So I did open a new subject and move the discussion to R-devel
now.
"MM" == Martin Maechler <maechler at stat.math.ethz.ch>
on Mon, 24 Jul 2006 11:46:05 +0200 writes:
>>>>> "Gabor" == Gabor Grothendieck <ggrothendieck at gmail.com>
>>>>> on Sun, 23 Jul 2006 09:02:35 -0400 writes:
Gabor> Looking at the diff.POSIXt code we see the problem is
Gabor> that it takes the length of the input using length
Gabor> which is wrong since in the case of POSIXlt the
Gabor> length is always 9 (or maybe length should be defined
Gabor> differently for POSIXlt?).
MM> Though I agree with Spencer that a user may expect
MM> length() to behave differently, but I don't think this
MM> would be a good idea. Yes, length() is generic, but its
MM> help() emphasizes that for lists, length() should be the
MM> number of list elements. Of course anyone one *can*
MM> define length() methods that behave differently for
MM> his/her classes, but then one would also want to make
MM> sure that e.g. x[length(x)] or 'x[length(x)] <- value'
MM> works and -- in a case of simple S3 class built on a
MM> list, would work differently than if x was a the simple list.
Hmm, after thinking a bit more, and particularly after seeing
all the other methods we already have working for "POSIXlt",
I'm much less sure if (re)definining length(<POSIXlt>)
would be such a bad idea.
Indeed, "[", and "[<-" and "[[" already work with an implicit
length of 'length_of_series' and not length 9, and so does str().
So it seems to me, there is more reasonable momentum *for* a
length.POSIXlt() method than I first considered.
Still not sure what the wisest decision would be, and I am
asking for more opinions or arguments pro / con.
Since Brian Ripley has done the most work on the POSIX[cl]?t
classes, we should also wait on his opinion.
Martin
MM> In my view, I would only consider redefing length() for
MM> "non-basic" S4 classes, i.e. those with slots, where no
MM> confusion is possible, since these objects are
MM> definitely not simple vectors nor lists (aka "generic"
MM> vectors).