An additional complication with the kadm5 sonames

[Nicolas Williams]

On Fri, Jan 15, 2010 at 01:13:25PM -0800, Russ Allbery wrote:
> Love Hörnquist Åstrand <lha at kth.se> writes:
> 
> >> Love, are you watching this thread?  Is causing a large bump to the MIT
> >> libkadm5 SONAME versions likely to prevent this sort of version
> >> conflict with Heimdal in the future?
> 
> > heimdal uses libtool so I can't predict any resulting version
> > number. The only way to solve this is to not use same base name.
> 
> Oh, good point, I forgot about that.  libtool can even use different
> SONAME versions on different platforms, since the requirements of
> different platforms around SONAME changes vary.

This is one shocking failure of libtool.  SONAMEs and version numbers
are important aspects of ABI definitions in dynamic run-time
environments.  Therefore developer control of them is as important as
developer control of the symbol names they use.  Nor can libtool
possibly detect when it's appropriate to change version numbers, since
the matter of whether and when you're making a release is relevant to
that, as is the past (information about past releases).

Nico
--