Java 8 Stream “collect and group by” objects that map to multiple keys [duplicate]

Answer 1

I think, this is only possible with an intermediate “pair” value to remember the association between a person/manager and the originating item. In absence of a standard pair type in Java’s standard API, we have to resort to Map.Entry which comes closest to a Pair type:

Map<Person, List<Item>> map = list.stream()
  .flatMap(item->item.getOwners().stream()
    .map(p->new AbstractMap.SimpleEntry<>(p.getManager(), item)))
  .collect(Collectors.groupingBy(Map.Entry::getKey,
    Collectors.mapping(Map.Entry::getValue, Collectors.toList())));

after improving it using import static we get

Map<Person, List<Item>> map = list.stream()
  .flatMap(item->item.getOwners().stream().map(p->new SimpleEntry<>(p.getManager(), item)))
  .collect(groupingBy(Map.Entry::getKey, mapping(Map.Entry::getValue, toList())));

It results in

m1: [i1,i3]
m3: [i1,i2]
m2: [i1,i2]

which differs because first, the standard map has no defined ordering, second, I think you made a mistake regarding you expectation as m1 is not associated with i2 in your example data.

Answer 2

Such scenarios are well supported by my free StreamEx library which enhances standard Stream API:

Map<Person, List<Item>> map = StreamEx.of(list) // create an enhanced stream of Item
                // create a stream of Entry<Item, manager>
                .cross(item -> item.getOwners().stream().map(Person::getManager))
                // swap keys and values to get stream of Entry<manager, Item>
                .invert()
                .grouping();

Internally it's similar to @Holger solution.