Speed improvements

[gnodet]

No description provided.

[olamy]

not sure what it is improving?

[gnodet]

The long answer is here, the short one is that interfaces.toArray( new Class[0] ) is faster than interfaces.toArray( new Class[interfaces.size()] ) mainly because the JVM does not need to zero the array.

[olamy]

thanks that's interesting read. But it's from 2016 and the conclusion says Future VM optimizations may close this performance gap for toArray(new T[size]) so it could be interesting to have similar benchmark now almost 5 years later :)

[gnodet]

Here's the bench results on a recent JDK 14 JVM. It does not seem the results of the article have changed much, though the difference is quite small...

Benchmark            (size)     (type)  Mode  Cnt     Score     Error  Units
ToArrayBench.simple       0  arraylist  avgt    3     5.854 ±   1.322  ns/op
ToArrayBench.simple       0    hashset  avgt    3     5.641 ±   0.389  ns/op
ToArrayBench.simple       1  arraylist  avgt    3     8.003 ±   0.972  ns/op
ToArrayBench.simple       1    hashset  avgt    3    15.671 ±   3.531  ns/op
ToArrayBench.simple      10  arraylist  avgt    3     8.805 ±   8.413  ns/op
ToArrayBench.simple      10    hashset  avgt    3    28.156 ±   2.709  ns/op
ToArrayBench.simple     100  arraylist  avgt    3    28.388 ±   2.799  ns/op
ToArrayBench.simple     100    hashset  avgt    3   285.236 ±  56.555  ns/op
ToArrayBench.simple    1000  arraylist  avgt    3   288.888 ±  83.145  ns/op
ToArrayBench.simple    1000    hashset  avgt    3  2340.670 ± 651.868  ns/op
ToArrayBench.sized        0  arraylist  avgt    3     6.343 ±   1.680  ns/op
ToArrayBench.sized        0    hashset  avgt    3     5.568 ±   0.957  ns/op
ToArrayBench.sized        1  arraylist  avgt    3    11.708 ±   0.874  ns/op
ToArrayBench.sized        1    hashset  avgt    3    21.645 ±  10.234  ns/op
ToArrayBench.sized       10  arraylist  avgt    3    20.294 ±   0.902  ns/op
ToArrayBench.sized       10    hashset  avgt    3    45.420 ±  11.422  ns/op
ToArrayBench.sized      100  arraylist  avgt    3   129.215 ±  27.252  ns/op
ToArrayBench.sized      100    hashset  avgt    3   472.538 ±  75.969  ns/op
ToArrayBench.sized     1000  arraylist  avgt    3  1164.126 ± 207.183  ns/op
ToArrayBench.sized     1000    hashset  avgt    3  4204.794 ± 407.450  ns/op
ToArrayBench.zero         0  arraylist  avgt    3     4.554 ±   0.187  ns/op
ToArrayBench.zero         0    hashset  avgt    3     4.377 ±   2.012  ns/op
ToArrayBench.zero         1  arraylist  avgt    3    10.969 ±   1.892  ns/op
ToArrayBench.zero         1    hashset  avgt    3    21.890 ±   1.103  ns/op
ToArrayBench.zero        10  arraylist  avgt    3    18.158 ±   1.774  ns/op
ToArrayBench.zero        10    hashset  avgt    3    34.278 ±   4.633  ns/op
ToArrayBench.zero       100  arraylist  avgt    3   113.737 ±  31.070  ns/op
ToArrayBench.zero       100    hashset  avgt    3   420.547 ± 104.867  ns/op
ToArrayBench.zero      1000  arraylist  avgt    3   994.804 ± 400.316  ns/op
ToArrayBench.zero      1000    hashset  avgt    3  3975.771 ± 658.523  ns/op

[gnodet]

@olamy that said, if you don't like the change, I can remove it, it won't impact much on the overall performances one way or another.

[olamy]

oh no I'm fine with that. I never read this blog post and surprised to have to change a pattern I'm using for so many years now 🤣