+1 from me. The discussion on #168 is the relevant background information.

Not having it look for fink and macports packages is fine - if you have one of those you can just use their matplotlib packages.

@jdh2358: I have been able to test this successfully on OSX 10.7 with both homebrew and apple Python 2.7. @r-owen has tested on 10.6. Do you know who else should see this before I commit?

@efiring: I know you were involved in the original discussion. Do you have any objections to this change?

No objection. Let's try it and see what happens.

+1 (though cannot test now, travelling)

I think the big reason for recommending make.osx is all the build problems you get when setupext picks up incompatible versions of libraries, so reducing the number of directories it looks in should help. I think if /usr or /usr/X11 has a version of freetype that's incompatible with /usr/bin/python, it's an Apple bug, but people could have installed whatever libraries in /usr/local (perhaps indirectly - .pkg installers can leave stuff all over your system). Does the output from "python setup.py build" indicate which directories each dependency is found in? That information could help when people report build problems.

Does the output from "python setup.py build" indicate which directories each dependency is found in?

No. But it looks like the setupext could be made to do so with a bit of effort if we did want that ability. I'm going to merge this issue, but will open a subsequent issue for getting the path info for some of the build dependencies.