While trying to bring up hadoop 0.21 from source, I noted that there is a bunch of duplicate files between the projects. The config directory for example. Also hadoop-mapreduce seems to build webapps/datanode for some reason (?!).
Someone on irc mentioned using 'ant tar', but it seems to be currently broken, since it (a) requires forrest and (b) forrest fails to build the docs on OSX 10.6.1, complaining about some datatype nonsense. What is the new standard for building a distribution?
