How can I transform a collection into a Guava Multimap grouped by the elements of a nested collection property?

Question

I have a List and want a Guava Multimap where we\'ve grouped the Foos by each tag of their Collection

Answer 1

You can use custom collector for this:

Multimap map = list.stream().collect(
    ImmutableMultimap::builder,
    (builder, value) -> value.getTags().forEach(tag -> builder.put(tag, value)),
    (builder1, builder2) -> builder1.putAll(builder2.build())
).build();

This does not cause extra side effects (see here on this), is concurrent and more idiomatic.

You can also extract these ad-hoc lambdas into a full-fledged collector, something like this:

public static  Collector> toMultimapByKey(Function> keysMapper) {
    return new MultimapCollector<>(keysMapper);
}

private static class MultimapCollector implements Collector, Multimap> {
    private final Function> keysMapper;

    private MultimapCollector(Function> keysMapper) {
        this.keysMapper = keysMapper;
    }

    @Override
    public Supplier> supplier() {
        return ImmutableMultimap::builder;
    }

    @Override
    public BiConsumer, T> accumulator() {
        return (builder, value) -> keysMapper.apply(value).forEach(k -> builder.put(k, value));
    }

    @Override
    public BinaryOperator> combiner() {
        return (b1, b2) -> b1.putAll(b2.build());
    }

    @Override
    public Function, Multimap> finisher() {
        return ImmutableMultimap.Builder::build;
    }

    @Override
    public Set characteristics() {
        return Collections.emptySet();
    }
}

Then the collection would look like this:

Multimap map = list.stream().collect(toMultimapByKey(Foo::getTags));

You can also return EnumSet.of(Characteristics.UNORDERED) from characteristics() method if the order is not important for you. This can make internal collection machinery act more efficiently, especially in case of parallel reduction.