R devel - Dec 2005 - ???UNSURE??? Re: (PR#8363) R CMD INSTALL fails if cd prints

On Mon, 5 Dec 2005, Philip Lijnzaad wrote:
> On Friday 02 December 2005 18:20, Prof Brian Ripley wrote:
>
>> What shells are these?
>
> Bash, mostly, but also ksh and zsh; sorry for not mentioning this.
I still don't know what you did to be able to reproduce this (and I did 
ask).  And as it is a shell script running under /bin/sh, it must be 
whatever is masquerading as Bourne shell on your system(s) that is 
affected.
> I now see that the root account usually does not change the behaviour of 
> cd, so we may as well forget about the matter. My thought was: if a 
> small change helps avoid this problem (which I think can occur easily 
> enough), it could be helpful. Incidentally, IMHO the way it is coded in 
> INSTALL.in (I didn't know about INSTALL.in, sorry) is a bit unusual; cd
> does not seem to be required by the POSIX standard to have non-zero exit 
> status when it fails (and in fact, on older SunOSes, 'cd 
> non/existant/directory' returns 0). But that is
But it _is_ so required:

http://www.opengroup.org/onlinepubs/009695399/utilities/cd.html

How old are `older SunOSes'?  Solaris 8 (which is 4 years old) is 
POSIX-compilant here, and I would be surprised if any from the last decade 
was not (at least with suitable paths set).
> probably more a matter of style, more than anything else.
Our philosphy is to assume that the users tools are standard (e.g. POSIX) 
unless there is widespread evidence to the contrary.  First off, we need 
to know what was done and to be able to reproduce it.

-- 
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