R devel - Dec 2003 - negative numbers from object.size() on 64 bit systems

Has anyone running R 1.8.1 on a 64 bit system gotten negative numbers 
from object.size() on objects > 2GB or so?  For example, on 
Solaris/Sparc I get

 > b <- numeric(131072 * 2800)
 > object.size(b)
[1] -1358954440

I get similar behavior on the Opteron (SuSE Linux).  I'm not sure if 
this is an R problem or something wrong with the way it was compiled on 
either system.

I noticed in objectsize() in size.c, the `cnt' variable is an int, which 
I guess would max out at 2^31.  Might this be the problem?  Any hints 
would be appreciated.

Thanks,

-roger

Hi,
same behavior on this machine:
> b <- numeric(131072 * 2800)
> object.size(b)
[1] -1358954440
> version
         _
platform alphaev68-dec-osf4.0f
arch     alphaev68
os       osf4.0f
system   alphaev68, osf4.0f
status
major    1
minor    8.0
year     2003
month    10
day      08
language R


-------------------------------------
Wolfgang Huber
Division of Molecular Genome Analysis
German Cancer Research Center
Heidelberg, Germany
Phone: +49 6221 424709
Fax:   +49 6221 42524709
Http:  www.dkfz.de/abt0840/whuber
-------------------------------------

On Wed, 17 Dec 2003, Roger D. Peng wrote:
> Has anyone running R 1.8.1 on a 64 bit system gotten negative numbers
> from object.size() on objects > 2GB or so?  For example, on
> Solaris/Sparc I get
>
>  > b <- numeric(131072 * 2800)
>  > object.size(b)
> [1] -1358954440
>
> I get similar behavior on the Opteron (SuSE Linux).  I'm not sure if
> this is an R problem or something wrong with the way it was compiled on
> either system.
>
> I noticed in objectsize() in size.c, the `cnt' variable is an int,
which
> I guess would max out at 2^31.  Might this be the problem?  Any hints
> would be appreciated.
>
> Thanks,
>
> -roger

On Wed, 17 Dec 2003, Roger D. Peng wrote:
> Has anyone running R 1.8.1 on a 64 bit system gotten negative numbers 
> from object.size() on objects > 2GB or so?  For example, on 
> Solaris/Sparc I get
> 
>  > b <- numeric(131072 * 2800)
>  > object.size(b)
> [1] -1358954440
> 
> I get similar behavior on the Opteron (SuSE Linux).  I'm not sure if 
> this is an R problem or something wrong with the way it was compiled on 
> either system.
> 
> I noticed in objectsize() in size.c, the `cnt' variable is an int,
which
> I guess would max out at 2^31.  Might this be the problem?  Any hints 
> would be appreciated.
Yes, and so is objectsize, and the return type is integer.  At the time
that was written R could not use more than 2Gb.

I've modified the function but have no machine with more than 2Gb to test 
it on, so could you please try R-devel in a couple of days?

-- 
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