[R] R and Scheme

[Wacek Kusnierczyk]

Stavros Macrakis wrote:
> I've read in many places that R semantics are based on Scheme semantics.  As
> a long-time Lisp user and implementor, I've tried to make this more precise,
> and this is what I've found so far.  I've excluded trivial things that
> aren't basic semantic issues: support for arbitrary-precision integers;
> subscripting; general style; etc. I would appreciate corrections or
> additions from more experienced users of R -- I'm sure that some of the
> points below simply reflect my ignorance.
>
> ==Similarities to Scheme==
>
> R has first-class function closures. (i.e. correctly supports upward and
> downward funarg).
>
> R has a single namespace for functions and variables (Lisp-1).
>
> ==Important dissimilarities to Scheme (as opposed to other Lisps)==
>
> R is not properly tail-recursive.
>
> R does not have continuations or call-with-current-continuation or other
> mechanisms for implementing coroutines, general iterators, and the like.
>   

there is callCC, for example, which however seems kind of obsolete.

> R supports keyword arguments.
>
> ==Similarities to Lisp and other dynamic languages, including Scheme==
>
> R is runtime-typed and garbage-collected.
>
> R supports nested read-eval-print loops for debugging etc.
>
> R expressions are represented as user-manipulable data structures.
>
> ==Dissimilarities to all (modern) Lisps, including Scheme==
>
> R has call-by-need, not call-by-object-value.
>
> R does not have macros.
>
> R objects are values, not pointers, so a<-1:10; b<-a; b[1]<-999; a[1] =>
> 999.  Similarly, functions cannot modify the contents of their arguments.
>   

have you actually tried this code?  even if the objects are values not
pointers, assignment causes, in cases such as the above, copying the
value with modifications applied as needed.  thus, a[1] -> 1, not 999,
even though after b<-a b and a are the same value object.

try the following:

system.time(x<-1:(10^8))
system.time(y<-x)
system.time(y[1]<-0)
system.time(y[2]<-0)
head(x)
head(y)


with some trickery, functions can modify the contents of their
arguments, using deparse/substitute and assign:

a <- 1
f <- function(x) assign(deparse(substitute(x)), 0, parent.frame())
f(a)
a


the 'cannot modify the contents' does not apply to arguments that are
environments:

e <- new.env(parent=emptyenv())
l <- list()
f <- function(e) e$a = 0
f(e)
e$a
f(l)
l$a



> There is no equivalent to set-car!/rplaca (not even pairlists and
> expressions).  For example, r<-pairlist(1,2); r[[1]]<-r does not create a
> circular list. And in general there doesn't seem to be substructure sharing
> at the semantic level (though there may be in the implementation).
>   

computations on environment objects seem not to be subject to the
copy-value-on-assignment semantics:

e <- new.env(parent=emptyenv())
ee <- e
e$a <- 0
ee$a


> R does not have multiple value return in the Lisp sense.
>
> R assignment creates a new local variable on first assignment, dynamically.
> So static analysis is not enough to determine variable reference (R is not
> referentially transparent). Example: ff <- function(a){if (a) x<-1; x} ;
> x<-99; ff(T) -> 1; ff(F) -> 99.
>
> In R, most data types (including numeric vectors) do not have a standard
> external representation which can be read back in without evaluation.
>
> R coerces logicals to numbers and numbers to strings. Lisps are stricter
> about automatic type conversion -- except that false a.k.a. NIL == () in
> Lisps other than Scheme.
>   

types are not treated coherently.  in some situations, r coerces doubles
to complex (according to the hierarchy of types specified here and there
in the man pages), in others it won't:

x <- as.double(-1)
y <- as.complex(-1)
x == y

sqrt(x)
sqrt(y)

in certain cases, r will also do implicit inverse (downward) coercion:

is(y:y)



vQ