[Rd] Packages of snippets?
Roger D. Peng
rpeng at jhsph.edu
Thu Jun 24 05:34:54 CEST 2004
Duncan, I very much like your idea of a RNews/JSS "package" which
would contain code snippets from different contributors. In
fact, I would be willing to maintain such a package if it came
into existence. One note (I'm not sure if this is a flaw,
strictly speaking), the package would only be able to cover
functions. So general "tips", for example those covered on the
Rtips webpage (http://www.ku.edu/~pauljohn/R/Rtips.html), would
not be able to make their way into the package.
Another issue is that as of now, there are many useful tools for
checking for errors/problems in *packages*. But there's no real
way to check a single code file and a documentation file (say, an
.Rd file). One possibility might be to force contributors to
submit something like a DESCRIPTION file for their function (with
the necessary contact information), along with the code and a man
page. Then there would be anough material to make a fake package
on which to run R CMD check.
Duncan Murdoch wrote:
> [Moved from r-help to r-devel]
> On Thu, 24 Jun 2004 09:29:12 +1200, Patrick Connolly
> <p.connolly at hortresearch.co.nz> wrote:
>>I made myself a function in the S-PLUS days which I've modified to
>>work in R. It involved adding another few functions to add dates to
> This isn't something I'd use, since I prefer to keep objects in source
> form, but I bet there are quite a few people who would use it if it
> were generally available.
>>I've toyed with the idea of adding an object size column but it's not
>>important enough for my use. Since I revisit projects over a period
>>of years at times, the date is very useful information -- in fact,
>>it's the main reason why I wrote it.
>>My code is not elegant enough for an esteemed place on CRAN. I could
>>make it a lot better myself if I spent the time on it, but it works
>>well enough for me as it is, so in that sense, it ain't broke.
>>However, if anyone is interested in having such functionality my code
>>could be a good starting point.
> Inelegance is no excuse! However, I do understand the feeling of
> writing a nice little function, and not knowing exactly what to do
> with it: CRAN is crowded, and it does seem that an entire package
> just to support one or two simple functions is a bit of overkill.
> Can we work out a way to publish such things? Here's a proposal, with
> some serious flaws listed below. Can someone fix them?
> - It should be easy to publish a short article describing your nice
> little function, and making the code available to others. (I think
> this is already in place, either by publishing in RNews or Journal of
> Statistical Software, as a code snippet.)
> - It should be easy for users to get it, together with man pages,
> etc. This is not in place, unless you package your function and send
> it to CRAN. What I'd propose is that a volunteer should act as an
> editor of an "RNews" or "JSS" package, that contains *all* of the
> snippets that have been published in RNews or JSS.
> Some major flaws in this proposal are:
> - It needs a volunteer editor.
> - The editor would have to be organized, and willing to bug authors
> for updates whenever R changes and breaks their snippet (or makes it
> obsolete). If there's no one willing to do this, the editor should
> feel free to drop the snippet from the package.
> - There'd have to be a way for someone to see which snippets don't
> have maintainers so have been dropped, but could be resurrected.
> - If it's successful, there'll be name clashes. The editor should
> try to head these off from the start, by setting some rules for what
> kinds of names should be used for internal functions, and what public
> names should look like.
> - There aren't actually that many snippets being published, because
> it's hard work to convert something that works "well enough for me" to
> something we'd like others to see.
> Duncan Murdoch
> R-devel at stat.math.ethz.ch mailing list
Roger D. Peng
More information about the R-devel