We would only have to bump to 1.7 http://docs.scipy.org/doc/numpy-dev/reference/c-api.array.html#c.PyArray_SetBaseObject

attn @mdboom @jkseppan

I don't have any objections to dropping numpy 1.6 support.

We should probably add the magic incantation that enforces the Numpy 1.7 API if we do this so it doesn't regress again, but I don't know how many changes would be required... That can wait for another PR, of course.

If the fix is as simple as https://github.com/scipy/scipy/blob/master/scipy/signal/lfilter.c.src#L285-L292 I think we should just apply it. Is there a simple test case? I don't regularly use the gdk backend.

Oh, I guess it would be a compile-time error so the test case is easy if I can otherwise get gdk to compile?

Yeah -- I think if we can install the gtk development packages on Travis, this extension would get built and then we'd at least catch simple compilation errors.

I think the fix would be something like jkseppan@4ab9919 but I didn't even try to compile it yet.

Unfortunately it's not a compilation error: it's just

    src/_backend_gdk.c:55:5: warning: implicit declaration of function ‘PyArray_SetBaseObject’ [-Wimplicit-function-declaration]
         PyArray_SetBaseObject(array, (PyObject *)py_pixbuf);
         ^

But import matplotlib.backends.backend_gdk triggers the error for me.

Would appreciate review of #5080, especially by someone who uses backend_gdk actively.

Probably easier to do a shout-out on the mailing list for this? Not so many people get alerted to issues/PRs...

I mentioned it at http://permalink.gmane.org/gmane.comp.python.matplotlib.devel/13729