Fix for netcdf4 returning masked array with no masked points.

[pp-mo]

Alternative solution to #3179

Hi @corinnebosley
Let's see how this fix works out ?

[pp-mo]

Let's see how this fix works out ?

It seems that some existing cases are managing to load masked data with empty masks.
I think we can just fix those tests ... will look into these tomorrow.

[corinnebosley]

@pp-mo You've got a bunch of CMLs to regenerate (which I don't see a problem with), but that's all I can see as far as test failures are concerned.

[pp-mo]

a bunch of CMLs to regenerate (which I don't see a problem with)

I will also first check those out, as they don't fit my understanding of what netCDF "used to do".

[pp-mo]

a bunch of CMLs to regenerate ... check those out, as they don't fit my understanding of what netCDF "used to do".

Ok, I think I have grasped it now.
All these CDL files with mask_checksum="no-masked-elements" in them already changed before, when we upgraded for netcdf 1.4 : that occurred in #3031 just before we cut iris 2.1.0.
So the results of these operations (whether you get a masked array) already changed between v2.0 and v2.1.
And now we are proposing to change them all back.

Now I just need to form a view on whether that is a good idea (as some users have requested), or not ...

[pp-mo]

Discussed off-line with @dkillick @pelson @bjlittle.
Concluded : Probably a bad idea -- instead, let's follow netcdf behaviour + pursue performance improvements in other ways if+when required

Withdrawn !