R devel - Jul 1997 - R-alpha: More problems with R0.50.a1

More problems...

1) when -lf2c is used for linking as in the standard Makefile in
src/main, it does not include the library as shown by ldd on my
slackware linux 2.0.29 but does on RedHat linux 2.0.27. But with
slackware adding the -L/usr/lib just before -lf2c does include it. On
the other hand readline in the same directory is included both times.
(We need the f2c library for complex numbers in Richard Jones' carma
for kalman filtering which we are dynamically loading.)

2) matrix(2,nrow=3,ncol=3)[!is.na(matrix(2,nrow=3,ncol=3))]
gives a vector of nine 2's
if the matrix has row and column names, the vector has the 3 column
names completed with six NA names.

3) Robert's example on 'Environment strangeness' has 2 errors:
	- missing bracket on the second last line of his g function.
	- should not try to print before b has been retrieved from the
	  appropriate environment (on same line as above).

4) why does the following function yield errors with R scoping
rules?
> p <- function () {
	b <- 123
	s()
}
> s <- function () {
	print(b)
}
> p()
Error: Object "b" not found.

5) getting the call line and saving it as an attribute:
This works:
> f <- function () {
	x <- match.call()
	print(x)
}
This recurses in an infinite loop, recalling the 'f' function each
time the "attr" line is stored in the "call" attribute of
x.
> f <- function () {
	x <- 2
	print(x)
	attr(x,"call") <- match.call()
	print(attributes(x))
}

6) The following gives a error. How can this be solved?
> "[<-.newtype" <- function(...) {
	NextMethod("[<-")
}
> f <- function () {
	idx <- 2
	tmp <- matrix(3,ncol=4,nrow=4)
	class(tmp) <- "newtype"
	tmp[idx,] <- matrix(1,ncol=4,nrow=4)[2,]
}
> f()
Error: Object "idx" not found.

This did not work in R0.49 either, but does work in S..
(In R the name of the variable "idx" is passed to
"[<-.newtype" in the
row and col local variable as in S it is the contents of "idx" which
are passed through.)
I had solved this for R0.49 by adding the line:
	idx <- eval(idx,sys.frame(sys.parent()))
But this does not work any more in R0.50.a1...
So I modified the line to:
        idx <- eval(expression(idx),sys.frame(sys.parent()))
but this did not change anything.
It looks as if the environments were not always retrieved properly and
that R scoping rules are not always followed. I thought that R would
be able to find any variable defined in a previous environment so that
they did not have to be passed from function to function as in S.

Patrick.
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Patrick,

   I can easily respond to some of the questions.
   
   2) That's the same thing S does (they don't propagate names though).
There
    are many instances where coercion or subsetting causes a loss of
    dimensioning information. In this case you are asking for a subset of the
    values in x; you haven't specified a format for them so they come back
as
    a vector.
   
   3) Sorry about that; I was working without cut and paste as we transferred
   to the new machine.
   
   4) b is defined in the environment created for the evaluation of p.
   The function s has as its parent the global environment (that's the
   environment that was active when s was defined). There is no value for
   b stored there hence s cannot find it. What you seem to want here
   is really dynamic scope rather than lexical scope. Dynamic scope was
   used in the early Lisp implementations but was quickly seen to be a
   mistake and I don't believe that it is used by any active languages
   (apart from people trying to fake it in Splus). 
   It's reasonably easy to tweak R to have this rather than lexical scope
   but the down side is pretty large. Imagine a more complicated system
   with a much deeper call stack; now s will grab the definition of b that
   is closest to it in the call stack; this may not be the one you want and
   it can take forever to find the problem.
   
   5) I think Ross has fixed this. At least both of the functions you gave
   run without error in the version I've got now. We will probably put up
   some patches pretty soon.
   
   6) This will take some work. I'm still unconvinced that any of this
Method
   stuff deserves any effort whatsoever. It has to be a better idea to write
   a real object system rather than try to constantly patch this thing. But
   I will spend some time digging around. To see where idx has disappeared.
   
  
   robert
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-