[uClibc] Strange dynamic symbol in PPC

[Erik Andersen]

On Fri Jan 30, 2004 at 06:11:54PM +0100, Joakim Tjernlund wrote:
> > On Mon Jan 26, 2004 at 08:12:47PM +0100, Joakim Tjernlund wrote:
> > > OK, had a look at dlopen.c, but I didn't get any wiser. Is 
> > this bug PPC specific?
> > 
> > Nope.  Is is a general, arch-independent bug.  Actually, it is
> > more of a design problem which causes us not correctly handle
> > weak symbols in dlopen()d shared libraries.  It is this very
> > problem that, i.e. is preventing mozilla from running as has
> > been reported a few times.  It also causes problems with perl,
> > where weak symbols from libpthread are not correctly handled,
> > 
> >  -Erik
> 
> Managed to cross compile the stuff uClibc/test/dlopen for PPC.
> 
> dltest reports "Weak values resolved incorrectly". I guess that
> this test shows the problem described above?

Yep, that is indeed the problem.

> Also, I noticed that test1 and test3 failed to open "libtest1.so" until
> I copied libtest*.so to /lib. Should they behave like that?

In the test Makefile, I think I was using LD_LIBRARY_PATH
to avoid that.  I wonder of glibc includes the current
directory in the lib patch when doing a dlopen()?

 -Erik

--
Erik B. Andersen             http://codepoet-consulting.com/
--This message was written using 73% post-consumer electrons--