libmagick, libmagick++ soname confusion

[Henning Makholm <henning@makholm.net>]

tag 224287 pending
thanks

Here is my analysis of bug #224287:

autotrace 0.31.1-2 was built against libmagick++5.5.7 version
which 4:5.5.7.9-1 provided libMagick++.so.0.

Two days ago, libmagick++5.5.7 version 4:5.5.7.15-1 entered the
archive. It provides libMagick++.so.6 instead.  This caused the
autotrace binary from 0.31.1-2 to refuse to start.

The bug against autotrace can be fixed by recompiling (just as the
reporter suggested, in fact). I will bump the Debian revision to
provoke recompiles shortly.

For debian-devel: I supposed it was a bug in imagemagick 4.5.5.7.9-1
that it provided libraries with the wrong sonames [1]. However, now
that the sonames are correct, shouldn't
/var/lib/dpkg/info/libmagick++5.5.7.shlibs declare a *versioned*
dependency, such that people will not try to use a freshly built (say)
autotrace with 4:5.5.7.15-1, which does not provide .so.6?

I'm new to this shlibs business, so I'm not filing a bug against
libmagick++5.5.7 right away. Should I? And if I should, should I wait
until a fixed libmagick package is in the archive before I reupload
autotrace? Or should I hack autotrace's control file such that it
conflicts agains old versiond of libmagick?

[1] but the Debian changelog does not mention the change, so it may be
    that there's still a bug in its build scripts that causes the
    sonames to be wrong *sometimes*?

-- 
Henning Makholm                     "The practical reason for continuing our
                                  system is the same as the practical reason
                          for continuing anything: It works satisfactorily."