Skip to content

bug in chdir option of source

4 messages · usenet@s-boehringer.de, Brian Ripley, Duncan Murdoch +1 more

#
I apologize for not having investigated enough.

However I want to bring up the point that upgrading can be a very
tedious thing. I administer a cluster for which upgrading brings it to a
halt for several hours stopping all calculations going on. At the moment
I have no way to go from 2.1.0 to 2.1.1.

I just want to add this point to the release policy of R which
historically has very quick release cycles which on the one hand is a
very positive thing but on the other hand can be problematic with
respect to some aspects.

Thanks again, Stefan
On Tue, 2005-07-12 at 13:02, Prof Brian Ripley wrote:
#
On Tue, 12 Jul 2005 usenet at s-boehringer.de wrote:

            
I really don't understand.  Just install the current R in a different 
place and try it.  If it works better, use it for newly-started jobs.
We too run a large cluster (and a few hundred loosely connected machines) 
and are able to upgrade pretty transparently even when long jobs are 
running.

For those who are reluctant to upgrade, I would suggest never installing a 
.0 release (of any software, not just R).

  
    
#
usenet at s-boehringer.de wrote:
One simple solution is to test on a different system (e.g. a standalone 
one).  If the bug is present in 2.1.0 there but absent on 2.1.1, it's 
probably fixed on your cluster, too.

Another option which may or may not work is just to read the NEWS file 
for the newer releases.  It's not always obvious that a bug fix there 
fixes your own problem, but sometimes it is.
It's hard to decide which versions to install.  On the one hand, if 
installs are difficult, you want to do it infrequently -- so maybe skip 
all the x.y.0 releases.  On the other hand, if you have unusual 
hardware, you are likely to see bugs that others don't see -- so 
installing the betas and the .0 releases will help to get your bugs fixed.

This is why it's useful to have two systems, one for "production", one 
for testing.  Then you can follow both policies.

Duncan Murdoch
#
usenet at s-boehringer.de writes:
Hmm, I'll believe you, but the blame for that can be shifted in
various directions.
Actually, we have quite slow release cycles compared to other
OpenSource projects. It's basically twice a year plus minor patches.
According to freshmeat.net, a project's "vitality" essentially drops
to zero a month after the latest release.

Now if only people would actually try the betas rather than reporting
errors after the release, we'd have a much better chance of avoiding
problems like yours...