[STATSGRASS] Re: [GRASS5] CygWIN: compilation of current CVS HEAD?

[Roger Bivand]

Further to my comments on compiling the R/GRASS interface under Windows, 
and after trying various routes, I'm inclined to think that Glynn is 
right. I think libgis.a and libdatetime.a really ought to be compiled 
using the R build train (MinGW) - at present for a fresh CygWin install 
there are version differences that raise errors when the DLL is being 
created. When compilation works, it is probably more luck than design. I'm 
going to look at trying to build an R binary package for Windows using 
libgrass - has anyone tried this? I'm moving this thread to statgrass, as 
it is more relevant there.

Roger


On Fri, 7 Feb 2003, Roger Bivand wrote:

> On Fri, 7 Feb 2003, Glynn Clements wrote:
> 
> > 
> > Roger Bivand wrote:
> > 
> > > > If there exists a choice between Unix and Windows versions of R,
> > > > choosing the Unix version is likely to result in better compatibility
> > > > with GRASS.
> > > 
> > > R on Windows will not build under CygWin. On the Windows platform, MinGW
> > > is the chosen build train, following which R runs natively. One can
> > > complain that it ought to be CygWin, but the reaction is to invite the
> > > complainer to become R Windows maintainer.
> > 
> > I'm just pointing out that the only way to use GRASS on Windows is via
> > Cygwin. If R can't use Cygwin, and GRASS can only use Cygwin, that
> > suggests that the GRASS/R combination may be inherently problematic.
> > 
> If the CygWin gcc compiled lisgib.a makes other assumptions about its 
> environment (compile train setup etc.) than the MinGW compiled grassR.dll 
> to be dynamically loaded into R, yes, but experience (limited) so far 
> suggests that it is not a major issue. If there were major differences 
> between the two gcc header file trees, you would get unexpected and, yes, 
> problematic, results. R and its dlls will also use the underlying C 
> libraries of MinGW. But providing that the two gcc's present the same 
> API's to R and GRASS, experience (again limited) indicates that - against 
> my expectations - it does work. One advantage of the R package formalisms 
> is that the functions need to be able to run their examples to completion, 
> which they certainly wouldn't do if the two gcc's had diverged. 
> 
> Roger
> 
> > 
> 
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Breiviksveien 40, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 93 93
e-mail: Roger.Bivand at nhh.no