Re: libdb4.* madness in unstable
On Tue, 16 Oct 2007, Russ Allbery wrote:
> Michael Biebl <firstname.lastname@example.org> writes:
> > Today, while browsing through aptitude, I noticed that I had the
> > following bdb versions installed:
> > version: # of packages depending on it (apt-cache rdepends)
> > libdb4.2 40
> > libdb4.3 26
> > libdb4.4 55
> > libdb4.5 64
> > libdb4.6 40
> > Having 5 different versions of one library is just insane imho. What are
> > the reasons, that we still carry around the older versions, like 4.2 and
> > 4.3? Is there software which doesn't build against newer versions, are
> > there other reasons?
> BerkeleyDB 4.2 is still faster and more stable than any subsequent version
> for OpenLDAP, although 4.6 is looking promising.
> I don't have any good explanation for 4.3 through 4.5, though.
I do. We don't ask the cyrus people and openldap people before packaging a
new libdb version.
The ugly truth is that libdb compatibility across minor versions is crap.
You always have to fine-comb the API for changes, because the changes are
not compile-time-detection safe. Then you hit the bugs.
IMHO we should declare a quarantine of a minimum of 6 months on every new
libdb upstream, and only package it *if* openldap (*the* heavy-duty user of
libdb advanced features) and cyrus imap (*the* thousands-of-concurrent-
locks, mmap-happy user of libdb) did not have problems with it.
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot