Add option to hash based on method instead of class

[winhamwr]

For cases with test fixtures, it's likely that splitting by class will result in the best speed. When there are no gains from grouping tests by class, though, this can result in a less-even, and thus slower, distribution.

$ export NOSE_DISTRIBUTED_NODES=2;
$ export NOSE_DISTRIBUTED_NODE_NUMBER=1;
$ export NOSE_DISTRIBUTED_CLASS_HASHING=0;
$ nosetests long_test_suit

[simon-weber]

This is something we'd like, too =)

Any thoughts since you suggested this? The interface you described should be easy enough to implement (though I think it'd need to fall back to the test's file if the class isn't available, eg a nose test function or generator).

[simon-weber]

Oh, maybe things changed at some point? Currently tests aren't distributed by class, but individually; I'm looking for distribution by class.

[simon-weber]

though I think it'd need to fall back to the test's file if the class isn't available

actually, it's probably best to just keep the current fallback -- I doubt many people use file-level setup and teardown.

[winhamwr]

Currently tests aren't distributed by class, but individually; I'm looking for distribution by class.

It would be wonderful if we could switch between per-method and per-class distribution. Pull requests definitely welcome :)