Collections should override spliterator()

[cowwoc]

Because Guava's collection classes (e.g. ImmutableList) do not override spliterator() they inherit the default implementation which has a characteristics of SIZED and SUBSIZED. This means that they cannot be used for parallel processing with ForkJoin.

Expected behavior: Immutable* should provide a more efficient spliterator() implementation.

[lowasser]

This will absolutely be part of Guava as soon as we support Java 8, but I'm not sure there is much we can do until then?

[cowwoc]

I found posts going back 1-2 years talking about adding Java8 support in the near future: http://www.reddit.com/r/java/comments/1y9e6t/ama_were_the_google_team_behind_guava_dagger/cfjd2qi and https://groups.google.com/forum/#!topic/guava-discuss/fEdrMyNa8tA

Is there an open issue tracking this work?

Is there a short-term workaround I can use in the meantime?

[lowasser]

I don't think so - the primary issue is Google internal?

As it stands, for essentially all immutable collections, a good implementation is

IntStream.range(0, collection.size()).mapToObj(collection.asList()::get)

[cowwoc]

The primary issue is Google internal?

I don't understand the question.

As it stands, for essentially all immutable collections, a good implementation is

IntStream.range(0, collection.size()).mapToObj(collection.asList()::get)

This won't do because spliterator().characteristics() will return ORDERED | SIZED | SUBSIZED. It is missing IMMUTABLE.

[lowasser]

The Google internal Comment was a statement, not a question.

I'm not certain the IMMUTABLE characteristic is actually used for anything? That should still be an adequate substitute for most use cases I can think of...

[cowwoc]

The Google internal Comment was a statement, not a question.

It had a question mark after it... :)

I'm not certain the IMMUTABLE characteristic is actually used for anything?

Agreed. I thought that this was being used for parallelization but it looks like it is not. So for now I don't have a concrete use-case for this characteristic.

That should still be an adequate substitute for most use cases I can think of...

Okay, but it is worth noting that the only characteristic it adds over the default is ORDERED. That said, it is useful for clients that wishs to jump to the end of a list: http://stackoverflow.com/a/29553490/14731.