[Rd] core dump on 64-bit Solaris (PR#990)

Prof Brian D Ripley ripley@stats.ox.ac.uk
Wed, 20 Jun 2001 22:17:18 +0100 (BST)

On Wed, 20 Jun 2001, Kurt Hornik wrote:

> >>>>> tlumley  writes:
> > The crash is caused by a bus error: address alignment. Recompiling without
> > -lsunperf and with -dalign fixes it.
> > I think there is a configuration issue here (not for 1.3.0,
> > though). The documentation for the Sun BLAS says that -dalign is
> > needed, but the default configuration is to use the Sun BLAS and not
> > -dalign.

I will add that to R-admin.texi, though.

> Ok.  Note that this part of the configure code is taken from Octave,
> which used to be reliable.  It would be great if we could pass info on
> problems and solutions on to John Eaton as well (in particular as I
> think that at some point of time there was talk about adding the code to
> the autoconf contrib archive).

I am not 100% sure this can be automated.  For a start, there are lots of
versions of the Sun compilers, and Thomas' is not current.  Really the
issue is that the compiler flags had better match, and that is very hard to
ensure except manually.  My f77 man page does say

          If you compile one subprogram or file with -dalign,
          then all subprograms and files in the program unit must
          be compiled with -dalign.

and my system is current (re-installed a month ago).

This issue comes up with mixed compilers too: AFAIR gcc does the equivalent
of -dalign and so could not be used with S-PLUS 3.2 (or about then, built
with SunPro cc).

> We had a related problem last week, resulting in dropping CXML/DXML
> support due to problems on 64-bit platforms with gcc (if I recall
> correctly).

Just gcc on Alpha running whatever Compaq calls Dec's UNIX today. CXML/DXML
are compiled with a different compiler.

So the general issue is the you need to ensure that you use the same
compiler and the same flags to compile R as you used for the libraries you
use, or know the perils if you do not.  I've just added a warning to
R-admin.texi about that.

Brian D. Ripley,                  ripley@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 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch