Guide to the java.util.Arrays Class

1. Introduction

In this tutorial, we’ll take a look at java.util.Arrays, a utility class that has been part of Java since Java 1.2.

Using Arrays, we can create, compare, sort, search, stream, and transform arrays.

2. Creating

Let’s take a look at some of the ways we can create arrays: copyOf, copyOfRange, and fill.

2.1. copyOf and copyOfRange

To use copyOfRange, we need our original array and the beginning index (inclusive) and end index (exclusive) that we want to copy:

String[] intro = new String[] { "once", "upon", "a", "time" };
String[] abridgement = Arrays.copyOfRange(storyIntro, 0, 3); 

assertArrayEquals(new String[] { "once", "upon", "a" }, abridgement); 
assertFalse(Arrays.equals(intro, abridgement));

And to use copyOf, we’d take intro and a target array size and we’d get back a new array of that length:

String[] revised = Arrays.copyOf(intro, 3);
String[] expanded = Arrays.copyOf(intro, 5);

assertArrayEquals(Arrays.copyOfRange(intro, 0, 3), revised);
assertNull(expanded[4]);

Note that copyOf pads the array with nulls if our target size is bigger than the original size.

2.2. fill

Another way, we can create a fixed-length array, is fill, which is useful when we want an array where all elements are the same:

String[] stutter = new String[3];
Arrays.fill(stutter, "once");

assertTrue(Stream.of(stutter)
  .allMatch(el -> "once".equals(el));

Check out setAll to create an array where the elements are different.

Note that we need to instantiate the array ourselves beforehand–as opposed to something like String[] filled = Arrays.fill(“once”, 3);–since this feature was introduced before generics were available in the language.

3. Comparing

Now let’s switch to methods for comparing arrays.

3.1. equals and deepEquals

We can use equals for simple array comparison by size and contents.  If we add a null as one of the elements, the content check fails:

assertTrue(
  Arrays.equals(new String[] { "once", "upon", "a", "time" }, intro));
assertFalse(
  Arrays.equals(new String[] { "once", "upon", "a", null }, intro));

When we have nested or multi-dimensional arrays, we can use deepEquals to not only check the top-level elements but also perform the check recursively:

Object[] story = new Object[] 
  { intro, new String[] { "chapter one", "chapter two" }, end };
Object[] copy = new Object[] 
  { intro, new String[] { "chapter one", "chapter two" }, end };

assertTrue(Arrays.deepEquals(story, copy));
assertFalse(Arrays.equals(story, copy));

Note how deepEquals passes but equals fails.

This is because deepEquals ultimately calls itself each time it encounters an array, while equals will simply compare sub-arrays’ references.

Also, this makes it dangerous to call on an array with a self-reference!

3.2. hashCode and deepHashCode

The implementation of hashCode will give us the other part of the equals/hashCode contract that is recommended for Java objects.  We use hashCode to compute an integer based on the contents of the array:

Object[] looping = new Object[]{ intro, intro }; 
int hashBefore = Arrays.hashCode(looping);
int deepHashBefore = Arrays.deepHashCode(looping);

Now, we set an element of the original array to null and recompute the hash values:

intro[3] = null;
int hashAfter = Arrays.hashCode(looping);

Alternatively, deepHashCode checks the nested arrays for matching numbers of elements and contents.  If we recalculate with deepHashCode:

int deepHashAfter = Arrays.deepHashCode(looping);

Now, we can see the difference in the two methods:

assertEquals(hashAfter, hashBefore);
assertNotEquals(deepHashAfter, deepHashBefore);

deepHashCode is the underlying calculation used when we are working with data structures like HashMap and HashSet on arrays.

4. Sorting and Searching

Next, let’s take a look at sorting and searching arrays.

4.1. sort

If our elements are either primitives or they implement Comparable, we can use sort to perform an in-line sort:

String[] sorted = Arrays.copyOf(intro, 4);
Arrays.sort(sorted);

assertArrayEquals(
  new String[]{ "a", "once", "time", "upon" }, 
  sorted);

Take care that sort mutates the original reference, which is why we perform a copy here.

sort will use a different algorithm for different array element types. Primitive types use a dual-pivot quicksort and Object types use Timsort. Both have the average case of O(n log(n)) for a randomly-sorted array.

As of Java 8, parallelSort is available for a parallel sort-merge.  It offers a concurrent sorting method using several Arrays.sort tasks.

4.2. binarySearch

Searching in an unsorted array is linear, but if we have a sorted array, then we can do it in O(log n), which is what we can do with binarySearch:

int exact = Arrays.binarySearch(sorted, "time");
int caseInsensitive = Arrays.binarySearch(sorted, "TiMe", String::compareToIgnoreCase);

assertEquals("time", sorted[exact]);
assertEquals(2, exact);
assertEquals(exact, caseInsensitive);

If we don’t provide a Comparator as a third parameter, then binarySearch counts on our element type being of type Comparable.

And again, note that if our array isn’t first sorted, then binarySearch won’t work as we expect!

