+1 from me. Just a quick thought... do we have any similar concerns with the contouring code? If one passes in a fortran ordered image, would the code still work properly?

Cherry-picked to master as 2077a46

Thanks @ianthomas23 !

@WeatherGod: I've checked contouring and it is fine.

Gah, this breaks the Travis on the python 2.6 build because it uses numpy 1.6 which doesn't support the order keyword on np.copy (it only came in on 1.7 I think). I don't know why Travis missed that environment variable off when it tested this PR.

Do we still support numpy 1.6? If so then we can replace:

np.copy(a, order=order)

with

np.array(a, copy=True, order=order)

@OceanWolf You are right. This wasn't discovered because we don't test with Numpy 1.6 on the maintenance branch. The changes to Travis that specifically enabled Numpy 1.6 (with Python 2.6) have not been back ported to that branch. I think your solution is fine. AFAIK we would like to continue supporting Numpy 1.6 since it is the default version on a number of Linux distributions

@OceanWolf Can you put in a PR with that change?

Is there any plan on the np side to deprecate the copy kwarg to array?

Okay, I now see that this got merged into color-overhaul which doesn't have that environment option, though I would have thought a separate Travis check should happen before merging into master, as they have different criteria for passing (not just numpy version).

@tacaswell not quite sure how to do this, although the actual code fix looks straight forward enough, it goes against two different branches, I imagine that the color-overhaul branch should get the Travis numpy1.6 check backported to prevent similar incident.

I have no idea if numpy has a plan to deprecate the copy kw, when I saw the problem on Travis, I just turned to the docs for 1.6 http://docs.scipy.org/doc/numpy-1.6.0/reference/generated/numpy.copy.html which says:

numpy.copy(a)

Notes

This is equivalent to

>>> np.array(a, copy=True)                              

I would hope numpy don't deprecate this until 1.6 has disappeared from the various distros.

@OceanWolf Yes in principle the code should be tested on both branches before merging but that requires opening a new PR against the other branch. Back porting the changes to Travis makes sense

I don't see anything in the numpy docs about deprecating copy=True so I think that is fine.

The easiest is to open a PR with the fix against color-overhaul and we can then cherrypick the change onto master.

The copy kwarg is fundamental to the array() function; I can't imagine it ever going away.