R devel - Dec 2005 - FW: Bug in sum() with named data frame arguments

I got an email error message when I attempted to send this from my work
account.  I have manually added it to the bug tracker, and am resending from
my personal account.

-G

On 12/12/05, Warnes, Gregory R <gregory.r.warnes@pfizer.com>
wrote:
>
>
>
> >  -----Original Message-----
> > From:         Warnes, Gregory R
> > Sent: Monday, December 12, 2005 1:53 PM
> > To:   'bugs@cran.r-project.org'
> > Subject:      Bug in sum() with named data frame arguments
> >
> >
> > Sum fails if it is passed a data frame argument with a name. It also
> generated different error messages depending on exactly how the data frame
> is passed in:
> >
> > > df <- data.frame(m1=1:2,m2=3:4)
> > > df
> >   m1 m2
> > 1  1  3
> > 2  2  4
> > > sum(df)  # succeeds
> > [1] 10
> > > sum(df=df)  # fails
> > Error in as.matrix(x) : argument "x" is missing, with no
default
> > > sum(1,df=df) # fails, different error message
> > Error in sum(..., na.rm = na.rm) : invalid 'mode' of argument
> > > sum(df=df,1) # fails, different error message
> > Error in sum.default(..., na.rm = na.rm) :
> >       invalid 'mode' of argument
> > ## Note: no sum.default exists...
> >
> >
> > No similar problem exists for matrixes:
> >
> > > x <- as.matrix(df)
> > > sum(x)
> > [1] 10
> > > sum(df=x)
> > [1] 10
> > > sum(m1=x)
> > [1] 10
> > > sum(1,m1=x)
> > [1] 11
> > > sum(m1=x,1)
> > [1] 11
> > >
> >
> > The cause seems to be that the generic sum is defined as:
> >
> > > sum
> > function (..., na.rm = FALSE)
> > .Internal(sum(..., na.rm = na.rm))
> > <environment: namespace:base>
> >
> > but the S3 methods are defined as
> >
> > > Summary.data.frame
> > function (x, ...)
> > {
> >     x <- as.matrix(x)
> >     if (!is.numeric(x) && !is.complex(x))
> >         stop("only defined on a data frame with all numeric or
complex
> variables")
> >     NextMethod(.Generic)
> > }
> > <environment: namespace:base>
> >
> > I suspect that the proper resolution of this bug is to change the base
> 'sum' method to provide a 1st argument named 'x', and to
update the
> documentation.
> >
> > Note that the current implementation also failes when more than one
data
> frame argument is supplied:
> >
> > > sum(y,y,y)
> > Error in sum.default(..., na.rm = na.rm) :
> >       invalid 'mode' of argument
> >
> > while the same thing with matrixes works fine:
> >
> > > sum(y,y,y)
> > Error in sum.default(..., na.rm = na.rm) :
> >       invalid 'mode' of argument
> >
> > -Greg
> >
> > Gregory R. Warnes, Ph.D.
> > Associate Director, Non-Clinical Statistics
> > Pfizer Global Research and Development
> >
>
>
	[[alternative HTML version deleted]]

>>>>> "Greg" == Gregory Warnes <gregory.warnes at
gmail.com>
>>>>>     on Mon, 12 Dec 2005 14:03:00 -0500 writes:
    Greg> I got an email error message when I attempted to
    Greg> send this from my work account.  I have manually
    Greg> added it to the bug tracker, and am resending from
    Greg> my personal account.


    Greg> -G

    Greg> On 12/12/05, Warnes, Gregory R <gregory.r.warnes at
pfizer.com> wrote:
    >> 
    >> 
    >> 
    >> >  -----Original Message-----
    >> > From:         Warnes, Gregory R
    >> > Sent: Monday, December 12, 2005 1:53 PM
    >> > To:   'bugs at cran.r-project.org'
	        ^^^^^^^^^^^^^^^^^^^^^^^

	Can you tell where you took this address from?

We'd very much like that R bug reports be sent to
     R-bugs at r-project.org

(from where they are forwarded to the repository in Denmark,
 *after* having been virus- and spam-filtered).

Martin Maechler, ETH ZUrich