integer> comparator = (i1, i2) -> Integer.compare(i1, i2);

You may have noticed that this lambda expression can also be written with a very nice bound method reference in that way:

Comparator<Integer> comparator = Integer::compare;

Refrain from implementing this comparator with (i1 - i2). Even if this pattern seems to be working, there are corner cases where it will not produce a correct result.

This pattern can be extended to anything you need to compare, as long as you follow the contract of the comparator.

The Comparator API went one step further, by providing a very useful API to create comparators in a much more readable way.

 

Using a Factory Method to Create a Comparator

Suppose you need to create a comparator to compare strings of characters in a non-natural way: the shortest strings are lesser than the longest strings.

Such a comparator can be written in that way:

Comparator<String> comparator =
        (s1, s2) -> Integer.compare(s1.length(), s2.length());

You learned in the previous part that it is possible to chain and compose lambda expressions. This code is another example of such a composition. Indeed, you can rewrite it in that way:

Function<String, Integer> toLength = String::length;
Comparator<String> comparator =
        (s1, s2) -> Integer.compare(
                toLength.apply(s1),
                toLength.apply(s2));

Now you can see that the code of this Comparator only depends on the Function called toLength. So it becomes possible to create a factory method that takes this function as an argument and returns the corresponding Comparator<String>.

There is still a constraint of the returned type of the toLength function: it has to be comparable. Here it works well because you can always compare integers with their natural order, but you need to keep that in mind.

Such a factory method does exist in the JDK: it has been added to the Comparator interface directly. So you can write the previous code in that way:

Comparator<String> comparator = Comparator.comparing(String::length);

This comparing() method is a static method of the Comparator interface. It takes a Function as an argument, that should return a type that is an extension of Comparable.

Suppose you have a User class with a getName() getter, and you need to sort a list of users according to their name. The code you need to write is the following:

List<User> users = ...; // this is your list
Comparator<User> byName = Comparator.comparing(User::getName);
users.sort(byName);

 

Chaining Comparators

The company you are working for is currently very happy with the Comparable<User> you have delivered. But there is a new requirement in version 2: the User class now has a firstName and a lastName, and you need to produce a new Comparator to handle this change.

Writing each comparator follows the same pattern as the previous one:

Comparator<User> byFirstName = Comparator.comparing(User::getFirstName);
Comparator<User> byLastName = Comparator.comparing(User::getLastName);

Now what you need is a way to chain them, just as you chained instances of Predicate or Consumer. The Comparator API gives you a solution to do that:

Comparator<User> byFirstNameThenLastName =
        byFirstName.thenComparing(byLastName);

The thenComparing() method is a default method of the Comparator interface, that takes another comparator as an argument and returns a new comparator. When applied to two users, the comparator first compare these users using the byFirstName comparator. If the result is 0, then it will compare them using the byLastName comparator. In a nutshell: it works as expected.

The Comparator API went one step further: since byLastName only depends on the User::getLastName function, an overload of the thenComparing() method has been added to the API which takes this function as an argument. So the pattern becomes the following:

Comparator<User> byFirstNameThenLastName =
        Comparator.comparing(User::getFirstName)
                  .thenComparing(User::getLastName);

With lambda expressions, method references, chaining, and composition, creating comparators has never been so easy!

 

Specialized Comparators

Boxing and unboxing or primitive types can also occur with comparators, leading to the same performance hits as in the case of the functional interfaces of the java.util.function package. To deal with this problem, specialized versions of the comparing() factory method and the thenComparing() default method have been added.

You can also create an instance of Comparator<T> with:

You use these methods if you need to compare objects using a property that is a primitive type and need to avoid the boxing / unboxing of this primitive type.

There are also corresponding methods to chain Comparator<T>:

The idea is the same: using these methods, you can chain the comparison with a comparator built on a specialized function that returns a primitive type, without having any performance hit due to boxing / unboxing.

 

Comparing Comparable Objects Using Their Natural Order

There are several factory methods worth mentioning in this tutorial, that will help you create simple comparators.

Many classes in the JDK and probably in your application too are implementing a special interface of the JDK: the Comparable<T> interface. This interface has one method: compareTo(T other) that returns an int. This method is used to compare this instance of T with other, following the contract of the Comparator<T> interface.

Many classes of the JDK are already implementing this interface. This is the case for all the wrapper classes of primitive types (Integer, Long and the like), for the String class, and for date and time classes from the Date and Time API.

You can compare the instances of these classes using their natural order, that is, by using this compareTo() method. The Comparator API gives you a Comparator.naturalOrder() factory class. The comparator it builds does exactly that: it compares any Comparable object using its compareTo() method.

Having such a factory method can be very useful when you need to chain comparators. Here is an example, where you want to compare string of characters with their length, and then with their natural order (this example uses a static import for the naturalOrder() method to further improve readability):

Comparator<String> byLengthThenAlphabetically =
        Comparator.comparing(String::length)
                  .thenComparing(naturalOrder());
List<String> strings = Arrays.asList("one", "two", "three", "four", "five");
strings.sort(byLengthThenAlphabetically);
System.out.println(strings);

Running this code produces the following result:

[one, two, five, four, three]

 

Reversing a Comparator

One major use of comparators is of course to sort lists of objects. The JDK 8 saw the addition of a method on the List interface especially for that: List.sort(). This method takes a comparator as an argument.

If you need to sort the previous list in reverse order, you can use the nice reversed() method from the Comparator interface.

List<String> strings =
        Arrays.asList("one", "two", "three", "four", "five");
strings.sort(byLengthThenAlphabetically.reversed());
System.out.println(strings);

Running this code will produce the following result:

[three, four, five, two, one]

 

Dealing with Null Values

Comparing null objects may lead to nasty NullPointerException while running your code, something you want to avoid.

Suppose you need to write a null-safe comparator of integers to sort a list of integers. The convention you decide to follow is to push all the null values at the end of your list, meaning that a null value is greater than any other non-null value. And then you want to sort non-null values in their natural order.

Here is the kind of code you may write to implement this behavior:

Comparator<Integer> comparator =
        (i1, i2) -> {
            if (i1 == null && i1 != null) {
                return 1;
            } else if (i1 != null && i2 == null) {
                return -1;
            } else {
                return Integer.compare(i1, i2);
            }
        };

You can compare this code to the first comparator you wrote at the beginning of this part, and see that readability took a big hit.

Fortunately, there is a much easier way to write this comparator, using another factory method of the Comparator interface.

Comparator<Integer> naturalOrder = Comparator.naturalOrder();

Comparator<Integer> naturalOrderNullsLast =
        Comparator.nullsLast(naturalOrder());

The nullsLast() and its sibling method nullsFirst() are factory methods of the Comparator interface. Both take a comparator as an argument and do just that: handle the null values for you, pushing them to the end, or putting them first in your sorted list.

Here is an example:

List<String> strings =
        Arrays.asList("one", null, "two", "three", null, null, "four", "five");
Comparator<String> naturalNullsLast =
        Comparator.nullsLast(naturalOrder());
strings.sort(naturalNullsLast);
System.out.println(strings);

Running this code produces the following result:

[five, four, one, three, two, null, null, null]

More Learning


Last update: February 24, 2023


Previous in the Series
Current Tutorial
Writing and Combining Comparators
That's the end of the series!

Previous in the Series: Combining Lambda Expressions