As Scala 2.11 is the default for 2.4.x, we currently include _2.11 artifacts in our release. Our Python library depends on the Scala artifacts.
With Homebrew, we have -submit and -shell script wrappers for spark-submit and spark-shell, and those will break at runtime if we're using _2.11 artifacts on Spark built for Scala 2.12. > On Apr 26, 2019, at 11:12 AM, Sean Owen <[email protected]> wrote: > > Yeah I don't think the pyspark change was intentional; I'm trying to help > assess what the impact is though. > > It may be a dumb question, but, what problem does the change cause? is it > beyond what I mentioned below? you have a project with interdependent Python > and Scala components? > > On Fri, Apr 26, 2019 at 11:02 AM Michael Heuer <[email protected] > <mailto:[email protected]>> wrote: > We certainly can't be the only project downstream of Spark that includes > Scala versioned artifacts in our release. Our python library on PyPI depends > on pyspark, our Bioconda recipe depends on the pyspark Conda recipe, and our > Homebrew formula depends on the apache-spark Homebrew formula. > > Using Scala 2.12 in the binary distribution for Spark 2.4.2 was unintentional > and never voted on. There was a successful vote to default to Scala 2.12 in > Spark version 3.0. > > michael > > >> On Apr 26, 2019, at 9:52 AM, Sean Owen <[email protected] >> <mailto:[email protected]>> wrote: >> >> To be clear, what's the nature of the problem there... just Pyspark apps >> that are using a Scala-based library? Trying to make sure we understand what >> is and isn't a problem here. >
