I wonder if it’s possible to take the long awaited refactoring of mrlegacy to split Mahout into two builds one Scala only (with a little math java) and even use SBT for Scala leaving a pared down mvn for mrlegacy build? This would greatly speed and simplify builds and is a very natural divide.
Spark has several sub-projects and this would give us wiggle room for using new tools and techniques. We could even name the separate builds with more descriptive sub-project names.