R devel - Jan 2007 - Why does not the command 'length(a <- 1:5) <- 4' not work?

when running the command
  > length(a <- 1:5) <- 4

there are two responses.

If 'a' does not exist then the response is

Error in length(a <- 1:5) <- 4 : object "a" not found

If 'a' does exist then the response is

Error in length(a <- 1:5) <- 4 : could not find function
"<-<-"

I would assume that 'length(a <- 1:5) <- 4' should work because 
'length(a <- 1:5)' does work.

Thank you for your help.

Charles Dupont
-- 
Charles Dupont	Computer System Analyst		School of Medicine
		Department of Biostatistics	Vanderbilt University

On 1/16/2007 1:24 PM, Charles Dupont wrote:
> when running the command
>   > length(a <- 1:5) <- 4
> 
> there are two responses.
> 
> If 'a' does not exist then the response is
> 
> Error in length(a <- 1:5) <- 4 : object "a" not found
> 
> If 'a' does exist then the response is
> 
> Error in length(a <- 1:5) <- 4 : could not find function
"<-<-"
> 
> I would assume that 'length(a <- 1:5) <- 4' should work
because
> 'length(a <- 1:5)' does work.
I'm guessing you are assuming it would mean the same as

a <- 1:5
length(a) <- 4

But how would R know the name of the variable whose length is set in the 
second line?  In "a <- 1:5" the "a" is just part of a
larger expression,
it's not special as in some other languages, i.e. R sees that as

"<-"(a, 1:5)

So if you rewrote your original as

length("<-"(a, 1:5)) <- 4

it is a lot less clear that you really mean to create and then change
"a".

In general things like

foo(a) <- b

are evaluated as "foo<-"(a, b), where "foo<-" is a
function that expects
a variable reference as its first argument.  There's some magic going on 
to allow things like

a <- matrix(1:4, 2,2)
names(dim(a)) <- letters[1:2]

and I think the parser is trying to set things up for that kind of magic 
in your expression, though I haven't traced through the execution path 
to explain exactly why you saw the error messages you saw.

Duncan Murdoch

On Tue, 16 Jan 2007, Charles Dupont wrote:
> I would assume that 'length(a <- 1:5) <- 4' should work
because
> 'length(a <- 1:5)' does work.
Appearances can be deceptive.

length(a <- 1:5) evaluates a <- 1:5 and then passes its value to length(),
in ordinary call-by-value semantics.

That can't be how complex assignment functions work, though.

Consider

length(names(a))<-5

and suppose names(a) is c("A","B").  This code doesn't
just evaluate
names(a) and pass the result on, since
    length(c("A","B"))<-5
would be silly (and invalid).  It must be doing something more 
complicated to make sure that "a", buried deep inside the call, gets 
updated.

What actually happens is that
   f(x) <- y
is rewritten in terms of the assignment function "f<-" as
   x <- "f<-"(x, y)
so
   length(names(a))<-5
is
   names(a) <- "length<-"(names(a), 5)
which in turn is
   a <- "names<-"(a, "length<-"(names(a), 5))

This requires the assignment function "f<-" to exist, of course.

In your case
   length(a <- 1:5) <- 4
is rewritten as
   (a<-1:5) <- "length<-"(a<-1:5, 5)
and then as
   a <- "<-<-"(a, "length<-"(a<-1:5, 5))
which gives the error you report -- there is no function "<-<-".

You  might think the last stage of rewriting could just be avoided, but
   (a<-1:5) <- "length<-"(a<-1:5, 5)
is invalid if not rewritten, since it tries to modify 1:5, which is not a 
variable.

If you want to compute
    a <- 1:5
    length(a)<-4
then just do it like that.


 	-thomas