Peter D. Cowan
pdc at berkeley.edu
Mon Aug 24 22:14:08 CEST 2009
On Mon, Aug 24, 2009 at 01:06:38PM -0700, Jim Regetz wrote:
> Peter D. Cowan wrote:
> >The Mac OS X build finally completed!! Would someone with a mac running 10.4, and R 2.9.2 please confirm that the following works.
> >install.packages("phylobase", type="mac.binary", repos="http://R-Forge.R-project.org")
> I just did some testing on a PowerPC Mac running OS X 10.4 and R 2.9.2.
> First of all, after installing Xcode 2.5.0 Developer Tools on this
> otherwise clean test machine, the *source* installation of phylobase
> went without a hitch.
Thanks for doing this!
> Unfortunately, the binary install didn't work. However, the issue I
> encountered seems to involve the architecture rather than the OS
> version. The installation step itself doesn't actually produce any
> errors or warnings, but when loading the package I get this:
> > library(phylobase)
> Error: package "phylobase" is not installed for 'arch=ppc'
Okay, this is almost the same error that I reported with 64bit versions of R.
> After surveying a few other R-Forge packages and their corresponding
> Mac build logs, I wonder if there is some problem with universal
> builds of R-Forge packages containing C++ code? I'm speculating, but
> I discovered two other packages (rgl and RInline) that produce this
> same arch=ppc error when I attempted to load them, and both (like
> phylobase) include C++ code. Moreover, the R-Forge build logs for
> several packages having only C/Fortran code report cpp/gfortran
> compiler statements targeting both i386 and ppc architectures,
> whereas only i386 shows up in g++ statements in the R-Forge build
> logs of phylobase, rgl, and RInline. (I should point out that some
> of the packages on R-Forge with only C/Fortran code also failed to
> load on this machine after installing from binary, but with
> completely different errors.)
That's pretty much the conclusion that I came to regarding the 64bit issue, but I wasn't as enterprising about testing other packages.
> One bit of potentially good news is that the rgl binary on CRAN
> installed fine for me, so this whole problem might also be a
> non-issue for phylobase on CRAN as opposed to R-Forge.
Well that at least gives us hope. I'd say we wait until the package is submitted to CRAN, and then retest w/ ppc and 64 bit. If it works then we can file a feature request with R-Forge. If not perhaps the issue is with the configuration files in NCL or more likely the ones generated by Rcpp.
> As for binary installation on an Intel Mac, I don't have one that I
> can readily downgrade to OS X 10.4 for testing. But frankly, given
> that the build itself seems to succeed on all tested platforms now,
> I think the odds are pretty good that CRAN will produce working
> binaries. Just don't quote me on that :)
> >We may close #269 yet.
Alas, it appears I was hasty when claiming this.
> >Phylobase-devl mailing list
> >Phylobase-devl at lists.r-forge.r-project.org
More information about the Phylobase-devl