[R] R 2.0.0 not suffisantly reliable to be be used

Prof Brian Ripley ripley at stats.ox.ac.uk
Mon Oct 11 08:48:13 CEST 2004


[I wrote this yesterday, but decided not to send it.  In the light of 
subsequent correspondence I do now.]

On Sun, 10 Oct 2004, Fan wrote:

> After wasting one whole day, I've finally decided to stay with 1.9.1,
> some problems have been reported to R-Bugs.

Four reports have been received from Fan, one of which was a duplicate of
another.  Two had already been replied to, and another reply has crossed
this message.  The details can all be seen on R-bugs or in the R-devel 
archive.

PR#7274 is a question answered in the first section of the NEWS file,
`USER-VISIBLE CHANGES', and also in the rw-FAQ.  PR#7275 is incorrect code
that also fails in 1.9.1 (and even 1.8.0) despite Fan's claim to the
contrary.  PR#7272/3 contains one small point (not involving code) that as
far as we can tell is already in the archive (and if so was described
incorrectly) and is already fixed in R-patched, and another that has no
code to reproduce it but appears to be a previously untested problem in
his datasets (as also mentioned in the top section of the NEWS file).

> For occasional users, I would say, there's no worst thing than that:
> you installed the new release, and soem of your existing codes no
> longer work !

This is 2.0.0, a major release, so one might expect some incompatibility
(and there are a few points documented early in the NEWS file).  In
particular we do document that packages must be reinstalled, and that we
do check more stringently that packages work at both install and load
times.  But large amounts of code and several hundred packages have been
tested under 2.0.0, and some of the performance benefits come from not 
checking if packages are correctly installed at run time.

There have so far been very few verifiable reports of bugs introduced in
2.0.0 (indeed fewer than I can recall in the first week for any 1.x.0
release) and all of those have already been fixed in R-patched.  (The only
ones I could find are R CMD INSTALL with versioned installs on packages
which save images, setting "bg" on Windows devices and failing to find
aliases in Windows CHM help.)  Compare that with the list of bugs that
have been fixed since 1.9.1 (and even of earlier bugs fixed in R-patched).
It would be even better if more users (especially Windows users) would 
help with alpha/beta testing new releases -- bugs tend to more prevalent 
in things which can only be tested interactively.

(Unfortunately due to repeated SVN problems, R-patched is not as easy to
get hold of as it is intended to be.  We do have up
ftp://ftp.stat.math.ethz.ch/Software/R/R-patched.tar.bz2, a source tarball
from early this morning, and
http://cran.r-project.org/bin/windows/base/rpatched.html has a Windows
build from Saturday.)

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595




More information about the R-help mailing list