Re: [Gábor Gombás] Bug#564566: libkadm5clnt7: SONAME conflict with Heimdal

[Jeffrey Hutzelman]

--On Tuesday, January 12, 2010 05:21:10 PM -0500 Sam Hartman 
<hartmans at mit.edu> wrote:

>>>>>> "Jeffrey" == Jeffrey Hutzelman <jhutz at cmu.edu> writes:
>
>     Jeffrey> --On Monday, January 11, 2010 03:14:00 PM -0800 Russ
>     Jeffrey> Allbery
>     Jeffrey> <rra at stanford.edu> wrote:
>
>> Tim Mooney <mooney at dogbert.cc.ndsu.nodak.edu> writes:
>     >>
>     >>> How much pain would it cause developers if the SONAME was
>     >>> libkadm5clnt-mit.so.<N> instead of libkadm5clnt.so.<N>_mit?
>     >>
>     >> The problem here is that it changes the name of the library that
>     >> one specifies on the link line of programs, so anything that
>     >> linked against the library would need to potentially change its
>     >> build machinery, have code for detecting the new and old version,
>     >> and so forth.  Now, since the MIT libkadm5clnt API had not
>     >> previously been public, there probably isn't a lot of such
>     >> software (I'm only aware of one package off-hand), but it
>     >> definitely has more potential impact.
>
>     Jeffrey> Or, it means you have a library with filename
>     Jeffrey> libkadm5clnt.so and soname libkadm5clnt-mit.so.<N>, which
>     Jeffrey> would also be poor.
>
>
> Jeff, why would that be bad?  Several of us discussed and couldn't come
> up with significant problems with that approach.

I suppose it would be OK to have libkadm5clnt-mit.so, with libkadm5clnt.so 
being a symlink.  In fact, one could even decide that the latter is 
controlled by alternatives, on a Debian-like platform.  It just seems like 
it would be wrong to not have the name corresponding to the unversioned 
soname.