Github user vanzin commented on the issue: https://github.com/apache/spark/pull/20809 > This is on travis and no SPARK_HOME is set. That sounds a little odd. If that is true, then your proposed code wouldn't work either, since it requires SPARK_HOME to be known. In any case, there are two calls to `getScalaVersion()`. First is: ``` boolean prependClasses = !isEmpty(getenv("SPARK_PREPEND_CLASSES")); boolean isTesting = "1".equals(getenv("SPARK_TESTING")); if (prependClasses || isTesting) { String scala = getScalaVersion(); ``` And your code shouldn't be triggering that, since both env variables are for Spark development and other applications shouldn't be using them. Second call is a little later: ``` String jarsDir = findJarsDir(getSparkHome(), getScalaVersion(), !isTesting && !isTestingSql); ``` Here `getScalaVersion()` is only needed when running Spark from a git clone, not from the distribution package. So the right thing would be to move `getScalaVersion()` to `CommandBuilderUtils`, and call it from `findJarsDir` only if needed.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org