Refactoring code to Java 8 Stream API

You may be wondering why a blog post in 2021 is mentioning Java 8, as it came out over 7 years ago.
The reason is that I have come across a lot of Java codebase which are indeed using Java 8 or a later version, however they are not always leveraging the Java 8 API.
I often find code blocks which could be refactored. What I mean here by refactoring is that these code blocks could leverage the Java 8 APIs to become more readable, more concise (i.e. less verbose) and also less outdated.

I have found that doing this kind of very small refactoring is a good way to get familiar with an unknown codebase — which you can do while reading through the code.
Always make sure that there is a good suite of tests for the code you are refactoring, and to regularly run them every time you make changes.

In this blog post, I would like to share some common patterns that I have come across, and how they can be changed — focusing on the usage of Java 8 Stream API.

Examples

Filtering

One very common action in programming is to filter a collection of elements. To do this prior to Java 8, you would do the following:

List<String> longWords(final List<String> words, final int threshold) {
    List<String> longWords = new ArrayList<>();
    for (String word : words) {
        if (word.length() > threshold) {
            longWords.add(word);
        }
    }
    return longWords;
}

The above method could be replaced with the following:

List<String> longWords(final List<String> words, final int threshold) {
    return words.stream()
            .filter(word -> word.length() > threshold)
            .collect(toList());
}

This is more concise, and more readable.

Counting

Instead of filtering, you may simply want to count the long words.
Prior to Java 8 you could do:

List<String> longWords(final List<String> words, final int threshold) {
     List<String> longWords = new ArrayList<>();
     for (String word : words) {
        if (word.length() > threshold) {
            longWords.add(word);
        }
     }
     return longWords;
}

With Java 8 Stream API you can do:

long longWordsCount(final List<String> words, final int threshold) {
    return words.stream()
            .filter(word -> word.length() > threshold)
            .count();
}

Grouping

Another common action is to group elements from a list, based on a specific condition or predicate.
For example, you may want to group words by length. To do this prior to Java 8, you would do the following:

Map<Integer, List<String>> groupByWordLength(final List<String> words) {
    Map<Integer, List<String>> groups = new HashMap<>();
    for (String word : words) {
        int length = word.length();
        if (groups.containsKey(length)) {
            List<String> group = groups.get(length);
            group.add(word);
        } else {
            List<String> group = new ArrayList<>();
            group.add(word);
            groups.put(length, group);
        }
    }
    return groups;
}

This is starting to be already quite verbose for such a common task. Thankfully it becomes a lot simpler leveraging Java 8 Stream:

Map<Integer, List<String>> groupByWordLength(final List<String> words) {
    return words.stream()
            .collect(groupingBy(word -> word.length()));
}

This one liner is also a lot more readable.

Partitioning

One slightly less common but still frequent action is to partition a list, based on a specific condition or predicate. Continuing with our "words" example, we can partition words based on their lengths. To do this prior to Java 8, you would do the following:

List<List<String>> partitionByWordLength(final List<String> words, final int threshold) {
    List<String> shortWords = new ArrayList<>();
    List<String> longWords = new ArrayList<>();
    for (String word : words) {
        if (word.length() > threshold) {
            longWords.add(word);
        } else {
            shortWords.add(word);
        }
    }
    return Arrays.asList(shortWords, longWords);
}

With Java 8 Stream API, you can refactor the method to be:

List<List<String>> partitionByWordLength(final List<String> words, final int threshold) {
    Collection<List<String>> partition = words.stream()
        .collect(partitioningBy(word -> word.length() > threshold))
        .values();
    return new ArrayList<>(partition);
}

Here we need to convert the Collection to an ArrayList because the type returned by partitioningBy collector is a Map.
We could have simply returned the Map of Lists which contains two keys, true and false.

Transforming

Transformation are also very common, a transformation is when you want to apply a function to each element in a list which converts it to something else.
There are a lot of method which use the following patterns:

static List<String> fooTransform(final List<String> words) {
    List<String> fooStuff = new ArrayList<>();
    for (String word : words) {
        String foo1 = word.repeat(3);
        String foo2 = foo1.stripTrailing();
        String foo3 = foo2.toUpperCase();
        fooStuff.add(foo3);
    }
    return fooStuff;
}

Using streams, you can do the following:

List<String> fooTransform(final List<String> words) {
    return words.stream()
        .map(word -> fooThing(word))
        .collect(toList());
}

private String fooThing(final String word) {
    String foo1 = word.repeat(3);
    String foo2 = foo1.stripTrailing();
    return foo2.toUpperCase();
}

Extracting the inner block computation of the for loop in a helper method allows to then have a clean and readable transformation with map.

