Skip to content

Upgrade 'R'

7 messages · rkevinburton at charter.net, Leon Yee, Brian Ripley +2 more

rkevinburton at charter.net
# 
More and more I am getting warnings from packages that I install that the package was built with 2.7.2 (I am running 2.7.1). I would like to upgrade but don't want to loose all of the packages that I have installed and the settings. Is there a way to just "upgrade" without uninstalling and reinstalling 'R'?

Thank you.

Kevin
Leon Yee
# 
Hello, Kevin

    You can get some hints by browsing in this mailist with the subject 
of " Upgrading R means I lose my packages", which were posted several 
days ago.

HTH

Leon
rkevinburton at charter.net wrote:
Keith Jewell
# 
As possible help to others, and also as a request for comments on how I 
might do things better, I describe how I've recently altered my system to 
handle this.

I'm on a Windows Server 2003 network and the R installation is accessible to 
many others. Everyone has "read" access to all installation files, but only 
I have write access. I do _not_ have Administrator privileges on the server, 
so I cannot make/change registry entries.

R versions are in an R folder tree, which also holds the version independent 
library folder.

//Server02/stats/R
//Server02/stats/R/R-2.7.0
//Server02/stats/R/R-2.7.1
//Server02/stats/R/R-2.7.1pat
//Server02/stats/R/R-2.7.2
   :
//Server02/stats/R/library

In each version I have edited /etc/Rprofile.site to include the line
 .libPaths("//Server02/stats/R/library")

The "default" libraries (base, boot, class...) are installed into the 
relevant version specific .../R-n.n.n/library/ folder by the windows 
installer program (e.g. R-2.7.2-win32.exe).
Occasionaly, and after installing a new R-version,  I update all the 
downloaded libraries in the version independent //Server02/stats/R/library/ 
folder with a simple update.packages().

HTH. Comments welcome.

Keith Jewell
---------------------------------------
"Leon Yee" <yee.leon at gmail.com> wrote in message 
news:48BCA225.7060106 at gmail.com...
Brian Ripley
#
On Tue, 2 Sep 2008, Keith Jewell wrote:

            

      
      
      
    

        
Please read the rw-FAQ for more details and insight, especially in using 
Renviron.site (here you want to set R_LIBS_SITE: see ?libPaths) and using 
update.packages(checkBuilt=TRUE, ask=FALSE)

For example, my sysadmins have Windows R installed locally (via the MSI 
installer), and then have on each machine in etc/Renviron.site have

R_LIBS_SITE=N:/R/library/2.7
R_LIBS_USER=P:/R/win-library/2.7

This allows both common packages and user-specific packages to be stored 
on SMB-mounted drives.  (N: is common to all machines, P: is 'personal' 
-- we use mapped drives rather than shares to allow quick changes of 
server)

There are two reasons we install locally.  The first is performance: if 20 
machines in a lab start R up simultaneously the network load is high.  The 
second is that our security settings (and I believe the defaults these 
days) disallow use of CHM help on network drives -- we remove the chtml 
directories from packages installed on R_LIBS_SITE, so R defaults to text 
help for those packages.

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Keith Jewell
    

    

      
      #
    

  


  

      
"Prof Brian Ripley" <ripley at stats.ox.ac.uk> wrote in message 
news:alpine.LFD.1.10.0809021124300.12240 at gannet.stats.ox.ac.uk...

            

      
      
      
    

        
Thanks Professor Ripley. I'd looked at the R Windows FAQ but  not carefully
enough.Using Renviron.site does seem much more elegant than Rprofile.site.
I've now included a line in Renviron.site
R_LIBS_SITE=//Server02/stats/R/library

I also note your use of site and personal libraries specific to "first 
decimal"
versions of R. Would you recommend others to follow this practice?

Thanks again,

Keith Jewell.

  
  


  


      

    

    

      
      

        


  

        

      Gabor Grothendieck
    

    

      
      #
    

  


  

      
One possibility is to only keep one version when the third
digit in the version number changes and only create new fresh versions
when the first or second digit in the version number changes. Thus when
going from 2.7.1 to 2.7.2 you could just overwrite the 2.7.1 installation with
the new 2.7.2 installation but when going from 2.7.2 to 2.8.0 you would
create a new parallel installation of R.  AFAIK that has always worked in
the past and is what I normally do.

To overwrite your 2.7.1 with 2.7.2, when installing 2.7.2 just enter
the old folder
path in the "Select Destination Location" screen of the installer.
All your old
settings and libraries will be preserved.  (You will no longer have a
2.7.1 version.)

If you do want to create a new parallel installation when 2.8.0 comes out, say,
there are links in
http://batchfiles.googlecode.com
discussing different methods.  One of these
is provided by movedir.bat or copydir.bat in the batchfiles collection which
can be used to move or copy libraries safely as they will never overwrite.

        
On Mon, Sep 1, 2008 at 10:04 PM,  <rkevinburton at charter.net> wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On Tue, 2 Sep 2008, Keith Jewell wrote:

        

            

      
      
      
    

        
Yes: that is the level of compatibility.  (It is done automatically for 
personal libraries created by R: we set R_LIBS_USER because HOME is 
set incorrectly on some of our systems.)