The resources aren't "code" but people and time. You have a limited set of folks who consistently contribute and become reviewers/committers. This is all based

The resources aren't "code" but people and time. You have a limited set of folks who consistently contribute and become reviewers/committers. This is all based on volunteer time - no one is paying these folks to do it. So, asking these folks to divide their limited volunteered time between multiple versions of python is unfair. I think this is the right decision to take. If you feel you have "plenty of resources" you can fork the python2 version of numpy and maintain it.