new methods to optionally re-enable old numpy array type behaviour (issue #809)

[davidhassell]

No description provided.

[CLAassistant]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

David Hassell seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[jswhit]

Looks good, thanks @davidhassell.

Just needs a Changelog entry.

[akrherz]

Just a comment from the peanut gallery here and I know API design is hard, but the name set_auto_array_type does not seem intuitive to me. This setting is controlling whether variables should always be numpy masked arrays or not. Currently set_auto_mask(self, True_or_False) exists, it seems like a companion to this would be set_always_mask(self, True_or_False) ?

Just my unqualified two cents here.

[davidhassell]

Sounds good. set_always_mask is better than what I had, I think. I'll give it a while and then update the pull request if there are no objections.

[davidhassell]

@jswhit, are you happy to rename the methods, as per @akrherz's suggestion?

[jswhit]

Not quite clear on what set_always_mask would do. Would it be True but default, and to enable the old behavior it would be set to False?

[jswhit]

as per the discussion in issue #808, please go ahead and change

            # issue #785: always return masked array, if no values masked
            # set mask=False.
            data = ma.masked_array(data,mask=False,fill_value=fill_value)

to

            # issue #785: always return masked array by default, if no values masked
            data = ma.masked_array(data)

[davidhassell]

Yes - The logic needs reversing when we rename to set_auto_array_type to set_always_mask. The default would be True, signifying that masked arrays are returned in every instance. I'll add this.

[jswhit]

Looks like tst_multifile.py is failing somewhere in cftime. Restoring the 'mask=False' fixes it.

[davidhassell]

I suggest, then, that we restore mask=False here and revisit it in #808 (i.e. see if we can fix the downstream issues in cftime as pull request on that ticket (which I'm happy to look at, if you like)). I'll put mask=False back in this ticket's PR. Thanks