Java 8 Streams: Map the same object multiple times based on different properties

Question

I was presented with an interesting problem by a colleague of mine and I was unable to find a neat and pretty Java 8 solution. The problem is to stream through a list of POJOs and then collect them in a map based on multiple properties - the mapping causes the POJO to occur multiple times

Imagine the following POJO:

private static class Customer {
    public String first;
    public String last;

    public Customer(String first, String last) {
        this.first = first;
        this.last = last;
    }

    public String toString() {
        return "Customer(" + first + " " + last + ")";
    }
}

Set it up as a List<Customer>:

// The list of customers
List<Customer> customers = Arrays.asList(
        new Customer("Johnny", "Puma"),
        new Customer("Super", "Mac"));

Alternative 1: Use a Map outside of the "stream" (or rather outside forEach).

// Alt 1: not pretty since the resulting map is "outside" of
// the stream. If parallel streams are used it must be
// ConcurrentHashMap
Map<String, Customer> res1 = new HashMap<>();
customers.stream().forEach(c -> {
    res1.put(c.first, c);
    res1.put(c.last, c);
});

Alternative 2: Create map entries and stream them, then flatMap them. IMO it is a bit too verbose and not so easy to read.

// Alt 2: A bit verbose and "new AbstractMap.SimpleEntry" feels as
// a "hard" dependency to AbstractMap
Map<String, Customer> res2 =
        customers.stream()
                .map(p -> {
                    Map.Entry<String, Customer> firstEntry = new AbstractMap.SimpleEntry<>(p.first, p);
                    Map.Entry<String, Customer> lastEntry = new AbstractMap.SimpleEntry<>(p.last, p);
                    return Stream.of(firstEntry, lastEntry);
                })
                .flatMap(Function.identity())
                .collect(Collectors.toMap(
                        Map.Entry::getKey, Map.Entry::getValue));

Alternative 3: This is another one that I came up with the "prettiest" code so far but it uses the three-arg version of reduce and the third parameter is a bit dodgy as found in this question: Purpose of third argument to 'reduce' function in Java 8 functional programming. Furthermore, reduce does not seem like a good fit for this problem since it is mutating and parallel streams may not work with the approach below.

// Alt 3: using reduce. Not so pretty
Map<String, Customer> res3 = customers.stream().reduce(
        new HashMap<>(),
        (m, p) -> {
            m.put(p.first, p);
            m.put(p.last, p);
            return m;
        }, (m1, m2) -> m2 /* <- NOT USED UNLESS PARALLEL */);

If the above code is printed like this:

System.out.println(res1);
System.out.println(res2);
System.out.println(res3);

The result would be:

{Super=Customer(Super Mac), Johnny=Customer(Johnny Puma), Mac=Customer(Super Mac), Puma=Customer(Johnny Puma)}
{Super=Customer(Super Mac), Johnny=Customer(Johnny Puma), Mac=Customer(Super Mac), Puma=Customer(Johnny Puma)}
{Super=Customer(Super Mac), Johnny=Customer(Johnny Puma), Mac=Customer(Super Mac), Puma=Customer(Johnny Puma)}

So, now to my question: How should I, in a Java 8 orderly fashion, stream through the List<Customer> and then somehow collect it as a Map<String, Customer> where you split the whole thing as two keys (first AND last) i.e. the Customer is mapped twice. I do not want to use any 3rd party libraries, I do not want to use a map outside of the stream as in alt 1. Are there any other nice alternatives?

The full code can be found on hastebin for simple copy-paste to get the whole thing running.

Answer 1

I think your alternatives 2 and 3 can be re-written to be more clear:

Alternative 2:

Map<String, Customer> res2 = customers.stream()
    .flatMap(
        c -> Stream.of(c.first, c.last)
        .map(k -> new AbstractMap.SimpleImmutableEntry<>(k, c))
    ).collect(toMap(Map.Entry::getKey, Map.Entry::getValue));

Alternative 3: Your code abuses reduce by mutating the HashMap. To do mutable reduction, use collect:

Map<String, Customer> res3 = customers.stream()
    .collect(
        HashMap::new, 
        (m,c) -> {m.put(c.first, c); m.put(c.last, c);}, 
        HashMap::putAll
    );

Note that these are not identical. Alternative 2 will throw an exception if there are duplicate keys while Alternative 3 will silently overwrite the entries.

If overwriting entries in case of duplicate keys is what you want, I would personally prefer Alternative 3. It is immediately clear to me what it does. It most closely resembles the iterative solution. I would expect it to be more performant as Alternative 2 has to do a bunch of allocations per customer with all that flatmapping.

However, Alternative 2 has a huge advantage over Alternative 3 by separating the production of entries from their aggregation. This gives you a great deal of flexibility. For example, if you want to change Alternative 2 to overwrite entries on duplicate keys instead of throwing an exception, you would simply add (a,b) -> b to toMap(...). If you decide you want to collect matching entries into a list, all you would have to do is replace toMap(...) with groupingBy(...), etc.