Improved performance of @UniqueElements(by=NOT_SET)`

[DirkToewe]

Overview

This PR aims at significantly improving the performance of @UniqueElements(by=NOT_SET), mainly by replacing calls to uniqueElements(x->x) by uniqueElements() wherever possible.

Details

If we look at the following two property tests:

@PropertyDefaults( tries = 100_000 )
public class UniqueElementsTest
{
  @Property void testA( @ForAll @UniqueElements int[] a ) {
    Arrays.sort(a);
    for( int i=1; i < a.length; i++ )
      assert a[i-1] <= a[i];
  }

  @Provide Arbitrary<int[]> arrays() {
    return Arbitraries.integers().array(int[].class).uniqueElements();
  }

  @Property void testB( @ForAll("arrays") int[] a ) {
    Arrays.sort(a);
    for( int i=1; i < a.length; i++ )
      assert a[i-1] <= a[i];
  }
}

On my machine, it takes 20 seconds for testA to finish while testB only takes 2 seconds. The problem becomes much more emphasized with larger array sizes.

---

I hereby agree to the terms of the jqwik Contributor Agreement.

[vlsi]

This is a suboptimal way to check uniqueness. You can stop as soon as you detect the first non-unique element.

At the same time, elements in hashset are always unique, aren't they?

[DirkToewe]

Oh You're right, it would be much faster to stop early. I will fix that.

Yes that's why added the HashSet clause. It might speed things up sometimes.

[DirkToewe]

Changed the logic. Now it should stop early.

[vlsi]

Nice. There's one more case doing stream.distinct.count:

jqwik/engine/src/main/java/net/jqwik/engine/properties/FeatureExtractor.java

Line 32 in 124d7af

long uniqueCount = elements.stream().map(this::applySafe).distinct().count();

I wonder if it makes sense to unify them.

WDYT?

[jlink]

@DirkToewe As soon as you deem it ready, just rebase it, so that I can merge it.

[DirkToewe]

I did a soft reset and re-committed the changes. All merge conflicts should be gone now.

The PR now also includes the change to FeatureExtractor::areUnique suggested by @vlsi. It felt like it fits the theme of the PR. I hope that's okay.

[jlink]

Published in 1.7.4-SNAPSHOT