Skip to content

'R CMD build <mypkg>' fails to 'make clean' when <mypkg> has a Makefile

5 messages · Hervé Pagès, Brian Ripley

#
Hi,

This problem has already been reported here more than 1 year ago by Brian J. Lopes:

  https://stat.ethz.ch/pipermail/r-sig-mac/2007-May/003863.html

It's still present in R 2.8.0 beta (2008-10-07 r46631) on Mac OS X as you can see
on the last build/check report for the xps package (Biosconductor):

  http://bioconductor.org/checkResults/2.3/bioc-LATEST/xps/pitt-buildsrc.html

xps has a Makefile in its src/ folder (with a "clean:" target) and 'make clean'
works fine for me if I run it manually from within the src/ folder on build
machine pitt. However, 'R CMD build' fails to clean the src/ folder because
it's apparently trying to access
  /Library/Frameworks/R.framework/Versions/2.8/Resources/etc/Makeconf
instead of
  /Library/Frameworks/R.framework/Versions/2.8/Resources/etc/<arch>/Makeconf

Are there any chance that this could be addressed?

Thanks in advance,
H.
#
On Fri, 17 Oct 2008, Herve Pages wrote:

            
Maybe, but (AFAICS, see also below) no one submitted a patch to the 
appropriate list (r-devel) nor even a 'wish' to R-bugs.
Why are you using a test version over a week old?  CRAN tests on the 
current (when the test round started) version of R.
A chance, but not in code freeze for 2.8.0.  (Your timing is the worse 
possible.)

That's an unusual setup (developing on a system with sub-architectures). 
If you submit a tested patch it is more likely to be 'addressed' once the 
R-2-8-branch is open again.

  
    
2 days later
#
Hi,
Prof Brian Ripley wrote:
[...]
Are you saying that developing on Tiger (or Leopard) is unusual?
Thanks!

H.
#
On Mon, 20 Oct 2008, Herve Pages wrote:

            
Well, yes (and we know that because of the non-standard tarballs those 
systems produce by default and which R CMD build now works around).

But my observation was based on the fact that everyone I know who develops 
on Mac OS X (and almost everyone on Windows) builds R from the sources to 
suit their own preferences (and since that will be for the one 
architecture they are working on and sub-architectures are not the 
default), such people don't have sub-architectures.  It's not just Mac OS: 
I have 32- and 64-bit versions on Linux and Solaris, but I keep them 
separate for development (and e.g. RH/Fedora has gone a different route 
for multi-architecture support that is tantamount to separate installs 
with a common front-end script).
I think this is probably covered in changes now made in R-devel, but 
please do check for us.

  
    
2 days later
#
Hi,
Prof Brian Ripley wrote:
[...]
We will, as soon as our builds are up and running with R-2.9 on macosx.
I'll post again here to confirm.

Thanks!
H.