[Rd] stringsAsFactors .... in expand.grid() etc

Martin Maechler maechler at stat.math.ethz.ch
Tue May 19 15:49:13 CEST 2009

>>>>> "RT" == Rolf Turner <r.turner at auckland.ac.nz>
>>>>>     on Tue, 19 May 2009 11:02:08 +1200 writes:

    RT> On 19/05/2009, at 10:20 AM, Steve Lianoglou wrote:

    >> Hi all,
    >> I've (tried) to look through the bug tracker, and gmane-search the  
    >> R list to
    >> see if this has been mentioned before, and it looks like it hasn't.
    >> According to the R 2.9.0 release notes[1], the expand.grid function  
    >> should now
    >> take a stringsAsFactor=LOGICAL argument which controls whether or  
    >> not the
    >> function coerces strings as factors. While the parameter is indeed  
    >> in the
    >> function, a quick examination of the function's source shows that  
    >> the value
    >> of this argument is never checked, and all strings are converted to  
    >> factors
    >> as a matter of course.
    >> The fix is pretty easy, and I believe only requires changing the  
    >> `if` check
    >> here:
    >> if (!is.factor(x) && is.character(x))
    >> x <- factor(x, levels = unique(x))
    >> To:
    >> if (!is.factor(x) && is.character(x) && stringsAsFactors)
    >> x <- factor(x, levels = unique(x))
    >> I can open a ticket regarding this issue and add this there if  
    >> necessary.

Thank you, but it's  not necessary anymore;   we (R Core Team)
have already opened a ticket...

    >> Thanks,
    >> -steve
    >> [1] http://article.gmane.org/gmane.comp.lang.r.general/146891

    RT> While we're at it --- would it not make sense to have the  
    RT> stringsAsFactors
    RT> argument (once it's working) of expand.grid() default to options() 
    RT> $stringsAsFactors,
    RT> rather than to FALSE?


    RT> This would make no difference to me personally, since I set
    RT> options(stringsAsFactors=FALSE) in my .Rprofile.  But it might make some
    RT> people happier ....

As I have said several times (on R-devel rather than R-help),
I strongly believe that the introduction of such an option has
been one of very few very bad choices we (the R core team) have
made in the recent past.
Setting an option should *never* influence (basic) R
computational functionality; options() typically should only
influence print()ing, maybe plotting() and similar I/O.

I still hope that I'll find the stamina in the not so distant
future to convince a majority within R-core to abolish this
abominational option, of course with transitional help for the
stringsAsFactors_=_FALSE junkies.

Martin Maechler

More information about the R-devel mailing list