In 0.13.1 we had one binary tarball. A full spread would look something like this in 0.13.2-
Spark-1.6, Scala-2.10 Spark-2.0, Scala-2.10 Spark-2.1, Scala-2.10 Spark-1.6, Scala-2.11 Spark-2.0, Scala-2.11 Spark-2.1, Scala-2.11 Spark-1.6, Scala-2.10, viennacl Spark-2.0, Scala-2.10, viennacl Spark-2.1, Scala-2.10, viennacl Spark-1.6, Scala-2.11, viennacl Spark-2.0, Scala-2.11, viennacl Spark-2.1, Scala-2.11, viennacl Spark-1.6, Scala-2.10, viennacl-omp Spark-2.0, Scala-2.10, viennacl-omp Spark-2.1, Scala-2.10, viennacl-omp Spark-1.6, Scala-2.11, viennacl-omp Spark-2.0, Scala-2.11, viennacl-omp Spark-2.1, Scala-2.11, viennacl-omp Spark-1.6, Scala-2.10, viennacl, viennacl-omp Spark-2.0, Scala-2.10, viennacl, viennacl-omp Spark-2.1, Scala-2.10, viennacl, viennacl-omp Spark-1.6, Scala-2.11, viennacl, viennacl-omp Spark-2.0, Scala-2.11, viennacl, viennacl-omp Spark-2.1, Scala-2.11, viennacl, viennacl-omp That's 24 tarballs of pre-compiled binaries. The main thing I'm concerned about is getting all combos of spark/scala, viennacl/scala, viennacl-omp/scala into Maven repositories. This can be accomplished with 6 tarballs: Spark-1.6, Scala-2.10, viennacl, viennacl-omp Spark-2.0, Scala-2.10, viennacl, viennacl-omp Spark-2.1, Scala-2.10, viennacl, viennacl-omp Spark-1.6, Scala-2.11, viennacl, viennacl-omp Spark-2.0, Scala-2.11, viennacl, viennacl-omp Spark-2.1, Scala-2.11, viennacl, viennacl-omp Not all users want ViennaCL (I would imagine) - A compromise might be the first and last 6 combinations: Spark-1.6, Scala-2.10 Spark-2.0, Scala-2.10 Spark-2.1, Scala-2.10 Spark-1.6, Scala-2.11 Spark-2.0, Scala-2.11 Spark-2.1, Scala-2.11 Spark-1.6, Scala-2.10, viennacl, viennacl-omp Spark-2.0, Scala-2.10, viennacl, viennacl-omp Spark-2.1, Scala-2.10, viennacl, viennacl-omp Spark-1.6, Scala-2.11, viennacl, viennacl-omp Spark-2.0, Scala-2.11, viennacl, viennacl-omp Spark-2.1, Scala-2.11, viennacl, viennacl-omp Thoughts?
