[Rd] Negative integer subscripts in [[?

[William Dunlap]

> -----Original Message-----
> From: Duncan Murdoch [mailto:murdoch at stats.uwo.ca] 
> Sent: Monday, November 09, 2009 4:55 PM
> To: William Dunlap
> Cc: r-devel at r-project.org
> Subject: Re: [Rd] Negative integer subscripts in [[?
> 
> On 09/11/2009 4:38 PM, William Dunlap wrote:
> > Should negative subscripts be outlawed in
> >    x[[subscript]]
> > ?  
> > 
> > Currently, if subscript is a scalar then it can only
> > be negative if length(x)==1 (otherwise [[ throws an

   Oops, I meant: length(x)==2

> > error).  If length(subscript)>1 then it gets treated
> > as an attempt to recursively extract an element of
> > a nested list.
> > 
> >> list(10,20)[[-1]] # get the last element out of 2
> > [1] 20
> >> list(10,20,30)[[-(1:2)]] # get the last of 3?  No.
> > Error in list(10, 20, 30)[[-(1:2)]] :
> >   attempt to select more than one element
> >> list(10,list(20,30))[[-c(1:2)]] # see how recursive subscripting is
> > done
> > [1] 20
> > 
> > If negative subscripts were not allowed in [[ then 
> > there might be a little less confusion about [[.
> 
> I agree, it would be better not to allow negatives here, but as John 
> said it's probably too late to do away with them.
> 
> > (If recursive subscripting were done by a list instead
> > of by an integer or character vector there might be
> > less confusion and it would be more flexible.)
> 
> I don't follow this.  Recursive lists are trees, and you specify a 
> single element of a tree by specifying a sequence of indices. 
>  Why would 
> it be less confusing to give a list?  What extra flexibility 
> could there 
> be?  I suppose you could mix integer and character indices, but what 
> would be meant by x[[ list(1, list(2,3), 4) ]] ?

It would allow mixed integer, character, and logical indices
and, since lists are not allowed for other forms of subscripts,
it would make it clear that you wanted a recursive subcript.
At each level you would have to select just one element, just as is
the restriction now.
E.g., now we do
   x <- list(A=list(10,20), B=list(B1=30, B2=40))
   x[[c(1,2)]] # gives 20
but this could be done as
   x[[list(1,2)]]
or
   x[[list("A",2)]]
or
   x[[list(c(TRUE,FALSE), -1)]]
It would mean that the following 2 functions would return the
same result
   f1 <- function(x, subscriptList) x[[subscriptList]]
   f2 <- function(x, subscriptList) {
             for(i in subscriptList)
                 x <- x[[i]] # or x[i][[1]] to handle logical subscripts
             x
          }
when subscriptList is a list.  Currently they return the same
when subscriptList is an integer vector.

> I don't know the original motivation for allowing vector indexing to 
> lists, but I extended it to pairlists so that it would be possible to 
> specify a location within a function exactly, by walking down 
> the parse 
> tree.  I think it's something that would be rarely used, but when you 
> need it, it's very handy.

That is the sort of thing I've used it for, rarely.
We can always change the logical and character vectors
into integers so it isn't a big deal.  I was more concerned
by folks accidently using [[ with a vector subscript where [
was appropriate and getting a mysterious error message or
the wrong answer from the unintended recursive subscript.
However, it is far too late to get rid of the integer vector
version of it.

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com 

> 
> Duncan Murdoch
>