[R-sig-Geo] R 3.0.0 and spatial classes

[Roger Bivand]

On Sun, 16 Dec 2012, Barry Rowlingson wrote:

> There's been an announcement that the next version of R will be called
> 3.0.0. Although not a massive change (compared to Python 2.x and 3.x,
> or Perl 4 to Perl 5),

While I won't comment on the ideas discussed in this thread (now), I think 
that it is worth noting that the 2.15 -> 3.0 is no more than the Linux 
kernel shift to 3.*, there were simply getting to be too many minor 
versions in a system that had migrated a long way from 2.0.

One set of 2/3 changes has already been addressed. It is crucial to 
maintaining stability in R, and imposes a ban on calling compiled code 
across contributed package boundaries. We recently saw an example between 
rgdal and raster, where my changing the number of arguments passed to a C 
function in rgdal broke raster, which was calling directly to rgdal. All 
the rgdal C functions called from raster now have R wrappers, so that 
raster can call them safely. There are very many cases like this across 
package-sphere, and they make the maintenance of CRAN hard. It is in these 
kinds of developer-facing issues that we'll see differences, not in 
user-facing R.

The distinction between user and developer should however be tempered; I 
personally find user's wishes frustrating (sometimes very frustrating). If 
users have wishes, then the first alternative should be to try to 
contribute something better, preferably in terms of working code. Once the 
user has established that things aren't so obvious, and/or involve design 
choices, any dialogue becomes more fruitful. "Simple" in a non-simple 
conceptual setting is not necessarily as simple as it sounds, in my 
experience.

Hope this clarifies,

Roger

> it might be a good opportunity to revise all the
> spatial classes in order to:
>
> * clear out any cruft
> * remove any inconsistencies
> * add some new functionality
> * unify across spatial packages
>
> The first thing that springs to mind, for example, is getting sp and
> raster to use the same functions for coordinate reference system
> processing. Doubtless there are other opprtunities for synergy...
>
> Yes, this may well break existing code, but if R is going to jump
> from 2 to 3 then that will break existing code too.
>
> Your ponderances, please...
>
> Barry
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Department of Economics, NHH Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no