[Rd] lme4 install, was Re: [R] Which executable is associated with R CMD INSTALL?

[Douglas Bates]

On 31 Oct 2006 12:05:21 +0100, Peter Dalgaard <p.dalgaard at biostat.ku.dk> wrote:
>
> [move to r-devel, put maintainer in loop]
>
> Patrick Connolly <p_connolly at ihug.co.nz> writes:
>
> > On Mon, 30-Oct-2006 at 04:44PM -0500, Duncan Murdoch wrote:
> >
> >
> > |> Try "R CMD printenv R_HOME" and you'll find which R home directory it is
> > |> using.  You can see a lot more with "R CMD printenv" or various options
> > |> to "R CMD config".
> >
> > Thanks for that information.  It knocks my theory on the head.  Pity
> > that, because I might have been able to do something about it if that
> > was the problem.  Now I'm at a loss to work out why lme4 installation
> > cannot find Matrix, and more strangely, why nothing else gave a
> > similar problem.  I think I've tried every version of lme4 and Matrix
> > that has emerged since R-2.4.0 has been released.
> >
> > The fact that no other Red hat user has a problem indicates the
> > problem is this end; but I'm short of ideas about where to look.
> > Looks like it's back to versions 6 months old -- like proprietary
> > software users have to put up with. :-(
>
> Hmmm, I can provoke this on SUSE too (I was going to ask you to try
> this!):
>
> mkdir foo
> MY_R=~/misc/r-release-branch/BUILD/bin/R
> R_LIBS=foo $MY_R --vanilla << EOF
>  options(repos=c(CRAN="http://cran.dk.r-project.org"))
>  install.packages("lme4", depend=TRUE)
> EOF
>
> This installs Matrix alright, then dies with
>
> Warning in install.packages("lme4", depend = TRUE) :
>          argument 'lib' is missing: using foo
> trying URL 'http://cran.dk.r-project.org/src/contrib/lme4_0.9975-8.tar.gz'
> Content type 'application/x-tar' length 235617 bytes
> opened URL
> ==================================================
> downloaded 230Kb
>
> * Installing *source* package 'lme4' ...
> ** libs
>
> gcc -I/usr/local/src/pd/r-release-branch/BUILD/include
>   -I/usr/local/src/pd/r-release-branch/BUILD/include
>   -I/usr/local/include -fpic -g -O2 -std=gnu99 -c Wishart.c -o Wishart.o
>
> In file included from Wishart.h:4,
>                  from Wishart.c:1:
> lme4_utils.h:9:20: Matrix.h: No such file or directory
> In file included from Wishart.h:4,
>                  from Wishart.c:1:
> lme4_utils.h:25: error: syntax error before "c"
> lme4_utils.h:25: warning: type defaults to `int' in declaration of `c'
> lme4_utils.h:25: warning: data definition has no type or storage class
> lme4_utils.h:163: error: syntax error before "cholmod_factor"
> lme4_utils.h:177: error: syntax error before "cholmod_factor"
> make: *** [Wishart.o] Error 1
> ERROR: compilation failed for package 'lme4'
>
> And Matrix.h is sitting happily in foo/Matrix/include/ but obviously
> not getting included. Should there have been a copy inside lme4?

Not according to my understanding of how the "LinkingTo:"
specification in the DESCRIPTION file is to work.

The headers for the C functions exported by the Matrix package are
defined Matrix/inst/include so they can be changed in one place if the
API changes.  The "LinkingTo:" specification is designed to put these
headers on the -I path at the time of compilation of the lme4 package
source code.  For some reason it didn't do that in this case.

> What is not obvious to me is how this can work anywhere...
>
> I also tried unpacking the lme4 directory from its tarball, dropping
> all files from the Matrix installed include dir into lme4/src and then
>
> $MY_R CMD INSTALL -l foo lme4
>
> but no go
>
> gcc -shared -L/usr/local/lib64 -o lme4.so Matrix_stubs.o Wishart.o
> glmer.o init.o lme4_utils.o lmer.o local_stubs.o pedigree.o
> -L/usr/local/src/pd/r-release-branch/BUILD/lib -lRlapack
> -L/usr/local/src/pd/r-release-branch/BUILD/lib -lRblas -lg2c -lm
> -lgcc_s
>
> local_stubs.o(.text+0x0): In function `M_numeric_as_chm_dense':
> /home/bs/pd/tmp/lme4/src/Matrix_stubs.c:420: multiple definition of `M_numeric_as_chm_dense'
> Matrix_stubs.o(.text+0x0):/home/bs/pd/tmp/lme4/src/Matrix_stubs.c:420: first defined here

You don't want to compile both Matrix_stubs.c and local_stubs.c.  All
that local_stubs.c does is include Matrix_stubs.c, the reason being
that those wrapper functions are needed locally but it is much cleaner
if they are defined in only one place.  The file Matrix_stubs.c is
never compiled by itself.  It is just there to provide the definitions
of the functions.  Check Luke's documentation and examples for using
R_RegisterCCallable and R_GetCCallable.


> local_stubs.o(.text+0x70): In function `M_dpoMatrix_chol':
> /home/bs/pd/tmp/lme4/src/Matrix_stubs.c:412: multiple definition of `M_dpoMatrix_chol'
> Matrix_stubs.o(.text+0x70):/home/bs/pd/tmp/lme4/src/Matrix_stubs.c:412: first defined here
> ..etc..
>
> --
>    O__  ---- Peter Dalgaard             Øster Farimagsgade 5, Entr.B
>   c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
>  (*) \(*) -- University of Copenhagen   Denmark          Ph:  (+45) 35327918
> ~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)                  FAX: (+45) 35327907
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>