garydgregory commented on PR #1524: URL: https://github.com/apache/commons-lang/pull/1524#issuecomment-6921973990
Thank you for the proposal and the benchmarking work, @IcoreE. I’m closing this PR because the collision examples reported in [LANG-1802](https://issues.apache.org/jira/browse/LANG-1802) are already addressed by the current implementation and covered by existing regression tests: ```java return Objects.hash(end, negated, start); ``` This implementation satisfies Java’s `equals()`/`hashCode()` contract. Collisions between unequal objects are permitted; the practical question is whether a change improves collection performance. Unfortunately, the proposed bit packing introduces poor bucket distribution for single-character ranges. For `CharRange.is(c)`, the hash is `(c << 16) | c`. After the JDK’s hash spreading, this becomes `c << 16`, placing all such ranges in bucket zero for table capacities up to 65,536. This affects `CharSet`, which stores ranges in a `LinkedHashSet`. A focused reproduction with 26 lowercase letters used 26 buckets with the current implementation, but only one tree bucket with this change. The benchmark demonstrates faster calculation of the proposed formula, but it does not measure the resulting collection behavior. It also benchmarks separate implementations of the formulas rather than calling `CharRange.hashCode()`, and adds no JUnit regression assertions. Given that the reported examples are already handled and this replacement introduces a bucket-distribution regression, I think retaining the current implementation is the appropriate choice. A future optimization supported by representative collection benchmarks and regression tests would be welcome. Thank you again for investigating this. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
