[Rd] simplifying number of R installations on disk

Gabor Grothendieck ggrothendieck at gmail.com
Sun Dec 27 02:16:19 CET 2009


On Sat, Dec 26, 2009 at 6:26 PM, Charlotte Maia <maiagx at gmail.com> wrote:
> I think it's a very bad idea to truncate version numbers.
> Version numbers are important, and there may be a critical difference
> x.y.0 and x.y.1.
> Plus there is no guarantee that the updated version is better than the
> previous one.
> Hence a programmer may need to go back to x.y.0 until x.y.2 is released.
>
> Perhaps there should be an option somewhere to remove existing
> installations (maybe there is????), however by default, installing a
> new version of a programming language, should not uninstall existing
> versions.

A new version of R comes out almost daily or weekly.  I don't think
many would want to create a separate R directory daily or weekly so
there has to be some sensible cutoff somewhere. In particular, its
important to distinguish 1st and 2nd level versions with 3rd level
versions.  I would agree with you for 1st and 2nd level versions but
not for 3rd level versions which is what is being discussed here.  3rd
level versions

- don't have critical differences and
- you virtually never have to go back since third level versions are
basically bug fixes

so I don't think such arguments apply.  The fact that one bug fix can
generate another bug is more theoretical than a real problem that is
encountered.  If you did want to go back there is nothing to prevent
you from downloading the old version again (although in years of
following this scheme I have never had to do it) or using the last 2nd
level version.   Also if you don't like the whole scheme its only a
default, not an operating requirement, so for those few people who
want a separate installation for every third level version they could
still have it just as anyone can implement the proposed scheme even in
the absence making it the default; however, for most people I think
that generating a large number of alpha, beta, pat, *.0, *.1
directories is pointless since they will never use anything other than
the most recent in any x.y.* series.



More information about the R-devel mailing list