Symbol visibility revised

[Martin v. Loewis]

> Martin, your patch is ok by itself. By that, I mean it doesn't
> introduce any new bugs. However, it is incomplete. I'd like to
> check in your patch and at the same time to check in a testsuite,
> ld-elfvsb, for the visibility fearture.

Thanks for reviewing my patch. If you mean by "incomplete" that
protected is not fully supported: Yes, that is indeed the case.  As
we've discussed before, protected visibility support probably needs to
occur in the dynamic linker; if somebody finds a scheme to support it
in the static linker also, that can be added later.

I'd really like to see that patch (together with your test cases) in
binutils, so that people can try out the new C++ ABI __dso_handle
stuff of GCC - especially since glibc 2.1.3 already has the required
functions _cxa_atexit and _cxa_finalize.

I'm very glad that this part of the new ABI eventually works out,
since it allows to fix a very old bug in g++ (dlclosing of shared
libraries which had block-static objects registered with atexit).

Regards,
Martin