Re: [patch] testsuite fix for solaris 64bit testing

[Andreas Tobler <toa at pop dot agri dot ch>]

Tom Tromey wrote:
> > > > > > "Andreas" == Andreas Tobler <toa@pop.agri.ch> writes:
>
>
> Andreas> during some xmas testing I encountered tons of failures when testing
> Andreas> solaris -m64 with a 32 bit native compiler.
>
> Andreas> The reason was a simple library mismatch. Setting the
> Andreas> LD_LIBRARY_PATH_64 for solaris -m64 reduced the failure count to a
> Andreas> reasonable minimum with about 28000 passes and about 200
> Andreas> failures. 32/64bit libjava testing with mauve.

I think I found the reason why _I_ need this PATH set and others don't 
seem to need it.

Solaris >= 7 knows the LD_LIBRARY_PATH and the LD_LIBRARY_PATH_64. If 
the latter is not specified, the former tells the linker to look for 
_BOTH_ 32 and 64 bit libraries in this PATH. In case of the 
LD_LIBRARY_PATH_64 is used somewhere, the LD_LIBRARY_PATH gets overriden 
for 64 bit libraries.

Detailed explanation somehwere on the sun doc site.
http://docs.sun.com/source/817-0929/4_libraries.html

I have set this _64 PATH somehwere in my shell for a debugger. So my 
usual LD_LIBRARY_PATH was now only valid for 32 bit libraries.

At the moment I'm not sure what to do. It also affects other gcc libraries.

> Could you look at proc ${tool}_set_ld_library_path, near the end of
> testsuite/lib/libjava.exp?

I'll do so.

> I think the libjava_init part of your patch is fine.  For the other
> parts, I think I'd prefer we just consolidate all the environment
> variable settings into the aforementioned proc.  Could you try that
> change?  If it works, let's also move the SHLIB_PATH and
> DYLD_LIBRARY_PATH settings there.

If it works I come up with a patch.


Thank you for review.

Andreas