[R] What is a sane way to deal with changes in library loadings after 1.9.0?

[Prof Brian Ripley]

Your confusion is that you are trying to run scripts from .First.
You did not say so, and that is strongly not recommended, especially if 
they might have errors.

Use R CMD BATCH to run scripts and you will find it much easier.

BTW, please send replies to the individual asking, not just to the list.

On Mon, 5 Jul 2004 a.prioglio at city.ac.uk wrote:

> On Mon, Jul 05, 2004 at 11:51:25AM +0100, Prof Brian Ripley wrote:
> > > Well, I'm certainly new to R. Still without the explicit loading of the
> > > above libraries my scripts while loading on 1.8.1 were definetely not
> > > loading on 1.9.1. I'm using the same account on all machines so I expect
> > > to have the same environment. Or should I?
> > > 
> > > One thing I noticed now, a search() on a 1.8.1 machine after the loading
> > > of the scripts now returns nothing (after encountering an error though).
> > 
> > And what was the error?
> After a call to factanal()
> Error in sc %% S: non conformable arguments
> 
> If the offending line is removed search() returns its normal output
> 
> So it appears that the interpreter environment is affected by
> encountering errors in the script. I did not know or noticed before.
> > 
> > It seems that you have been fiddling with the default set of packages.
> > Have you set R_DEFAULT_PACKAGES?  Have you set option "defaultPackages"?
> > If so, set them to "", or try starting R with --vanilla.
> 
> No the R environment is "vanilla", i.e. no variables set and a call to
> "set" from bash does not show any R variable. R is from official RPM binaries
> for RH9 (on RH9) and the 1.9.1 from the RPM for SuSE 9.0 (on SuSE 9.0).
> 
> The only option changed in the scripts is the width of output for text
> files, which is restored at the end of the script.
> 
> No other option is changed.
> 
> > 
> > Packages utils, graphics and stats are loaded by default on 1.9.1.  And
> > package mva is loaded by default on 1.8.1.  People who answer here tend to
> > assume that people who know enough to change the default packages loaded
> > know enough to revert the changes ....
> 
> This is not my experience. The Changelog document for 1.9.0 indicates 
> clearly that scripts that used to work before 1.9.0 now need to
> explicitly load stats, utils and graphics libraries.
> Indeed these pachages are loaded by default if
> one starts R as an interpreter. This is not my the case, as I explicitly
> stated as I run R with a script loaded with the .First mechanism in
> .Rprofile.
> 
> My confusion derives from the fact that after I had to explicitly
> load/attach the libraries to cope with the changes as described in the
> Changelog document for 1.9.0, I started to experience problems that I did
> not notice before and more intersting even when using the 1.8.1
> version with the same script.
> .
> 
> > 
> > > How can one know what environments are loaded? Calling R --verbose did
> > > not seem to clarify this point.
> > 
> > search()  (attached, not loaded, BTW, as there are loaded namespace 
> > environments which are not attached).
> >
> Well, what I was hoping for was a way to see something from the unix
> shell like R --show-environment (I know it's not on the man page ...) or
> something like --verbose that did report the working environment. By the
> time I get to the R prompt I see nothing.
> 
> 

-- 
Brian D. Ripley,                  ripley at 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