[PROJ] NTv1/NTv2/CTable2 and west-oriented longitude shifts

[Even Rouault]

Jochem,

> I agree. Please, abandon the strange conventions of NTv2 and use east
> positive, normal column ordering and preferably also decimal degrees
> instead of arcseconds.

For the NTv2 to GeoTIFF conversion script, I'll keep arcseconds, so that we 
can "trivially" check the strict binary equivalence. But degree/radian/arc-
second will all be possible.
 
> I would prefer very accurate (0.1 mm) grid shift files and an encoding that
> makes it simple to write and read it. To me this is more important than the
> best possible data compression. So if possible, use signed integers with no
> offset (0 is no correction) and 1*10^-9 degree as (default) unit.

For the NTv2 to GeoTIFF conversion script, I'll keep IEEE-Float32. As you see 
value for the encoding you suggest, I will also add support for reading 
signed/unsigned Int32 with offset/scale.
 
> By the way, GeoTIFFs also have a map projection. I assume you are planning
> to use latlon thus platte-carree projection. It would be interesting to
> also allow other projections (with 1*10^-4 metre as (default) unit).
> Although this is against ISO, it would make it possible for countries that
> use horizontal grid shifts on projected coordinates to implement their
> transformation in PROJ., so they don't have to redefine their national
> coordinate system like the Netherlands did to accomplish this.

The GeoTIFF encoding by itself would indeed allow to use CRS other than 
geographic one, but this will not be in the scope of my RFC-4 work. Further 
work in PROJ will be necessary.

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com