[Rd] RE: [R] Bus error with xyplot

ripley@stats.ox.ac.uk ripley@stats.ox.ac.uk
Mon Feb 3 20:38:02 2003

On Mon, 3 Feb 2003, Liaw, Andy wrote:

> > From: Martin Maechler [mailto:maechler@stat.math.ethz.ch]
> > 
> > >>>>> "Giovanni" == Giovanni Petris <GPetris@uark.edu>
> > >>>>>     on Mon, 3 Feb 2003 08:36:06 -0600 (CST) writes:
> > 
> >     Giovanni> Has anybody else experienced something like the 
> > example below? 
> > not recently.
> > Could it be that your version of the `grid' package (which is
> > loaded by `lattice') or `lattice' are incompatible (i.e. older
> > than) to your R version?
> > After   library(lattice), use 
> >  .path.package()
> > to see where it was loaded from.
> This brings up (IIRC) a topic that was discussed on R-devel a while ago:
> Version checks for required packages.  Is this feasible?  Seems quite
> worthwhile to me...

It's not feasible.  If package A is upgraded and makes package B need an 
upgrade, there is no way for package B's maintainer to anticipate this.
Now one could be conservative and have B require exactly one version of A,
but then upgrading A makes B unusable.  (The latter is what has happened a 
few times with ghostscript and gsview/windvi.)

I have suggested before that grid be more closely integrated with R, and 
become a base package.  Then grid would only be changed when the R version 
changes, and the critical incompatibility goes away.


Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595