Re: [PATCH 0/7] Move nfs.conf to a shared library

[bfields@fieldses.org (J. Bruce Fields)]

On Fri, Sep 01, 2017 at 02:28:29PM +0100, Justin Mitchell wrote:
> On Wed, 2017-08-30 at 16:33 -0400, Steve Dickson wrote:
> > 
> > On 08/30/2017 10:20 AM, Chuck Lever wrote:
> > > 
> > > > On Aug 30, 2017, at 8:15 AM, Christoph Hellwig <hch@infradead.org
> > > > > wrote:
> > > > 
> > > > On Wed, Aug 30, 2017 at 12:51:06PM +0100, Justin Mitchell wrote:
> > > > > Moving the config parsing code out to a shared library fixes
> > > > > both of these issues with minimal code changes. I appreciate
> > > > > that the API this library presents is less than ideal, it was
> > > > > left this way to reduce code changes at this point, and i
> > > > > welcome suggestions for a sympathetic approach to improving it.
> > > > 
> > > > The general shard code approach look good, but is there any good
> > > > reason not to merge libnfsidmap into nfs-utils and make it a
> > > > static (or at least a private shared) library instead of
> > > > introducing
> > > > a new ABI?
> > > 
> > > Agreed: who else is using libnfsidmap? Is there currently a
> > > packaging requirement that keeps it outside of nfs-utils?
> > 
> > I hope not... and I agree... lets use this effort to roll
> > libnfsidmap into nfs-utils as static library.
> > 
> > The pain is more of a logistics issue for distro... IMHO...
> 
> afaics the three things that depend on libnfsidmap are nfs-ganesha,
> freeipa-client, and sssd-nfs-idmap. the first two pull in nfs-utils
> anyway, the latter would gain a few extra dependencies from pulling in
> nfs-utils instead of just libnfsidmap.
> 
> Although that is somewhat academic as a merged tree could still be
> packaged out as separate items, just as it is now, but internally can
> share the code.
> 
> So I will look at doing that instead.

Originally we tried to make it possible to manipulate NFSv4 ACLs using
setfacl/getfacl, by patching in NFSv4<->POSIX ACL mapping into libacl.
That added a libacl->libnfsidmap dependency.  I think that was the only
original motivation.  And, yeah, that's not necessarily an argument for
building it from a separate source tree.  I think it was just a mistake.
I'm fine with merging it into nfs-utils.

--b.