Re: [Rd] R CMD check warning about boot ?

From: Dirk Eddelbuettel <edd_at_debian.org>
Date: Fri 01 Dec 2006 - 13:42:49 GMT

( I added Kurt + Uwe back here again, they had replied earlier in the thread.)

On 1 December 2006 at 11:27, Prof Brian Ripley wrote:
| On Fri, 1 Dec 2006, Martin Maechler wrote:
|
| >>>>>> "Dirk" == Dirk Eddelbuettel <edd@debian.org>

| >>>>>> on Thu, 30 Nov 2006 08:06:46 -0600 writes:
| >
| > [.............]
| >
| > Dirk> The boot package loads fine, and is current. Why does
| > Dirk> R CMD check think it is an error that boot happens to
| > Dirk> live in a particular directory, for as long as that
| > Dirk> directory is known to .libPaths() ?
| >
| > There's quite a difference between
| > .libPaths() in interactive and in "R CMD check" use :
| >
| > 'R CMD check' does not use your R_ENVIRON where you probably set
| > your R_LIBS .
|
| That's not the issue here. Even if R_LIBS is set, R CMD check looks for
| the copies of the standard and recommended packages in .Library
| (specifically .check_Rd_xrefs does): that is a safety measure as those are
| the ones against which the checks should be done even if you have
| development versions elsewhere in R_LIBS. (Not my design, but one that
| seems highly defensible.)

Recall that on Debian (and hence Ubuntu etc), and based on a suggestion by KH+FL made during DSC2003, we set R_LIBS in the _global_ Renviron file:

edd@basebud:~> grep R_LIBS /usr/lib/R/etc/Renviron R_LIBS=${R_LIBS-'/usr/local/lib/R/site-library:/usr/lib/R/site-library:/usr/lib/R/library'}

So Martin's point about a 'your R_LIBS' does not really hold. It's not via ~/.Renviron that this is governed.

Moreover, I have for several years now configured R itself (aka "r-base-core" in Debian package space) to build --without-recommended-packages so if we put the two together then a) R knows it cannot assume recommended packages, and b) every function call using .libPaths() always saw the three values set in R_LIBS --- which is why on the particular setup I use at work,

        /usr/local/lib/R/site-library holds "everything", incl rec'd pkgs

        /usr/lib/R/site-library is empty, no r-cran-* packages here

	/usr/lib/R/library		has what r-base-core installs
					but no recommended pkgs either

This has the advantage that the 'littler' shell wrapper update.r will update everything (in /usr/local/lib/R/site-library) with one simple call; apt-get and Debian take care of /usr/lib/R/library, and there is no overlap between R's update.packages() and Debian. It's more radical than what I tried before, but it works *great*.

Apart from the R CMD check whine. I think I should patch that, but I'd need a hint to where I should start looking.

Or else I have to cave in and put the recommended packages back into /usr/lib/R/library --- and violate in spirit the implicit contract of leaving /usr et al to the package management system, and /usr/local to the user. I don't like breaking that contract.

| There is no provision in the R scripts to install the recommended
| packages elsewhere than .Library, but I believe Dirk has a post-install
| script that breaks the assumptions so this is a Debian-packaging-specific
| issue.

Nothing post-install. For normal Debian packages, we have

	r-base-core	
	+ recommended pkgs	using		/usr/lib/R/library

	r-cran-* from Debian	using		/usr/lib/R/site-library

	R/user installed pkgs	using		/usr/local/lib/R/site-library

where the last one is the default by virtue of its first place in R_LIBS.  

| > I vaguely remember that there a good reasons for this behavior,
| > but I don't recall them at the moment.
|
| R CMD does not make use of ~/.Renviron nor R_HOME/etc/Renviron.site:
| it is like R --vanilla, which is what is used for a lot of batch testing
| code.

Right, but I had used R_HOME/etc/Renviron, not $R_HOME/etc/Renviron.site (though arguably "Renviron.site in spirit").

Sorry for the long post, but this question of how to "best" manage R and Debian packages on Debian/Ubuntu also pops up frequently on the r-sig-debian list (or, as seen this week, in Doug and my mail box, even from a formerly prominent CRAN author).

Comments highly welcome!

Cheers from snowy Chicago, Dirk

-- 
Hell, there are no rules here - we're trying to accomplish something. 
                                                  -- Thomas A. Edison

______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
Received on Sat Dec 02 00:46:31 2006

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.1.8, at Fri 01 Dec 2006 - 17:30:49 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-devel. Please read the posting guide before posting to the list.