5. Streaming

As we saw earlier, Arrays was updated in Java 8 to include methods using the Stream API such as parallelSort (mentioned above), stream and setAll.

5.1. stream

stream gives us full access to the Stream API for our array:

Assert.assertEquals(Arrays.stream(intro).count(), 4);

exception.expect(ArrayIndexOutOfBoundsException.class);
Arrays.stream(intro, 2, 1).count();

We can provide inclusive and exclusive indices for the stream however we should expect an ArrayIndexOutOfBoundsException if the indices are out of order,  negative, or out of range.

6. Transforming

Finally, toString, asList, and setAll give us a couple different ways to transform arrays.

6.1. toString and deepToString

A great way we can get a readable version of our original array is with toString:

assertEquals("[once, upon, a, time]", Arrays.toString(storyIntro));

Again we must use the deep version to print the contents of nested arrays:

assertEquals(
  "[[once, upon, a, time], [chapter one, chapter two], [the, end]]",
  Arrays.deepToString(story));

6.2. asList

Most convenient of all the Arrays methods for us to use is the asList. We have an easy way to turn an array into a list:

List<String> rets = Arrays.asList(storyIntro);

assertTrue(rets.contains("upon"));
assertTrue(rets.contains("time"));
assertEquals(rets.size(), 4);

However, the returned List will be a fixed length so we won’t be able to add or remove elements.

Note also that, curiously, java.util.Arrays has its own ArrayList subclass, which asList returns. This can be very deceptive when debugging!

6.3. setAll

With setAll, we can set all of the elements of an array with a functional interface. The generator implementation takes the positional index as a parameter:

String[] longAgo = new String[4];
Arrays.setAll(longAgo, i -> this.getWord(i)); 
assertArrayEquals(longAgo, new String[]{"a","long","time","ago"});

And, of course, exception handling is one of the more dicey parts of using lambdas. So remember that here, if the lambda throws an exception, then Java doesn’t define the final state of the array.

7. Parallel Prefix

Another new method in Arrays introduced since Java 8 is parallelPrefix. With parallelPrefix, we can operate on each element of the input array in a cumulative fashion.

7.1. parallelPrefix

If the operator performs addition like in the following sample, [1, 2, 3, 4] will result in [1, 3, 6, 10]:

int[] arr = new int[] { 1, 2, 3, 4};
Arrays.parallelPrefix(arr, (left, right) -> left + right);
assertThat(arr, is(new int[] { 1, 3, 6, 10}));

Also, we can specify a subrange for the operation:

int[] arri = new int[] { 1, 2, 3, 4, 5 };
Arrays.parallelPrefix(arri, 1, 4, (left, right) -> left + right);
assertThat(arri, is(new int[] { 1, 2, 5, 9, 5 }));

Notice that the method is performed in parallel, so the cumulative operation should be side-effect-free and associative.

For a non-associative function:

int nonassociativeFunc(int left, int right) {
    return left + right*left;
}

using parallelPrefix would yield inconsistent results:

@Test
public void whenPrefixNonAssociative_thenError() {
    boolean consistent = true;
    Random r = new Random();
    for (int k = 0; k < 100_000; k++) {
        int[] arrA = r.ints(100, 1, 5).toArray();
        int[] arrB = Arrays.copyOf(arrA, arrA.length);

        Arrays.parallelPrefix(arrA, this::nonassociativeFunc);

        for (int i = 1; i < arrB.length; i++) {
            arrB[i] = nonassociativeFunc(arrB[i - 1], arrB[i]);
        }

        consistent = Arrays.equals(arrA, arrB);
        if(!consistent) break;
    }
    assertFalse(consistent);
}

7.2. Performance

Parallel prefix computation is usually more efficient than sequential loops, especially for large arrays. When running micro-benchmark on an Intel Xeon machine(6 cores) with JMH, we can see a great performance improvement:

Benchmark                      Mode        Cnt       Score   Error        Units
largeArrayLoopSum             thrpt         5        9.428 ± 0.075        ops/s
largeArrayParallelPrefixSum   thrpt         5       15.235 ± 0.075        ops/s

Benchmark                     Mode         Cnt       Score   Error        Units
largeArrayLoopSum             avgt          5      105.825 ± 0.846        ops/s
largeArrayParallelPrefixSum   avgt          5       65.676 ± 0.828        ops/s

Here is the benchmark code:

@Benchmark
public void largeArrayLoopSum(BigArray bigArray, Blackhole blackhole) {
  for (int i = 0; i < ARRAY_SIZE - 1; i++) {
    bigArray.data[i + 1] += bigArray.data[i];
  }
  blackhole.consume(bigArray.data);
}

@Benchmark
public void largeArrayParallelPrefixSum(BigArray bigArray, Blackhole blackhole) {
  Arrays.parallelPrefix(bigArray.data, (left, right) -> left + right);
  blackhole.consume(bigArray.data);
}

7. Conclusion

In this article, we learned how some methods for creating, searching, sorting and transforming arrays using the java.util.Arrays class.

This class has been expanded in more recent Java releases with the inclusion of stream producing and consuming methods in Java 8 and mismatch methods in Java 9.