Cloudera recently released mrunit which is a unit testing framework for hadoop. http://www.cloudera.com/blog/2009/07/03/debugging-mapreduce-programs-with-mrunit/
I see the upside of being able to specify inputs and output in the source code. Does this approach make sense for hive? Hive does have some complex situations to deal with. For example, the raw output of an explain or parse probably is better off in its own file, but when dealing with a small number of rows of tabular output working with the outfiles and diff having the input and expected output inline might be nice. Thoughts? Edward
