test-jar is generally evil. I'm sure that remark is not especially helpful, but pretty much everyone who touches it that I've ever heard of regrets it.
Practically for mahout, you can't grab a source tree and to mvn -Pfastinstall. You have to sit through running all the tests. Now, I could fix fastinstall to use the finer-grained options that keep test compilation and just suppress test running, I confess. But my first instinct was to squash the use of test-jar altogether. If you all cluster 'round and express a preference in the direction of keeping test-jar, I'll make the other fix. And, yes, MahoutTestCase may be the singular item that causes all this grief. Unfortunately, since it extends 'Assert', my 'proposal (3)' doesn't work. We need a jar file that contains RandomUtils, and then we need a jar file that contains MahoutTestCase. Ouch. Even I am beginning to think that the profile fix is preferable. On Thu, Apr 7, 2011 at 4:19 PM, Sean Owen <[email protected]> wrote: > What's the problem being solved here? it all works OK to me. > My guess is there's not a need to create a super-core nameless toplevel > module just for this. Are we talking about MahoutTestCase? > > On Thu, Apr 7, 2011 at 4:27 PM, Benson Margulies <[email protected]> > wrote: >> >> The use of test-jar leads to problems, notably that you can't build >> the complete tree from clean without getting tangled up with tests. I >> propose to make more directories in the top level to contain the >> common test classes that are currently in the test-jars. This is two >> new dirs, one for math and one for core. >> >> Do people want prior review of this on a JIRA? > >
