spark_connect() should handle openjdk 11 gracefully

[sz-cgt]

Error:

sc <- sparklyr::spark_connect(master = "local[*]")
#> Error in validate_java_version_line(master, version): Java version detected but couldn't parse version from: openjdk version "11.0.2" 2019-01-15

^{Created on 2019-02-20 by the reprex package (v0.2.1)}

Root Cause:
The version string for openjdk 10 and above includes a release date. Starting with openjdk 11.0.2, that release date is in 2019. validate_java_version_line() fails to handle dates in 2019 (and presumably earlier release dats with a 9 in the month or the day) due to the following code

splatVersion <- splat[grepl("9|[0-9]+\\.[0-9]+\\.[0-9]+", splat)]

if the intention of that regexp is to match jdk 9 versions strings, it should either have a caret (^) before the 9 or something after it to prevent matches against a ISO date string.

Given that java 9 is end of life, a better answer is simply to remove the 9| from the regexp entirely

[zero323]

@sz-cgt At the moment (Spark 2.4) running Spark on JDK9 or later is not supported.

AFAIK SPARK-24417 Build and Run Spark on JDK11 - serves as an umbrella ticket for different JDK compatibility issues, and informally supporting new JDKs should happen in 3.0 (time for Apache Spark 3.0?, and target versions for remaining sub-tasks).