Functions that write functions in R packages

On 2/23/2007 11:05 AM, hadley wickham wrote:

Dear all,

Another question related to my ggplot package:  I have made some
substantial changes to the backend of my package so that plot objects
can now describe themselves much better.  A consequence of this is
that a number of convenience functions that previously I wrote by
hand, can now be written automatically.  What is the best practice for
creating these functions for bundling in a package?  I see three
possible solutions:

 * dump function specifications out to a .r file
 * dynamically create at package build time so they are including in
the package rdata file
 * dynamically create at package load time

Can anyone offer any advice as to which is preferable? (or if there's
a better way I haven't thought of)

My code currently looks like this (experimenting with two ways of
creating the functions)

create_accessors <- function(objects, name, short=NULL) {
      lapply(objects, function(x) {
              assign(paste(name, x$objname, sep="_"), x$new, pos=globalenv())
              if (!is.null(short)) {
                      eval(
                              substitute(
                                      f <- function(plot, ...) plot + add(...),
                                      list(
                                              add = as.name(paste(name, x$objname, sep="_")),
                                              f = as.name(paste(short, x$objname, sep=""))
                                      )
                              ), envir = globalenv()
                      )

              }
      })
}

I'd say it's not a great idea to write to globalenv.  What if your
function stomps on my object of the same name?  It would be better to
set up your own environment and write these objects there.  You could
then attach that environment, and the user would see your functions (if
he hadn't defined his own).

If others of your functions need to call these, then you need to be
careful with environments so that they don't accidentally call the
user's function of the same name instead.

With the considerations above, I think it would be easiest to do the
creation at install time or earlier, rather than at load time.