[PATCH RFC] cifs-utils: new plugin architecture for ID mapping code

[simo]

On Thu, 2012-12-06 at 10:16 -0500, Scott Lovenberg wrote:
> On Thu, Dec 6, 2012 at 7:37 AM, Jeff Layton <jlayton at samba.org> wrote:
> > 1/ Should we switch this code to use a config file of some sort instead
> > of this symlink? The symlink would probably be more efficient, but it
> > is a little odd and might confuse people. It also might make it hard to
> > expand the idmapping interfaces later.
> 
> Please no symlinks.  We could end up with something like alternatives
> (/etc/alternatives) where you need a utility to track and change
> symlinks pointing to symlinks.  I don't even know where java is on my
> machine or which JVM I'm running.  Symlinks have a tendency to turn
> into symlink farms.  Look at your PAM install, it's a bunch of links
> to a single file on most installs.  That way lies madness and dragons.
> 
> I think most people are comfortable with config files and they're
> intuitive.  Also, is anyone at any time going to be using more than
> one mapping interface?

Probably not, which is why a symlink is all we need.
I think starting with a symlink is fine, I do not particularly love it,
but makes a lot of stuff simpler for starter.
Having to parse a config file at each program invocation (for utilities)
adds up in latency and also requires YACF (Yet Another Config File)
where all you do is say: use plugin X, or a very complex version of a
symlink.

Now if you need to actually configure something then yes we could move
to use a config file eventually.

But unless you already know there is something the cifs utils (*not* the
plugins) need to configure then it seem a lot of overhead both in terms
of coding that needs to be done, file formats to choose and associated
bikeshed, and runtime churn.

Simo.


-- 
Simo Sorce
Samba Team GPL Compliance Officer <simo at samba.org>
Principal Software Engineer at Red Hat, Inc. <simo at redhat.com>