Java 8 Stream API - Does any stateful intermediate operation guarantee a new source collection?

Question

Is the following statement true?

The sorted() operation is a “stateful intermediate operation”, which means that subsequent operations no longer operate on the backing collection, but on an internal state.

^{(Source and source - they seem to copy from each other or come from the same source.)}

Disclaimer: I am aware the following snippets are not legit usages of Java Stream API. Don't use in the production code.

I have tested Stream::sorted as a snippet from sources above:

final List<Integer> list = IntStream.range(0, 10).boxed().collect(Collectors.toList());      list.stream()     .filter(i -> i > 5)     .sorted()     .forEach(list::remove);  System.out.println(list);            // Prints [0, 1, 2, 3, 4, 5] 

It works. I replaced Stream::sorted with Stream::distinct, Stream::limit and Stream::skip:

final List<Integer> list = IntStream.range(0, 10).boxed().collect(Collectors.toList());      list.stream()     .filter(i -> i > 5)     .distinct()     .forEach(list::remove);          // Throws NullPointerException 

To my surprise, the NullPointerException is thrown.

All the tested methods follow the stateful intermediate operation characteristics. Yet, this unique behavior of Stream::sorted is not documented nor the Stream operations and pipelines part explains whether the stateful intermediate operations really guarantee a new source collection.

Where my confusion comes from and what is the explanation of the behavior above?

Answer 1

The API documentation makes no such guarantee “that subsequent operations no longer operate on the backing collection”, hence, you should never rely on such a behavior of a particular implementation.

Your example happens to do the desired thing by accident; there’s not even a guarantee that the List created by collect(Collectors.toList()) supports the remove operation.

To show a counter-example

Set<Integer> set = IntStream.range(0, 10).boxed()     .collect(Collectors.toCollection(TreeSet::new)); set.stream()     .filter(i -> i > 5)     .sorted()     .forEach(set::remove); 

throws a ConcurrentModificationException. The reason is that the implementation optimizes this scenario, as the source is already sorted. In principle, it could do the same optimization to your original example, as forEach is explicitly performing the action in no specified order, hence, the sorting is unnecessary.

There are other optimizations imaginable, e.g. sorted().findFirst() could get converted to a “find the minimum” operation, without the need to copy the element into a new storage for sorting.

So the bottom line is, when relying on unspecified behavior, what may happen to work today, may break tomorrow, when new optimizations are added.