[OPENNLP-1466] Convert the use of Collection#toArray(..) more performant version under modern JVMs

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.1.0, 2.1.1
• Fix Version/s: 2.2.0
• Assignee: Martin Wiesner
• Reporter: Martin Wiesner
• Created: 17/Feb/23 21:27
• Updated: 22/Apr/23 17:39
• Resolved: 25/Feb/23 22:12

Description

In older Java versions, using a pre-sized array was preferred, as the reflection call necessary to create an array of pre-computed size was rather slow. However, since Java 6+, the performance of the empty array version the same, and sometimes even better. [1] [2]

The code of many OpenNLP classes still relies on the old style of converting collections to arrays, via someCollection.toArray(...size()) or similar, in essence: via a pre-computed length. However, the zero-sized initialization is considered (and demonstrated) to be faster under Java 11+ vms. [3]

Aims

• Migrate the old  way of creating / converting to arrays to the more performant version.
• Ensure all tests hold no drawbacks show up.

 

[1]: https://shipilev.net/blog/2016/arrays-wisdom-ancients/#_new_reflective_array

[2]: https://www.baeldung.com/java-collection-toarray-methods#performance-trials

[3]: https://www.baeldung.com/java-collection-toarray-methods#benchmarks-on-newer-jdks