在狭窄情况下如何调试R CMD检查失败的情况

Question

Apologies for yet another package build debugging, but I'm completely stumped here.

Using R-devel.sh CMD check taRifx.geo_1.0.6.tar.gz --as-cran on (X)Ubuntu 14.04 LTS, installed via this script , I get the following error:

* checking examples ... ERROR
Running examples in ‘taRifx.geo-Ex.R’ failed
The error occurred in:
[ ...SNIP... ]
> ### * </HEADER>
> library('taRifx.geo')
Loading required package: sp
Error in loadNamespace(i, c(lib.loc, .libPaths()), versionCheck = vI[[i]]) : 
  there is no package called ‘lattice’
Error: package ‘sp’ could not be loaded
Execution halted

The manual for package checking says:

Before using these tools, please check that your package can be installed (which checked it can be loaded). R CMD check will inter alia do this, but you may get more detailed error messages doing the install directly.

But when I do that (being sure to use R-devel, loaded with --vanilla ), I get no error:

# 1) 
install.packages("taRifx.geo_1.0.6.tar.gz",repos=NULL)
# 2) 
library(taRifx.geo)
# 3) Profit.

If I run R CMD check through RStudio or command line R (but not the -devel version), I also get no error.

Thing is, when I search the entire project's code for "lattice", I can't find any case where I use it, even in the example files. lattice is installed and I've tried checking with lattice in Suggests and without.

Project code here: https://github.com/gsk3/taRifx.geo

The check logs are here .

Now for the question: How do I go about debugging this? I have some thoughts as to where the problem might lie (Depends rather than Imports for a few packages, and the spatial packages tend to have very intertangled namespaces), but I can't really begin to solve it unless I can reproduce the error in an environment other than R CMD check run from the command line via R-devel.sh .

This, by the way, is Under development (unstable) (2014-05-26 r65771) -- "Unsuffered Consequences" . Appropriately named.

Edit

Here are the contents of /usr/local/lib/R/site-library :

akima       digest      labeling  plyr          RJSONIO      stringr
bitops      evaluate    lattice   proto         R.methodsS3  survival
boot        fields      mapproj   pspline       R.oo         taRifx
caTools     functional  maps      RandomFields  RSQLite      taRifx.geo
cluster     gdata       maptools  RColorBrewer  R.utils      testthat
colorspace  geoR        MASS      Rcpp          scales       whisker
data.table  ggplot2     Matrix    RCurl         shapefiles   xtable
DBI         gtable      memoise   reshape2      sp
devtools    gtools      munsell   rgdal         spam
dichromat   httr        nlme      rgeos         splancs

The lattice description file says:

Depends: R (>= 2.15.1)
Built: R 3.2.0; x86_64-unknown-linux-gnu; 2014-05-29 17:14:03 UTC; unix

Still the same error on the CMD check .

Answer 1

I had been attempting to install to the site-wide library at "/usr/local/lib/R/site-library" which did not solve it, but I instead brute-forced a solution.

First, find all the places R might look:

> .libPaths()
[1] "/usr/local/lib/R/site-library"       
[2] "/usr/lib/R/site-library"             
[3] "/usr/lib/R/library"                  
[4] "/usr/local/lib/R-devel/lib/R/library"

Then, as root, install to each of those in turn until the error goes away. As it turns out, it's "/usr/local/lib/R-devel/lib/R/library" :

install.packages(c('lattice','sp'),lib="/usr/local/lib/R-devel/lib/R/library")

Then q() that R-devel session and run R-devel.sh CMD check mypackage.tar.gz --as-cran .

As for why it needs lattice in the first place, I suspect it's pulling in an .Rdata file at launch which has lattice in the namespace.

Many thanks to @DirkEddelbuettel and @GavinSimpson for the various shoves in the right direction.