Re: libpam and symbol versioning

On Tue, Aug 23, Wichmann, Mats D wrote:

> 
> >the next release of Linux-PAM will have symbol versioning enabled.
> >While this is not a problem at runtime for old binaries, lsblibchk
> >does complain that it cannot find the functions without symbol
> >versioning.
> >
> >What can we do here?
> 
> That's going to take some redesign, and I don't really see
> how we can figure this out.  lsblibchk thinks a symbol is
> either versioned, or unversioned, and doesn't know how to
> mix them.  It wouldn't be too hard to say "found versioned
> symbol when expecting unversioned - okay", but that seems to 
> open the door to not catching real changes. For example,
> this would make both foo@VER_1.0 and foo@ver_1.1 acceptable,
> which is pretty clearly the wrong thing.

Thinking twice about this: Why would it be the wrong thing to have
a symbol with different version numbers? I think this could be the
right thing, only think about glibc 2.0 (unversioned) and 2.1 (versioned).

LSB conform applications using PAM would not have symbol versions for
the functions they call, so it is in the response of the dynamic
loader to choose the right one. Whatever it is.
Since the same dynamic loader is used for the runtime tests, we know
that the choosen function will pass the LSB tests.

  Thorsten

-- 
Thorsten Kukuk         http://www.suse.de/~kukuk/      [email protected]
SUSE LINUX Products GmbH       Maxfeldstr. 5       D-90409 Nuernberg
--------------------------------------------------------------------    
Key fingerprint = A368 676B 5E1B 3E46 CFCE  2D97 F8FD 4E23 56C6 FB4B