Github user bavardage commented on the issue: https://github.com/apache/spark/pull/21794 There is a way in which the `nanSafeCompare*` methods do differ from java built-in, but that wasn't captured in the test cases (or in the suggestive naming of the method), namely the handling of `-0.0` vs `0.0`. `java.lang.Double.compare(-0.0d, 0.0d) == -1` `Utils.nanSafeCompareDoubles(-0.0d, 0.0d) == 0` If that behaviour is indeed required, I could abandon this PR and instead fix the test cases to capture this.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org