[GenABEL-dev] Missing file for unit test in GenABEL

[Andreas Tille]

Hi Lennart,

On Wed, Jul 02, 2014 at 08:30:09AM +0200, L.C. Karssen wrote:
> > ??? I admit it sounds pretty strange to me that there should be a policy
> > to avoid testing software.  At least in Debian we are doing the contrary
> > and try to add tests whereever possible that can run at package build
> > time and in periodical tests of the full archive.
> 
> Maybe we should try to enable the tests again and see what the CRAN
> people say?

+1
 
> >> 1. use CRAN release, no unit tests
> >>
> >> 2. use SVN tag for latest release - easiest for us (but I understand you'd prefer tarball)
> >>
> >> 3. we provide the tarball for latest release, including RUnit part on the code on our web-site http://www.genabel.org/
> >>
> >> what do you think? I am somewhat afraid that it may be difficult to implement/keep coherent option 3. - to be fair, with me being overstretched, we struggle a bit with keeping sustainable maintenance of the GenABEL-package. 
> >>
> >> Any other options, anyone?
> 
> 
> Using a simple SVN checkout is not going to work. Currently we use a
> shell script (also in SVN,
> pkg/GenABEL-general/distrib_scripts/makedistrib_GenABEL.sh) to prepare
> the package before releasing to CRAN. This script copies some stuff
> around (e.g. some .cpp/.h files that are shared with DatABEL, but also
> files that are in subdirectories in GenABEL/src/ in SVN (e.g. GAlib/,
> ITERlib/), but get copied to GenABEL/src/. I guess that a proper
> Makevars file would obsolete the need for this file juggling. It's been
> on my list for quite some time (also for other packages), but as Yurii
> mentioned, we're a bit short on man power.

... as always.

> > I admit that having a canonical way to obtain a tarball from CRAN is
> > quite attractive.  Is there any chance to somehow "hide" the tests -
> > perhaps by renaming the directory the test files are located in - before
> > uploading the tarball to CRAN so the test will not run there but we can
> > "uncover" the files again and run the tests.  For me this sounds like a
> > good compromise to fullfill all needs without changing the workflow
> > drastically.
> 
> That should definitely be possible. I'll have a look.

So my prefered solution would be in this sequence

  1. Follow your suggestion to re-enable the tests on CRAN
  2. Follow my suggestion to provide "hidden" tests
  3. anything else

> > If you might document this in some README file also other potential
> > distributors and users could profit from this. 
> 
> Thanks for the suggestion. We should definitely do that.
> 
> Thanks for thinking along!

Thanks also to you for the constructive cooperation

      Andreas.

-- 
http://fam-tille.de