rotated geodetic

[bblay]

For SciTools/iris#67

[pelson]

The way this has been designed, the ellipse is not propagated as is the case for the superclass. I think it needs to be.

[pelson]

@bblay : There is an outstanding action on you before this can be merged. (handle ellipse and datum better)

[pelson]

Seems strange that I should set the ellipse keyword, but it only kicks in if I provide proj params.
Worse, I don't think we should be exposing the user to the proj4_params - it is an implementation detail that we are using proj.4 under the hood.

[rhattersley]

Seems strange...

Indeed. It looks a bit like a square peg has been driven into a round hole. Supplying the proj4_params argument completely circumvents the definition of the Geodetic class! One could now do: crs = Geodetic({'proj': 'tmerc', ...}) and end up with a "geodetic" CRS which is actually a transverse Mercator!

The way to avoid this would be to allow only the parameters relevant to the rotated geodetic, e.g.:

class Geodetic
    def __init__(self, pole_longitude, pole_latitude, ellipse, datum)

But now the Geodetic class looks just like the RotatedGeodetic class! Which highlights the need for either:

1. Make Geodetic a subclass of RotatedGeodetic.
2. Get rid of the inheritance between Geodetic and RotatedGeodetic.

Out of those two I'd rather have option (2). It gives us what we need with the minimum of complexity.

[pelson]

Obviously, I have some concerns about the implementation details with regards to this PR, but I think this is a great new feature and I am very eager to merge! Thanks fro your work @bblay!

[bblay]

Thanks, doing these actions.

Why are Geodetic and Geocentric in _crs.pyx rather than crs.py?
I suppose we'd prefer RotatedGeodetic to live in the same place as those classes?

[rhattersley]

Why are Geodetic and Geocentric in _crs.pyx rather than crs.py?
I suppose we'd prefer RotatedGeodetic to live in the same place as those classes?

I don't think there's a reason they have to be in _crs.pyx, I think it's just because they're so fundamental and because they're used (slightly naughtily, one might argue) within the CRS class. The RotatedGeodetic class isn't fundamental, so I've no problem with it living in crs.py.

[rhattersley]

And this one.