See the bug description for details. I propose a simple solution. Let's allow ArraySpliterator to be non-SIZED and report artificial estimatedSize(), much bigger than the real one. This will allow AbstractSpliterator and IteratorSpliterator to produce prefix whose size is comparable to Long.MAX_VALUE (say, starting with Long.MAX_VALUE/2), and this will enable further splitting of the prefix. This change will drastically improve parallel streaming for affected streams of size <= 1024 and significantly improve for streams of size 1025..20000. The cost is higher-grained splitting for huge streams of unknown size. This might add a minor overhead for such scenarios which, I believe, is completely tolerable.
No public API changes are necessary, sequential processing should not be affected, except an extra field in ArraySpliterator which increases a footprint by 8 bytes. I added a simple test to ensure that at least two threads are actually used when parallelizing Stream.iterate source. More testing ideas are welcome. ------------- Commit messages: - JDK-8280915 Better parallelization for AbstractSpliterator and IteratorSpliterator when size is unknown Changes: https://git.openjdk.java.net/jdk/pull/7279/files Webrev: https://webrevs.openjdk.java.net/?repo=jdk&pr=7279&range=00 Issue: https://bugs.openjdk.java.net/browse/JDK-8280915 Stats: 128 lines in 2 files changed: 96 ins; 0 del; 32 mod Patch: https://git.openjdk.java.net/jdk/pull/7279.diff Fetch: git fetch https://git.openjdk.java.net/jdk pull/7279/head:pull/7279 PR: https://git.openjdk.java.net/jdk/pull/7279
