Kafka has very simple build requirements and a system like Ant is well suited for a clean and concise build. I have an experimental patch that does just this - replaces SBT with Ant+Ivy. IMO, this approach is cleaner, clearer, and more developer friendly.
Dependencies are localized to one directory in the project rather than living in ~/.ivy2 and elsewhere. This makes manual classpath building very simple (just one glob) and also makes packaging the libs very easy.
Testing is done through junit rather than scalatest. The Kafka tests use `org.scalatest.junit.JUnitSuite` which allow the tests to be executed through the junit test runner.
Management of the Scala version is handled through Ivy. The way I have laid out the Ant script, the Scala version can be changed by setting a different runtime property (-Dscala.version=2.8.2). Cross-compilation of the Kafka artifact would be simple to add.
The one downside to this approach is lack of an incremental build. `scalac` is deprecating its incremental build capabilities in coming versions. The suggested solution to this is to use an IDE that supports incremental builds.
The main motivation for this approach, to me at least, is that a developer can look at build.xml and immediately understand what is going on (with the exception maybe of the <ivy: .../> actions which would not be changing). This is largely not true for SBT unless someone is already familiar with SBT.