[PROJ] included tests and grids

[Greg Troxel]

Sebastiaan Couwenberg <sebastic at xs4all.nl> writes:

> On 12/30/19 3:56 AM, Greg Troxel wrote:
>> Are tests of 6.2.1 expected to pass?  With the following grids, or
>> something else?
>
> The tests are expected to pass with only the grids from proj-datumgrid
> available, when also adding grids from the regional archives you should
> expect failures like these.

Thanks for explaining that to me (again!).

All:

I have added to my in-progress proj6 package the notion of an option for
grids, and added a faux test target to fail when the option is enabled.

I understand how we got here, but the combination of it being good for
users to have the grids (assuming they have the disk space, which I
think most do) and it being bad to have grids for testing means that
it's hard to run the tests.  (Yes, I realize one could split the
packages.)

So it would be nice if either the tests expected what seems to be the
standard set of extra grids (i.e., what is listed at
https://proj.org/download.html under "Current Release"), or somehow
avoided reading things that aren't part of base.

To me, this is part of a larger repeatability issue.  I understand that
there are transforms of different accuracy, but the computer scientist
in me wants the property that "install proj, do transform x" will either
get a consistent answer or fail.  But I realize there are lots of other
opinions, so I'll say this briefly and try not to do so more than once
a year.