[Rd] No [[<-.factor()

Martin Maechler maechler at stat.math.ethz.ch
Mon Aug 30 11:48:57 CEST 2010


>>>>> Prof Brian Ripley <ripley at stats.ox.ac.uk>
>>>>>     on Mon, 30 Aug 2010 08:28:24 +0100 (BST) writes:

    > On Thu, 26 Aug 2010, Martin Maechler wrote:
    >>>>>>> "WD" == William Dunlap <wdunlap at tibco.com>
    >>>>>>> on Wed, 25 Aug 2010 17:31:27 -0700 writes:
    >> 
    WD> Should there be a [[<-.factor() that either throws
    WD> an error or acts like [<-.factor() to avoid making
    WD> an illegal object of class factor?
    >> 
    >> Yes, one or the other.
    >> Note that both `[<-` and `[[<-` are "Primitive" and do some
    >> "informal dispatch" in addition to the formal method dispatch,

    > I don't think so.  do_subassign2 does formal method dispatch, albeit 
    > internally, but nothing else (do_assign2_dflt selects only by 
    > typeof()).

One consider that as an "informal dispatch", notably
distinguishing "list"s and atomics...

    >> so there's no need for an explicit '[[<-.factor' method definition.
    >> For factors, it seems rational and "expectable" that
    >> double bracket behaves as single bracket when *one* element is
    >> assigned.
    >> z[[2:3]] <- c("Two, Three")
    >> however  should continue to give an error.

    > I added a [[<-.factor yesterday.  Note that we do have a [<-.factor 
    > which works on the integer codes, and I simply copied its logic.

thank you!

    > BTW, you probably meant

    > z[[2:3]] <- c("Two", "Three")

yes, indeed!
Martin



    >> >> z <- factor(c("Two","Two","Three"), levels=c("One","Two","Three"))
    >> >> z
    WD> [1] Two   Two   Three
    WD> Levels: One Two Three
    >> >> str(z)
    WD> Factor w/ 3 levels "One","Two","Three": 2 2 3
    >> >> z[[2]] <- "One"
    >> >> str(z) # the .Data part is now character
    WD> Factor w/ 3 levels "One","Two","Three": 2 One 3
    >> >> z
    WD> [1] <NA> <NA> <NA>
    WD> Levels: One Two Three
    >> >> z[2] <- "One"
    WD> Error in class(x) <- cx : adding class "factor" to an invalid object
    >> 
    WD> Bill Dunlap
    WD> Spotfire, TIBCO Software
    WD> wdunlap tibco.com



More information about the R-devel mailing list