Save GDT20 - Polar Stereographic - take 2

[trexfeathers]

Closes #122

I believe the outstanding requests have been addressed:

• Save GDT20 - Polar Stereographic #122 (comment)
  • ff8fc0d
  • 8f735bf
  • 756bb56
• Save GDT20 - Polar Stereographic #122 (comment)
  • 26eb55a
• Save GDT20 - Polar Stereographic #122 (comment)
  • 36dd8ef
  • 6b0d2a1

[CLAassistant]

All committers have signed the CLA.

[trexfeathers]

@pp-mo I'm looking at this comment - #122 (comment):

I'm thinking that at least one of NorthPolar/SouthPolar will currently fail a save-then-load roundtrip check.
That's a good reason for adding one before merging this.

Could you offer some guidance on this? There is a new integration test but I suspect that's not what you meant.

[pp-mo]

@pp-mo I'm looking at this comment - #122 (comment):

I'm thinking that at least one of NorthPolar/SouthPolar will currently fail a save-then-load roundtrip check.
That's a good reason for adding one before merging this.

Could you offer some guidance on this? There is a new integration test but I suspect that's not what you meant.

I think what this meant is that we should also test with Iris iris.coord_systems.PolarStereographic coordinates systems -- i.e. as well as iris.coord_systems.Stereographic.

In CF terms, a "polar_stereographic" grid mapping must have latitude_of_projection_origin = +90 or -90. It's effectively just a special case of "stereographic", but I guess the polar case was put in first so they needed to keep it.
Iris completely follows CF in this, in having a separate PolarStereographic coordinate system type, presumably so that we can sensibly distinguish "polar" ones, but the PolarStereographic class actually inherits from the Stereographic class, and only adds a distinct type and a modified constructor -- no different behaviour.

So, in terms of iris-grib, I think my old comment meant that data with an Iris PolarStereographic coordinate system (either north or south) should be handled by iris-grib, being treated as a generic 'Stereographic' case. But it's saying that that "would have failed", at some point, for some reason.

So it's probably still worth doing a roundtrip cube-grib-cube test on both North and South cases, and in each case when presented as either a PolarStereographic or Stereographic Iris coordinate system.
Though obviously when loading back, you will always get a PolarStereographic Iris coordinate system (since AFAICT GRIB does not seem to provide a more general case).

[pp-mo]

when loading back, you will always get a PolarStereographic Iris coordinate system

Actually no, to my surprise.
we do it like this
So, we always construct a "generic" Stereographic coord-system. I don't know if there is a particular reason for that.

[pp-mo]

A couple of suggestions, but looks very comprehensive to me. Practically good to go.

[trexfeathers]

Thanks @pp-mo 😊