R devel - Mar 2009 - surprising behaviour of names<-

playing with 'names<-', i observed the following:
  
    x = 1
    names(x)
    # NULL
    'names<-'(x, 'foo')
    # c(foo=1)
    names(x)
    # NULL

where 'names<-' has a functional flavour (does not change x), but:

    x = 1:2
    names(x)
    # NULL
    'names<-'(x, 'foo')
    # c(foo=1, 2)
    names(x)
    # "foo" NA
  
where 'names<-' seems to perform a side effect on x (destructively
modifies x).  furthermore:

    x = c(foo=1)
    names(x)
    # "foo"
    'names<-'(x, NULL)
    names(x)
    # NULL
    'names<-'(x, 'bar')
    names(x)
    # "bar" !!!

    x = c(foo=1)
    names(x)
    # "foo"
    'names<-'(x, 'bar')
    names(x)
    # "bar" !!!

where 'names<-' is not only able to destructively remove names from
x,
but also destructively add or modify them (quite unlike in the first
example above).

analogous code but using 'dimnames<-' on a matrix performs a side
effect
on the matrix even if it initially does not have dimnames:

    x = matrix(1,1,1)
    dimnames(x)
    # NULL
    'dimnames<-'(x, list('foo', 'bar'))
    dimnames(x)
    # list("foo", "bar")

this is incoherent with the first example above, in that in both cases
the structure initially has no names or dimnames attribute, but the end
result is different in the two examples.

is there something i misunderstand here?


there is another, minor issue with names:

    'names<-'(1, c('foo', 'bar'))
    # error: 'names' attribute [2] must be the same length as the vector
[1]

    'names<-'(1:2, 'foo')
    # no error

since ?names says that "If 'value' is shorter than 'x', it
is extended
by character 'NA's to the length of 'x'" (where x is the
vector and
value is the names vector), the error message above should say that the
names attribute must be *at most*, not *exactly*, of the length of the
vector.

regards,
vQ

Wacek Kusnierczyk wrote:
> playing with 'names<-', i observed the following:
>   
>     x = 1
>     names(x)
>     # NULL
>     'names<-'(x, 'foo')
>     # c(foo=1)
>     names(x)
>     # NULL
> 
> where 'names<-' has a functional flavour (does not change x),
but:
> 
>     x = 1:2
>     names(x)
>     # NULL
>     'names<-'(x, 'foo')
>     # c(foo=1, 2)
>     names(x)
>     # "foo" NA
>   
> where 'names<-' seems to perform a side effect on x
(destructively
> modifies x).  furthermore:
> 
>     x = c(foo=1)
>     names(x)
>     # "foo"
>     'names<-'(x, NULL)
>     names(x)
>     # NULL
>     'names<-'(x, 'bar')
>     names(x)
>     # "bar" !!!
> 
>     x = c(foo=1)
>     names(x)
>     # "foo"
>     'names<-'(x, 'bar')
>     names(x)
>     # "bar" !!!
> 
> where 'names<-' is not only able to destructively remove names
from x,
> but also destructively add or modify them (quite unlike in the first
> example above).
> 
> analogous code but using 'dimnames<-' on a matrix performs a
side effect
> on the matrix even if it initially does not have dimnames:
> 
>     x = matrix(1,1,1)
>     dimnames(x)
>     # NULL
>     'dimnames<-'(x, list('foo', 'bar'))
>     dimnames(x)
>     # list("foo", "bar")
> 
> this is incoherent with the first example above, in that in both cases
> the structure initially has no names or dimnames attribute, but the end
> result is different in the two examples.
> 
> is there something i misunderstand here?
Only the ideology/pragmatism... In principle, R has call-by-value
semantics and a function does not destructively modify its arguments(*),
and foo(x)<-bar behaves like x <- "foo<-"(x, bar). HOWEVER,
this has
obvious performance repercussions (think x <- rnorm(1e7); x[1] <- 0), so
we do allow destructive modification by replacement functions, PROVIDED
that the x is not used by anything else. On the least suspicion that
something else is using the object, a copy of x is made before the
modification.

So

(A) you should not use code like y <- "foo<-"(x, bar)

because

(B) you cannot (easily) predict whether or not x will be modified
destructively

--------
(*) unless you mess with match.call() or substitute() and the like. But
that's a different story.


-- 
   O__  ---- Peter Dalgaard             ?ster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark      Ph:  (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)              FAX: (+45) 35327907