Re: Solaris: patches to run-octave.in

[Moritz Borgmann]

At 0:22 Uhr +0100 2007-11-30, David Bateman wrote:
> John W. Eaton wrote:
> >  On 29-Nov-2007, Moritz Borgmann wrote:
> >
> >  | hm. We could make it conditional and not set LD_PRELOAD on Solaris.
> >  | Other than that, I fail to see immediate solutions. Or we could try
> >  | to track down what the reason was that the other systems (which
> >  | ones?) didn't honor the LD_LIBRARY_PATH. Maybe it's called
> >  | differently on that system (like, e.g., DYLD_LIBRARY_PATH under
> >  | Darwin, which configure sets correctly).
> >
> >  The thread where this was originally discussed is here:
> >
> >
> > http://www.nabble.com/segfault-in-build-of-octave-CVS-tf2587656.html#a7226545
> >
> >  I dont' think it mentions exactly what system had the problem, but
> >  maybe David remembers.
> >
> >  jwe
>
> Yes the issue occurred on my Mandriva 2006 linux machine. I have no idea
> why it wasn't respecting the LD_LIBRARY_PATH variable, and was loading
> the installed versions. I didn't use rpath in the build, so that's not it.

I think it's a subtle problem with the loader, see, e.g.

http://www.perlmonks.org/index.pl?node_id=126587
http://www.nabble.com/Re%3A-dynaloader-issue-p5667945.html

(unfortunately, man ld.so doesn't say this, at least on my Debian Linux).

Do we have to "exec" octave? Why can't we just spawn it as another 
process? That should solve the problem and make LD_PRELOAD 
superfluous...

Note that the evilness of LD_PRELOAD on Solaris has already bitten others:

http://groups.google.com/group/opera.os.solaris/browse_frm/thread/f350f95505a83477/f7c67646330d6a96?#f7c67646330d6a96

-M