[PATCH] wbclient password change calls

[Steve Langasek]

On Fri, Aug 29, 2008 at 11:12:48AM -0500, Gerald (Jerry) Carter wrote:
> > If this one just adds a function it is 
> > technically not necessary. Or is this change
> > actually changing a function signature ?

> > sonames bumps are really necessary only if 
> > the ABI changes.

> Thought we already talked about this.  If it is still
> libwbclient.so.0 and someone write a new app that
> calls wbcChangeUserpassword(), it will break on a v3.2
> installation even though it will find libwbcliebt.so.0.

I wonder why this is considered important.  If you change the name to
libwbclient.so.1 and someone writes a new app that calls
wbcChangeUserpassword(), it will /still/ break on a v3.2 installation, but
with a different error.  Why does libwbclient need a different solution than
the standard one that's been used on, e.g., Linux for the past decade?
Using symbol versioning gives you a nicer error message when a symbol
version isn't found, but plenty of libraries do this without symbol
versioning at all even on platforms that support it - if a newer symbol is
added, they adjust the library filename (e.g., libwbclient.so.0.0 ->
libwbclient.so.0.1) to signal the change while keeping the soname the same,
and trust that anyone copying around binaries keeps track of the needed
dependencies in some way (either via a package manager, or by noting the
filename differences).

The /only/ benefit you get with symbol versioning or an SONAME bump over
this method, in the case of a backwards-compatible ABI change, is a somewhat
clearer error message.  In all else, the management requirements are the
same, and administrators have been managing them for years...

-- 
Steve Langasek                   Give me a lever long enough and a Free OS
Debian Developer                   to set it on, and I can move the world.
Ubuntu Developer                                    http://www.debian.org/
slangasek at ubuntu.com                                     vorlon at debian.org