CLANG 3.2 breaks security/pam_ssh_agent_auth on stable/9

[Brooks Davis]

On Thu, Jan 31, 2013 at 10:22:44PM -0600, Mark Linimon wrote:
> On Thu, Jan 17, 2013 at 09:15:02AM -0600, Brooks Davis wrote:
> > Not unless you consider adding new functions in a reserved namespace
> > (str*) to be ABI breakage.
> 
> Well, what often happens is that when we add new functions, ports break.
> I think deciding whether this is or is not "ABI breakage" is semantics.
> The fact is that regressions get introduced with these types of changes.
> 
> > The port should have continued to work unless it was recompiled so it
> > should have preferred it's own version of the strnvis symbol.  If its
> > makefiles were properly constructed it would have failed to compile
> > due to the signature mismatch.
> 
> The mantra should be "every possible combination of ways that a port's
> internal build glue can be wrong, is already included in the Ports Collection."
> In case after case we see fragile code that is written by people who are
> clearly not professionally trained.  They "get it to work on their system"
> and then shove it out the door.
> 
> Claiming that "they shouldn't do that" is correct but self-defeating.
> It's just the reality of open-source software.

I'm not sure why I'm being jumped on me in this weeks old report of a
now-fixed problem.  I did determine to root cause and others produced a
patch.  If no one else had stepped up I would have done so my self.

> IMHO, the burden should be on whoever makes the change to find out whether
> or not regressions will be introduced.  (And yes, I am very aware that we
> don't have -exp run capability right now, but this is one of the cases
> where I would like to suggest it would have helped.)

I would likely have done an exp run had there been the capability of
doing one, but this bug would not have been found since it's a runtime
crash caused by a combination of two different BSD projects not talking
to each other and poorly chosen CFLAGS in the upstream software allowing
it to compile.

One could probably write a tool to detect some forms this sort of issue
(even premptively), but it's probably not worth doing.

-- Brooks
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 188 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20130201/8a91e64a/attachment.sig>