Concatenation

Another common practice is to fetch data from two or more different place and then merge the elements into a final list.
To do this prior to Java 8, you could do the following:

List<String> concat(final List<String> words1, final List<String> words2) {
    List<String> concat = new ArrayList<>();

    for (String word : words1) {
        concat.add(word);
    }

    for (String word : words2) {
        concat.add(word);
    }

    return concat;
}

This can be transformed to a (fairly long) one liner :

List<String> concat(final List<String> words1, final List<String> words2) {
    return Stream.concat(words1.stream(), words2.stream()).collect(toList());
}

The above snippet is leveraging the concat method for Streams, and then collecting back to a List.
If you want to add a condition in you concatenation, you can easily do it by using the filter method on each word stream.

Summation

Having a raw collection of elements does not always give good insights, and we often need to do computation on the list like calculting the average, the sum etc.
For example, we might want to know the sum of characters in a list of words for words longer than a given threshold.
To do this prior to Java 8, you could do the following:

int characterSum(final List<String> words, final int threshold) {
    int sum = 0;
    for (String word : words) {
        int length = word.length();
        if (length > threshold) {
            sum += length;
        }
    }
    return sum;
}

This can be achieved by leveraging the reduce method of Streams. This method perform what is called a reduction on the elements of the stream. It takes on two parameters, an initial value for the reduction and an accumulator.
The simplest example is a sum:

int characterSum(final List<String> words, final int threshold) {
    return words.stream()
            .map(word -> word.length())
            .filter(length -> length > threshold)
            .reduce(0, (sum, l) -> sum + l);
}

The accumulator function is applied at each iteration with the current temporary sum at that given iteration.
The initial value here is 0.

Performance comparison

From what I have read, using streams does not degrade performance. However, I did not want to simply guess, rather I wanted to test it myself using the above examples.
I have made very simple and approximate (or "quick and dirty") timer tests using System.nanoTime() before and after executing the method on my laptop.

For all the examples, I have generated either 5 hundred thousands or 5 millions random words, and executed the method 100 times.
The results I will share below is the average executing time in milliseconds.

I will also use both the stream() and the parallelStream() method for comparison.
When using the parallelStream() method, it creates multiple substreams which are processed in parallel by different threads. The intermediary results of each substream then gets combined when collecting.
However, note that you have to be careful when using parallelStream():

• iteration operation should not depend/interfere on/with each other
• iteration operation should not modify a common mutable variable

Filtering

Reusing my filtering example, the results are the following:

• Java 7: 23,66 ms
• Java 8 sequential stream: 24,01 ms
• Java 8 parallel stream: 11,82 ms

Counting

Reusing my counting example, the results are the following:

• Java 7: 20,75 ms
• Java 8 sequential stream: 20,79 ms
• Java 8 parallel stream: 9,11 ms

Grouping

Reusing my grouping example, the results are the following:

• Java 7: 17,24 ms
• Java 8 sequential stream: 17,98 ms
• Java 8 parallel stream: 13,36 ms

Partitioning

Reusing my partitioning example, the results are the following:

• Java 7: 8,71 ms
• Java 8 sequential stream: 10,40 ms
• Java 8 parallel stream: 7,40 ms

Transforming

Reusing my transforming example, the results are the following:

• Java 7: 92,78 ms
• Java 8 sequential stream: 77,86 ms
• Java 8 parallel stream: 67,22 ms

Concatenation

Reusing my concatenation example, the results are the following:

• Java 7: 26,72 ms
• Java 8 sequential stream: 22,54 ms
• Java 8 parallel stream: 26,52 ms

Summation

Reusing my summation example, the results are the following:

• Java 7: 22,13 ms
• Java 8 sequential stream: 22,52 ms
• Java 8 parallel stream: 11,64 ms

Results

Disclaimer that my testing was "quick and dirty" and that execution time depends on a lot of factor, especially in Java, such as startup time, number of threads available, number of resource available for the JVM, etc.
So take the results with as approximations.

Sequential streams

It seems that sequential streams perform slightly worse that for/loop in most cases.
However, from my point of view — unless you are optimizing for performance — the tradeoff between having code that is more readable and concise, meaning more testable and maintainable versus slight worse performance makes using Java 8 Stream completely worth it.

Parallel streams

Parallel streams usually performs a lot better that for/loop.
When the operation does not depend on the order of the iteration, and each iteration does not modify any shared variable — using parallel streams is a good way to increase the performance of your looping logic.

Conlusion

Java 8 Stream API allows to have code that is more readable and more concise. This is very valuable for making code easier to test and maintain.
Refactoring to using Java 8 Stream is a good habit and can allow you to make small improvements to a new codebase while learning how things work.
So I would recommend you start trying to identify classes or methods that are using old looping patterns and replace them with streams when applicable.