Checked out and built Mahout using IntelliJ and it was completely
painless. Doing the same with Eclipse (Indigo) and following our wiki
instructions to install m2e was not. The 1.0.100 version of m2e only
allowed me to check the integration subproject. This is not unique to
the particular version of m2e; with other versions I've seen similar
erratic performance. Even when it does finally install correctly, it
doesn't work very well, losing SVN connectivity, etc. I really hate it.
I am able to run mvn eclipse:eclipse and, using the .classpath file it
created in examples/, am able to create a single Eclipse project with
only three glitches:
* Trunk has two copies of
o/a/m/cf/taste/grouplens/GroupLensRecommender.java: one in
integration/ and one in examples/.
* Trunk has two copies of
o/a/m/cf/taste/grouplens/GroupLensDataModel.java: also in
integration/ and examples/.
These files are duplicates, the only difference is the order of imports.
Can I delete the examples/ copies?
* Trunk has two copies of o/a/m/classifier/bayes/package-info.java:
one in core/ and one in examples/. Actually there are many such
files which contain no actual Java artifacts (there's even one in
Hadoop).
These files seem to be related to package annotations, but do we need
them in both locations? If I remove the one in examples/ my project
compiles.
I'd switch to IntelliJ except for the huge effort in learning a new IDE.
Also, many corporate R&D labs (all the ones I've worked for in the last
10 years) still use Eclipse so it seems like it ought not to be so
painful. Did we paint ourselves into an IDE corner on purpose?
