[PROJ] Big grid files

[Greg Troxel]

Javier Jimenez Shaw via PROJ <proj at lists.osgeo.org> writes:

> What I did is using int16. The max error is 1.2 mm (if I did the maths
> correctly). The size with that change is about 24 MB
> If anybody wants even more accuracy, they can use int32. The reduction is
> smaller, but still less than 100 MB.

It's hard to believe that there is true accuracy to 1 mm; that seems
state of the art for multi-year observations determining HAE at a
national reference station.  But it might be nice to numerically match
NGS tools ~exactly as a check.

Perhaps having makefile support to create the int32 version, and not
putting it in tarballs.  It could be hosted elsewhere; it feels like we
are running up against github limitations that perhaps should not
control how things are, but doing something different may be
hard/expensive.

> (any benefit between uint16 vs int16? the positive and negative ranges are
> almost the same, so I do not need offset with int16)

Using a signed type for a signed value seems like a win in terms of
reduced cognitive load.