[Rd] R 2.15.0 alpha: R CMD check --as-cran / tools:::..check_package_CRAN_incoming() crash

Henrik Bengtsson hb at biostat.ucsf.edu
Tue Mar 6 03:33:16 CET 2012

For what it's worth, with

  R --no-init-file CMD check --as-cran ${pkg}_${version}.tar.gz

on R version 2.15.0 alpha (2012-03-03 r58572) on Windows I just
managed to generate a crash:

Checking package affxparser...
* using log directory 'X:/affxparser,BioC-devel/R2.15.0/affxparser.Rcheck'
* using R version 2.15.0 alpha (2012-03-03 r58572)
* using platform: x86_64-pc-mingw32 (64-bit)
* using session charset: ISO8859-1
* checking for file 'affxparser/DESCRIPTION' ... OK
* this is package 'affxparser' version '1.27.4'
* checking CRAN incoming feasibility ...Warning in
url(sprintf("%s/src/contrib/Meta/archive.rds", CRAN), "rb") :
  unable to connect to 'CRAN.R-project.org' on port 80.
Error in url(sprintf("%s/src/contrib/Meta/archive.rds", CRAN), "rb") :
  cannot open the connection
Execution halted

I can *not* reproduce it.  Also, if I disable internet access, it
continues with a warning:

* this is package 'affxparser' version '1.27.4'
* checking CRAN incoming feasibility ...Warning in
url(sprintf("%s/src/contrib/PACKAGES.gz", u), "rb") :
  unable to resolve 'CRAN.R-project.org'
NB: need Internet access to use CRAN incoming checks
* checking package namespace information ... OK

Looking at tools:::..check_package_CRAN_incoming() [in
src/library/tools/R/QC.R], there are few places where different online

URL 1: %s/src/contrib/PACKAGES.gz,
URL 2: %s/src/contrib/Meta/archive.rds,
URL 3: %s/src/contrib/Meta/current.rds,
URL 4: %s/web/packages/packages.rds

are downloaded, but it is only the first one that is downloaded with
exception handling.  Thus, if URL 2, 3 or 4 is broken/being updated,
or internet access goes down, the above crash would occur.

Also, in tools:::.check_Rd_xrefs() another URL is downloaded assuming
working internet access.

Just feedback. I can live the above, but it could be that others
report on this in the future.


More information about the R-devel mailing list