R devel - Aug 2008 - ifelse

I find it slightly surprising, that

ifelse(TRUE, character(0), "")

returns NA instead of character(0). 

[WNT 2.6.2 Patched]

-- 
Heikki Kaskelma

on 08/18/2008 06:35 PM Heikki Kaskelma wrote:
> I find it slightly surprising, that
> 
> ifelse(TRUE, character(0), "")
> 
> returns NA instead of character(0). 
> 
> [WNT 2.6.2 Patched]
Time to upgrade...  :-)


The same behavior is in:

  R version 2.7.2 beta (2008-08-16 r46368)


The reason for this is that the internal code for ifelse() has:

    if (any(test[!nas]))
        ans[test & !nas] <- rep(yes, length.out = length(ans))[test &
            !nas]

where the key code is:

  rep(yes, length.out = length(ans))

In this case, 'yes' is character(0) and 'ans' is TRUE.

Thus, you get:

  > rep(character(0), length.out = length(TRUE))
  [1] NA


This behavior is documented in ?rep, where in the Details section you
will find:

"If x has length zero and length.out is supplied and is positive, the
values are filled in using the extraction rules, that is by an NA of the
appropriate class for an atomic vector (0 for raw vectors) and NULL for
a list. "

Thus:
> class(rep(character(0), length.out = length(TRUE)))
[1] "character"


which shows that the NA that is returned is of class character, which is
the class for character(0).

HTH,

Marc Schwartz

Others have commented on why this holds.

There is an alternative, 'ifelse1', part of the splus2R package, that
does what you'd like here.

Tim Hesterberg
>I find it slightly surprising, that
>   ifelse(TRUE, character(0), "")
>returns NA instead of character(0). 
>
>-- 
>Heikki Kaskelma