Peter Dalgaard wrote:
Dirk Eddelbuettel wrote:
On 14 October 2009 at 18:54, Duncan Murdoch wrote:
| On 14/10/2009 6:42 PM, Dirk Eddelbuettel wrote:
| > Maybe I missed an announcement somewhere but having started with
the usual
| > practice of putting R pre-releases into Debian unstable leading up
to the
| > release of the next R version, I now find myself unable to do my
daily
| > package maintenance as I only see VR 7.2-49 with its
| > | > Depends: R (≥ 2.5.0), R (≤ 2.9.2)
| > | > preventing use on R 2.10.0. That stopped the update of the
Debian package
| > r-cran-effects (which wants MASS) in its tracks. | > | > I see
the somewhat hidden MASS 7.3-0 and 7.3-1 in
| > | > http://cran.r-project.org/src/contrib/Archive/MASS/
| > | > so should I be using these? Along with the other components
of VR ?
| | Yes, see this news item:
| | o Bundles of packages are deprecated. See 'Writing R
| Extensions' for the steps needed to unbundle a bundle.
That of course I knew as a reader of both your NEWS feed [1] as well
as my
own CRANberries [2] which carried e.g. Paul Gilbert's dse unbundling.
But that still does not answer my new MASS, nnet, spatial, class
release are
not on CRAN in the usual location (eg $CRAN/src/contrib/) but socked
away in
the Archive/ subdir.
| The VR bundle contained packages class, MASS, nnet, and spatial,
which | are now separate recommended packages, whose source is
included in the R | source tarball, or on CRAN in
| | http://cran.r-project.org/src/contrib/2.10.0/Recommended/
I guess the fact that VR is special and included in R itself explains
part of
it. I forget about that angle.
I don't think that's it. Rather, the point is that packages directly
under src/contrib/ refers to the CURRENT version of R and gets updated
only on release of the next version. The MASS in src/contrib/2.10.0 has
Depends: R (>= 2.10.0), grDevices, graphics, stats, utils
which means that trouble will occur if it gets put into src/contrib at
this stage and someone tries to install it.
This does occasionally also affect non-recommended packages, currently:
orientlib, which is in src/contrib/2.10.0/Other.
So, the point is that if you are building stuff for unreleased versions,
the process should be taught to look in the appropriate part of CRAN.
As I'm never quite sure what you're up to, I can't tell whether the
issue is on your side or ours. There does seem to be some potential
issues, both with dependency searches and with the effects of
unbundling. (Do we, e.g., want a virtual VR package which just depends
on the packages that used to be inside, for the benefit of packages that
depends on VR?)
Can packages depend on VR? Untested, but at least there is no such
package on CRAN and dependency rules should mention the package, not the
bundle.
The PACKAGES file in src/contrib also points to the packages in
src/contrib/2.10.0. It is up to Dirk to use the information in PACKAGES
correctly as it is done for the Windows and Mac binaries on CRAN.
Best,
Uwe
-p
Dirk
[1] http://developer.r-project.org/blosxom.cgi/R-devel
[2] http://dirk.eddelbuettel.com/cranberries/
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
