[Bioc-devel] Sweave changes (keep.source = TRUE or FALSE?)

Friedrich Leisch friedrich.leisch at stat.uni-muenchen.de
Wed Dec 6 17:07:03 CET 2006


>>>>> On Tue, 05 Dec 2006 15:27:02 +0000,
>>>>> Wolfgang Huber (WH) wrote:

[...]

  > This discussion seems to move into the direction of having different
  > "flavours" of R. In addition to --vanilla, we could add a --bioc
  > flavour. Alternatively, a really minimal "Biobase"-type package that
  > needs to be loaded for any use with Bioconductor and contained such
  > goodies as setting the Sweave default behaviour and the biocLite script?

  > I am not sure whether this a direction that we really want to take, and
  > I am here not suggesting actually taking it, but isn't this the logical
  > consequence?

Unfortunately yes ... but I'd say this is a more general problem and
may also apply a to other user groups, so I would not hardcode a
--bioc exception into the code.

IMO R must be allowed to change, but if possible it should be easy to
protect yourself against changes when possible. I always considered
the parse&deparse behavior of Sweave a bug for which I had no simple
solution (and am somewhat surprised that there are people who think it
is a feature). So now that the bug is fixed, it would be natural for
me to make that available as default.

The unix way of solving something like that is having site-wide
configuration files which the sysadmin sets for all users. R has the
same concept, but R CMD check and build do not use it (to make the
process as portable as possible). That behavior is good as a default,
but we should allow users and sysadmins to override it, i.e., load
other configurations especially in R CMD build.

The basic concept was always that

1) The author runs R CMD build, and it is OK if that works only on the
   author's machine, especially for vignettes. It is strange the
   building vignettes finds and happily uses the local TeX
   configuration including style and bibtes files, but ignores the
   local R configuration.

2) Everybody, especially repositories, should be able to run R CMD
   check on a package.

I'd propose to fix 1) in R 2.5.0, i.e., add an option --no-vanilla to
R CMD build such that local configurations are read.

Best,
Fritz



More information about the Bioc-devel mailing list