I've almost suggested the same thing while looking at a couple of the issues Patrick has raised. Too much has changed / is still changing between the hadoop 1 and 2 APIs. It's pretty much impossible already to cater to both and I don't see the situation improving. We already produce 2 artifacts, so we might as well produce them from different branches. The end result for the user will be no different.
Big +1 from me. Cheers Dave On 21 December 2012 19:40, Jarek Jarcec Cecho <jar...@apache.org> wrote: > We had recently similar incompatibility fight in Sqoop, so I can > understand your frustration. > > I'm +1 on the proposal. > > Jarcec > > On Fri, Dec 21, 2012 at 01:29:19PM -0600, Brock Noland wrote: > > Hi, > > > > In MRUNIT-31 we used mockito to allow a single branch of code work > > with hadoop 1 and hadoop 2. Patrick Hunt has filed a number of tickets > > for items which don't work with hadoop 2. I do not see an easy way to > > fix MRUNIT-167 and MRUNIT-161 in a single branch since hadoop 1 is > > lacking a large number of methods that hadoop 2 supports. > > > > As such, I think that we should think about the following: > > > > 1) Create hadoop-1 to support hadoop 1 with trunk supporting hadoop 2. > > 2) Resolve MRUNIT-167 and MRUNIT-161 in trunk. > > 3) Going forward we will have to backport from trunk to the hadoop-1 > > branch. At first this will be simple since they are so similar. Over > > time this will become more difficult. Hopefully by the time it's too > > difficult to support hadoop-1 most people will have already moved to > > hadoop 2. > > 4) During releases we'll have to generate artifacts from both branches. > > > > Thoughts? > > > > -- > > Apache MRUnit - Unit testing MapReduce - > http://incubator.apache.org/mrunit/ >
