The Avro project faced this same issue and they resolved it in Avro 1.7.3. Afaik, they used maven classifiers. Thus, an avro-mapred jar compiled against hadoop2 is avro-mapred-1.7.3-hadoop2.jar. See https://issues.apache.org/jira/browse/AVRO-1170. - Wing Yew
On Fri, Mar 29, 2013 at 8:50 AM, Stack <[email protected]> wrote: > (cc'ing bigtop dev in case the bigtoppers have an opinion) > > Long story short, we need to publish two artifacts up to the maven > repository: one for hadoop1 and one for hadoop2. Maven doesn't let you do > this 'naturally'; a maven build produces one artifact only. The only > dimension left to us to pervert is version (Maven 'classifiers' can not be > pulled around to serve this purpose after some trying). So, I'm suggesting > that our version be 0.95.0-hadoop1-SNAPSHOT instead of 0.95.0-SNAPSHOT; > i.e. we include the gross hadoop version the jar was built against. The > jars we publish will look like: hbase-client-0.95.0-hadoop1-SNAPSHOT, or > hbase-client-0.95.0-hadoop2-SNAPSHOT, and so on. > > The above is ugly but it makes it clear what version of hadoop the jar was > built for. > > Opinions? > > Thanks, > St.Ack
