[Rd] sum overflow (PR#1091)

[Martin Maechler]

>>>>> "BDR" == Prof Brian Ripley <ripley@stats.ox.ac.uk> writes:

    BDR> On Thu, 13 Sep 2001, Bill Simpson wrote:
    >> > It's not a problem with sum:
    >> >
    >> > > sum(a*a)
    >> > [1] 333833500
    >> > > sum(b*b)
    >> > [1] 333833500
    >> >
    >> > are accurate.
    >> >
    >> > The overflow is in the integer arithmetic for *.  That's a question for
    >> > your C run-time system.  On a 64-bit machine you might get different
    >> > results (although most use 32-bit ints, including mine).
    >> >
    >> > If you use integers you need to be aware of the consequences.  It's a
    >> > feature not a bug.
    >> I thought R used an internal rep that was double in all cases.

    BDR> Assumed, I guess, as I've never seen that in the manuals.

    >> Now I'm confused:
    >> > a<-(1:1000)
    >> > b<-(1:1000)
    >> > sum(a*a)*sum(b*b)
    >> [1] -652010736
    >> > a<-(1:1000)/1.0
    >> > b<-(1:1000)/1.0
    >> > sum(a*a)*sum(b*b)
    >> [1] 1.114448e+17
    >> 
    >> So  R somehow decides whether to use an integer or a double
    >> representation? Please tell me the rule used by R so I will know in the
    >> future.

    BDR> It uses what you tell it to in general.  Integers will be handled as
    BDR> integers, but as in C integer/numeric gives a numeric.  Rather than
    BDR> learn rules, use typeof or storage.mode as in
    >> typeof(1:1000)
    BDR> [1] "integer"

    BDR> to find out, or make explicit coercions (as.numeric) if you want
    BDR> to make assumptions about objects.

    BDR> At present you are only likely to get integers from m:n and
    BDR> similar seq() calls, and from unclass on a factor.  

also  length() and dim()
on which I even rely in .C() calls when not using explicit as.integer()
there  [but many will tell me that was not "defensive" and un-wise].

    BDR>  But no one is saying that will not change in the future (and it
    BDR> did from S3 to S4).  And you could always have a user who used
    BDR> as.integer explicitly.

indeed!

-- 
Martin Maechler <maechler@stat.math.ethz.ch>	http://stat.ethz.ch/~maechler/
Seminar fuer Statistik, ETH-Zentrum  LEO D10	Leonhardstr. 27
ETH (Federal Inst. Technology)	8092 Zurich	SWITZERLAND
phone: x-41-1-632-3408		fax: ...-1228			<><
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._