[R] Problem with installing home-made package under Windoze.

[Prof Brian Ripley]

I tried this with aaMI and had no problems, from either i386 or x86_64 
Linux.  (I think we have only ever said that i386 Linux will work and I 
suspect that a big-endian host would not.)

It look as if the transferred files have got corrupted. Marc's suggestion 
is unlikely, as these errors are coming from C code and it does look as if 
.readRDS is calling the right C code.

If you create a MD5 checksum file in Linux, it can be checked on Windows. 
See the help on checkMD5sums in package tools.  I believe that if you put 
a MD5 file in the package on Linux, it will automatically be checked when 
installing on Windows: it certainly happens for installs from a 
repository.

On Sat, 5 Nov 2005, Marc Schwartz wrote:

> On Sat, 2005-11-05 at 17:35 -0400, Rolf Turner wrote:
>> I am using/trying to use R 2.2.0.
>>
>> I have created a package of ``supplementary'' time series functions
>> for use by my students in a time series course.  The package involves
>> only raw R code; no dynamic loading to complicate matters.  I need to
>> install this package in a location on a local area network where the
>> students can access it under Windoze.
>>
>> In the past I learned that I could do this by:
>>
>> 	o installing the package under Linux
>> 	o changing directories to the local library
>> 	o executing
>>
>> 		zip -r9l ts.sup.zip ts.sup
>>
>> 	o uploading the resulting ts.sup.zip file to a convenient
>> 	  Windoze-running machine
>>
>> 	o excuting
>>
>> 	 > install.packages("ts.sup.zip",repos=NULL,lib="L:/statdata/Rlib")
>>
>> 	the value of the ``lib'' argument being a location in the LAN
>> 	where I have write permission and which is accessible by the
>> 	students.
>>
>> In the past this worked.  Now when I try this I get the error message
>>
>> 	``Error in .readRDS(pfile) : cannot read workspace version
>> 	  167772160 written by R 512.2.2; need R 0.1.4 or newer''
>>
>> I experimented with other small locally built packages and got the
>> same phenomenon.  Then I tried installing a package from CRAN
>> choosing to install ``aaMI'' just because it was first on the
>> list.  Got the same error message.
>>
>> I tried firing up R version 2.1.1, which is still around.  Strangely,
>> this allowed me to install ``aaMI'' from CRAN, but threw up the same
>> error message when I tried to install from local zip files.
>>
>> This is very frustrating --- I'm trying to install an updated package
>> with some bugs removed and new facilities added.  Can anyone suggest
>> what might be causing the problem and how I might fix it?
>
> Rolf,
>
> What happens if you start up R on Windows from the command line using
>
>  Rgui --vanilla
>
> ?
>
> Any chance that you have a corrupted workspace getting restored on
> startup that is subtle enough to not be immediately evident, but causes
> this problem in later steps?
>
> You may need to remove any .RData files that are getting loaded on
> startup and recreate your package(s) in a clean R session before further
> attempts at installation.
>
> HTH,
>
> Marc Schwartz

-- 
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