Duncan Murdoch wrote:
On Mon, 12 Jul 2004 15:14:27 +0200, Uwe Ligges <ligges@statistik.uni-dortmund.de> wrote :
Gabor Grothendieck wrote:
The objective should be that creating a package is as easy as this:
f <- function()1; g <- function()2; d <- 3; e <- 4:5
package.skeleton(list=c("f","g","d","e"), name="AnExample")
library(AnExample)
f()
which means that the package needs to be inserted where library will
find it. It should not be necessary to have an understanding of this.
OK, I understand what you are going to do, but in that case you can use dump() into an *.R or save() into an *.RData file and use source()/load() to load it again. I don't see any advantage of a package if you don't want to modify documentation or other stuff in the package. Also, you would need the tools to make a binary package from the source package somewhere. package.skeleton() is clearly not intended to be used for that purpose, but to create the template for your source package. Hence the default should not be changed.
I agree with both of you on this. Currently the method that Uwe
describes is a lot easier than creating a package, but I think the
objective should be to make things almost as easy as Gabor describes.
Not completely as easy: he's missing the step where the package is
installed. I think we want to keep that (because the distinction
between the source of a package and the installed copy of it is
important), but it should be easier to install a new package than it
is now, especially in Windows. So I'm suggesting that it would be
nice to be able to do something like this:
f <- function()1; g <- function()2; d <- 3; e <- 4:5
package.skeleton(list=c("f","g","d","e"), name="AnExample")
install.packages("AnExample", build = TRUE)
library(AnExample)
f()
[moved the thread to R-devel....] Yes. That's the ideal solution. Anyway, these days you still need perl, and what happens if you have got some C or Fortran sources? Additionally, we can never do all the stuff without the tools. And I think you don't want to be able to do so, because maintaining the R sources will get much harder if development of the package management is that different under Unix-alikes and Windows. From my point of view the aim should be to join, not to diverge here .... Uwe BTW: Note that creating packages in such an easy manner does not remind people to write their documentation. And if nobody tells me to write documentation, my function will be useless (even for myself in a couple of weeks). But that's just the pedagogical part. ;-)
but currently install.packages doesn't know how to build, and for most Windows users, a fairly substantial effort is necessary to obtain all the tools.
>
Duncan Murdoch