Does it still happen with R-2.15.0 patched? Maybe an old version of
Matrix in a library that is first in the search path? Unfortunately
glmnet does not import from Matrix and search path issues are likely to
happen. This needs to be sorted out on R-forge , I think.
Best,
Uwe Ligges
On 01.06.2012 11:19, Nick Sabbe wrote:
Dear all,
I've recently had some issues getting my package to successfully "check".
This was on R-Forge, so it's not obvious for me to provide SessionInfo or
the likes (if necessary, Stefan can chime in?).
After some research (mainly by Stefan Theussler, driving force behind
R-Forge), this turned out to be the root cause:
On R-Forge, the version of R installed was the latest release (2.15.0),
which includes Matrix 1.0-5 as a recommended package.
My own package depends upon glmnet, which itself depends upon Matrix 1.0-6.
(Note: to avoid issues with that, as a test I've also included an explicit
dependency upon Matrix 1.0-6 in my own package, but this didn't change
anything).
During build and most of R CMD check (with the -as-cran flag), everything
worked out fine (which suggests to me that among others, Matrix 1.0-6 is
present and can be loaded, since one of the checks is that the package and
all its dependencies can be loaded), but while running the examples, somehow
loading of the package failed, with an error message that Matrix 1.0-6 could
not be loaded. Here's the relevant part of the log:
library('addendum')
Loading required package: glmnet
Error: package 'Matrix' 1.0-5 was found, but>= 1.0.6 is required by
'glmnet'
Execution halted
Run time: 92.43 seconds.
There are some issues with this comparison, but in the Windows build/check
chain on R-Forge, R2.15.0 patched was installed (including Matrix 1.0-6) and
the build _was_ successful.
Stefan suspected this to be an artifact of Matrix being a recommended
package (so R CMD check would somehow prefer the version installed with R),
and to a certain degree the successful build/check with R patched confirms
this suspicion.
I've neither read up on this nor tried other options, but if this is indeed
the case then I suggest to alter this behaviour. Unless, of course, there is
any reason to treat "recommended" packages any different wrt versioning and
R CMD check.
I suspect other people may run into this sooner or later.
If not or if my/our analysis is wrong/incomplete I will gladly stand
corrected.
I'll leave it to you gurus to judge.
Thanks for R! (And another shout out to Stefan, who diligently helped me)
Nick Sabbe
--
ping: nick.sa...@ugent.be
link:<http://biomath.ugent.be/> http://biomath.ugent.be
wink: A1.056, Coupure Links 653, 9000 Gent
ring: 09/264.59.36
-- Do Not Disapprove
[[alternative HTML version deleted]]
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
