From owner-cvs-ports  Mon Dec  9 02:29:55 1996
Return-Path: <owner-cvs-ports>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.4/8.8.4) id CAA26892
          for cvs-ports-outgoing; Mon, 9 Dec 1996 02:29:55 -0800 (PST)
Received: from dfw-ix3.ix.netcom.com (dfw-ix3.ix.netcom.com [206.214.98.3])
          by freefall.freebsd.org (8.8.4/8.8.4) with SMTP id CAA26887;
          Mon, 9 Dec 1996 02:29:53 -0800 (PST)
Received: from silvia.HIP.Berkeley.EDU (ala-ca19-15.ix.netcom.com [205.187.212.47]) by dfw-ix3.ix.netcom.com (8.6.13/8.6.12) with ESMTP id CAA10410; Mon, 9 Dec 1996 02:29:11 -0800
Received: (from asami@localhost) by silvia.HIP.Berkeley.EDU (8.8.4/8.6.9) id CAA03507; Mon, 9 Dec 1996 02:29:09 -0800 (PST)
Date: Mon, 9 Dec 1996 02:29:09 -0800 (PST)
Message-Id: <199612091029.CAA03507@silvia.HIP.Berkeley.EDU>
To: thomas@ghpc8.ihf.rwth-aachen.de
CC: thomas@ghpc8.ihf.rwth-aachen.de, CVS-committers@freefall.freebsd.org,
        cvs-all@freefall.freebsd.org, cvs-ports@freefall.freebsd.org
In-reply-to: <199612091006.LAA06465@ghpc6.ihf.rwth-aachen.de> (message from Thomas Gellekum on Mon, 9 Dec 1996 11:06:13 +0100 (MET))
Subject: Re: cvs commit:  ports/graphics/ImageMagick/pkg PLIST
From: asami@freebsd.org (Satoshi Asami)
Sender: owner-cvs-ports@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

 * > Ok, so they are incompatible.  Is that one direction only?  (I.e., do
 * > old binaries work if they are fed libMagick.so.3.7.8?)
 * 
 * Haven't tried that.

That would involve a major number change so it's actually quite
important....

 * > Also, you didn't say how it "doesn't work".  Does ld.so correctly
 * > print out a warning ("shlib minor >= 8 expected, only found 7, using
 * > it anyway" or some such) or does it just bomb on you with a core dump
 * > or mising function?
 * 
 * Missing function.

Without any warning from ld.so at all?  That's bad, since users might
not even know the cause. :(

Of course, we don't have a port that depends on shared libMagick, so
it probably isn't a problem for most people.  But that doesn't mean we
can ship releases with packages containing inconsistent shared
libraries, if a user copies a binary from FreeBSD system A to system
B, and they are both running RELEASE with ports/packages that came
with them, ld.so is supposed to handle any situation gracefully.

 * > If it is the latter, that means the third digit is ignored by
 * > ldconfig/ld.so.
 * 
 * Ah. I didn't realize that. Guess I'll have to look at the port again.

Ok...I guess that means if it's backward compatible, it should be
bumped to 3.8, if it's incompatible both ways, 4.0.  (And let's not
have any more x.y.z shlibs, it's confusing as it gives us the illusion 
that all digits are used by the linker.)

Satoshi