[Phylobase-devl] Rcpp and OS X compiliation

[Dirk Eddelbuettel]

Brian,

On 17 December 2009 at 11:43, Brian O'Meara wrote:
| On Dec 16, 2009, at 2:15 PM, Dirk Eddelbuettel wrote:
| 
| >
| > On 16 December 2009 at 10:16, Peter Cowan wrote:
| > | On Dec 16, 2009, at 7:06 AM, Dirk Eddelbuettel wrote:
| > |
| > | >
| > | > On 16 December 2009 at 09:19, Hilmar Lapp wrote:
| > | > | On Dec 16, 2009, at 7:36 AM, Dirk Eddelbuettel wrote:
| > | > |
| > | > | > What I don't understand is why the lib that R loads itself  
| > (in the
| > | > | > Rcpp/libs/
| > | > | > directory) is a .so.  Is that how it is supposed to be OS X?
| > | > |
| > | > |
| > | > | No, it's the .dylib that it's supposed to be loading.
| > | >
| > | > Ok, maybe Hilmar, Brian and I should take this off-list then.   
| > My best idea
| > | > (being the OS X ignoramus that I am) were to look at two other  
| > (and small !!)
| > | > package that also link against Rcpp:
| > | >
| > | > -- Earth Mover Distance
| > | >    https://r-forge.r-project.org/projects/earthmovdist/
| > | >    No log at R-Forge because no recent changes
| > | >
| > | > -- RInside
| > | >    https://r-forge.r-project.org/projects/rinside/
| > | >    No log at R-Forge because I had a stupid mistake in a change
| > | >    on Sunday that I fixed yesterday -- will work now but hasn't  
| > been
| > | >    rebuilt on R-Forge yet
| > | >
| > | > We simply need a smaller working example, see the exact gcc  
| > invocation on OS
| > | > X and then compare what the more complicated (as ncl enters etc)  
| > phylobase
| > | > does.  This ain't brain surgery and we will get this fixed.
| > |
| > | I don't mind if this stays on list.  Many of us are OS X users,  
| > but don't understand the build process very well.
| >
| > Ok. I keep railing then :) I understand the build process [ and  
| > there is only
| > one lesson: let R do as much as possible and don;t get in the way;  
| > Simon
| > Urbanek drilled that into my head with some success ... ] but am  
| > clueless
| > with OS X.
| >
| > | I've had similar problems in the past, and attributed it to NCL  
| > not being
| > | build using the R make variables.  This is what I *thought* was  
| > going on,
| > | it might be completely off base.  From 10.5 to 10.6 Apple changed  
| > the
| > | default build architecture from i386 to x86_64.  R has typically  
| > been
| > | distributed as a 32 bit (i386) build.  So all was fine when NCL was
| > | building as the default in 10.5, but when moving to 10.6 32 bit R  
| > and the
| > | default build doesn't work.  So I use the 64bit build of R to  
| > install and
| > | run phylobase and it works fine.  But, doesn't constitute a general
| > | solution.
| >
| > I got to this via the automated generation of binary CRAN package  
| > in .deb
| > form which I do at http://debian.cran.r-project.org.  phylobase failed
| > miserably, and I blame ncl.  The configure files are old and should  
| > get
| > replaced.  For Linux, I essentially sidestepped ncl's configure and  
| > used
| > Makefile.win with one modification. Then it worked. That would not  
| > work for
| > OS X due to multiarch.  But we will get to that.
| >
| > That said, we currently have an issue with a part I am maintaining  
| > (Rcpp) and
| > I would like that squared off first.
| >
| > Hence my request (see above in the quoted part) for an OS X  
| > volunteer to
| > look with me off-line at success / failure of RInside or EarthMovDist.
| 
| I would be able to do this after Jan 9 (trying to get a grant in by  
| NSF's deadline), but someone else should feel free to jump in.
| 
| Thanks, all.

I am still game. I see a three-pronged approach:

  i)    Suggestion Let's fix NCL and have a locally built version outside 
        of phylobase on all test platforms

  ii)   With that, fix/simplify Rcpp usage. Should be easy

  iii)  With that, get back to i) and clean up NCL's configure.

I think it would be easier to disentangle Rcpp and NCL inside phylobase while
we make this more.

Who else can work with us on this, and in particular provide OS X skillz and
testing?

Dirk

 
| Brian
| 
| 
| >
| > | I'm sure the make files for NCL could be greatly simplified which  
| > could
| > | help. But, it is currently beyond my abilities, and I don't  
| > currently have
| > | the time to learn the make.
| >
| > Ideally we would able to assume that the user has already installed  
| > ncl. In
| > the real world down here that may never be the case, alas.
| >
| > Dirk
| >
| > -- 
| > Three out of two people have difficulties with fractions.
| 
| ------------------------------------------------------
| Brian O'Meara
| http://www.brianomeara.info
| Assistant Prof.
| Dept. Ecology & Evolutionary Biology
| U. of Tennessee, Knoxville
| 

-- 
Three out of two people have difficulties with fractions.