[PROPOSAL] Re-bundle (stop producing tarballs for) ldb?

[Andreas Schneider]

On Thursday, April 11, 2019 5:01:29 AM CEST Jeremy Allison wrote:
> Red Hat is the most popular Linux distro with a large
> user base and we need to take their engineering
> needs into account.

I think if we release a library as a separate tarball and make an API promise, 
we should not stop in doing so. If we can't make such a promise we should not 
have released libldb in the first place.

Yes, it happens that an API design is not good. That happened several times 
for me with libssh, so I introduced a complete new API keeping the old API and 
semantics functioning and I have probably to keep them around forever. For the 
latest version I rewrote known_hosts handling from scratch as the initial 
implementation was not flexible enough. If an API should not be used anymore 
you can mark them as deprecated in a later release.

libldb is not the only library out there having these issues. Also what would 
you say if lmdb would change it's internals so that the API is the same but it 
would function completely differently? We expect stable code from others and 
if we release a library with a promise we should keep that.

There are users of libldb beside of Samba and they choose libldb because they 
thought we provide a great project which is maintained, taken care of and 
offers a stable API.

My fear is that if we start changing libldb internals and break something 
which another party depends on, they will just fork it. And this is something 
I really want to avoid.

What makes it impossible to provide a stable API for libldb and keep libldb as 
it is?




Best regards,


	Andreas


-- 
Andreas Schneider                      asn at samba.org
Samba Team                             www.samba.org
GPG-ID:     8DFF53E18F2ABC8D8F3C92237EE0FC4DCC014E3D