Skip to content

R CMD check error

7 messages · Martyn Plummer, Martin Maechler, Terry Therneau

Original
Terry Therneau
# 
I have a local library which depends on the expm library.  The expm library is loaded into 
my personal space and I have the environment variable R_LIBS_USER set appropriately.  The 
command "library(expm)" works just fine from the command line, and in fact the package 
works if I do the source() and dyn.load() commands by hand.

The following sequence works:

tmt% R CMD build --no-build-vignettes hmm
tmt% R CMD INSTALL hmm*gz
tmt% R
 > library(hmm)
 >  run some commands from the hmm library

But "R CMD check hmm.gz" fails with
   ERROR: dependency ?expm? is not available for package ?hmm?
* removing ?/people/biostat2/therneau/consult/alzheimer/hmm.Rcheck/hmm?

The R CMD build command fails similarly if I let it try to build the vignettes.

What's up?

Terry T


 > sessionInfo()
R version 3.3.1 (2016-06-21)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: CentOS release 6.8 (Final)
Martyn Plummer
#
On Wed, 2017-02-08 at 15:51 -0600, Therneau, Terry M., Ph.D. wrote:
If you are setting the environment variable R_LIBS_USER in
R_HOME/site/Renviron.site or in .Renviron then this file will not be
read when you run R CMD check or R CMD build, as R is then run with --
vanilla which implies --no-environ.?

You also need to set it in these files:

~/.R/build.Renviron
~/.R/check.Renviron

See R-exts section 1.3 and ?Startup.

Martyn
Terry Therneau
# 
Martin,
  I am aware of --vanilla; I use it myself for some testing.  In this case R_LIBS_USER was 
set externally (part of my login) and does not involve any of the R scripts.  That means 
it is inherited by any subprocess.  For example:

tmt1495% R --vanilla --no-environ

R version 3.3.1 (2016-06-21) -- "Bug in Your Hair"
Copyright (C) 2016 The R Foundation for Statistical Computing
Platform: x86_64-pc-linux-gnu (64-bit)
R_LIBS_SITE=
R_LIBS_USER=/people/biostat2/therneau/Rlib

So, per the manual R CMD check inherits the path.  The question is why does it ignore it?

Terry T.
On 02/09/2017 02:54 AM, Martyn Plummer wrote:
Martyn Plummer
#
On Thu, 2017-02-09 at 09:52 -0600, Therneau, Terry M., Ph.D. wrote:
Hmmm. Perhaps it is being overwritten. ?Does this work?

$ export R_CHECK_ENVIRON=
$ R CMD check hmm.gz

Martyn
Terry Therneau
# 
Martyn,
   No, that didn't work.
   One other thing in the mix (which I don't think is the issue) is that I call one of the 
C-entry points of expm.  So the DESCRIPTION file imports expm, the NAMESPACE file imports 
expm, and the init.c file is

#include "R.h"
#include "R_ext/Rdynload.h"

/* Interface to expm package. */
typedef enum {Ward_2, Ward_1, Ward_buggy_octave} precond_type;
void (*expm)(double *x, int n, double *z, precond_type precond_kind);
void R_init_hmm(DllInfo *dll)
{
     expm = (void (*)) R_GetCCallable("expm", "expm");
}

I don't expect that this is the problem since I stole the above almost verbatim from the 
msm package.

Terry T.
On 02/09/2017 11:23 AM, Martyn Plummer wrote:
Martin Maechler
# 
> Martyn,
    > No, that didn't work.
    > One other thing in the mix (which I don't think is the issue) is that I call one of the 
    > C-entry points of expm.  So the DESCRIPTION file imports expm, the NAMESPACE file imports 
    > expm, and the init.c file is

    > #include "R.h"
    > #include "R_ext/Rdynload.h"

    > /* Interface to expm package. */
    > typedef enum {Ward_2, Ward_1, Ward_buggy_octave} precond_type;
    > void (*expm)(double *x, int n, double *z, precond_type precond_kind);
    > void R_init_hmm(DllInfo *dll)
    > {
    >      expm = (void (*)) R_GetCCallable("expm", "expm");
    > }

    > I don't expect that this is the problem since I stole the
    > above almost verbatim from the msm package.

    > Terry T.

Hmm.  Yes, I can see that the CRAN package  msm  does do this, indeed.

It is interesting if/why that does not produce any notes or rather even warnings.
In principle, if you use the C API of 'expm'  you should use
  'LinkingTo: expm'

see *the* manual, specifically the section

    https://cran.r-project.org/doc/manuals/r-release/R-exts.html#Linking-to-native-routines-in-other-packages

and that section does mention that (unfortunately in my view)
you also should use 'Imports:' or 'Depends:' in addition to the 'LinkingTo:'

Note howver that  'expm' would not have to mentioned 
in the NAMESPACE file.... unless your R functions do use some of
expm's R level functionality.


Martin
Terry Therneau
# 
Martin,
   That was it-  I forgot the "LinkingTo" line.  I had read that section of the manual 
twice in the last 2 days, yet somehow missed that critical line both times.  And even 
worse, the final sentence of said section references my own coxme package as an example of 
how to do it correctly!

Thank you all for the help.  My only remaining defense, but a very weak one, is that the 
error message could have been better since it led me to believe that R couldn't find the 
library at all.

Terry Therneau
On 02/10/2017 10:26 AM, Martin Maechler wrote: