The Wrapper Project

[Andreas Schneider]

On Monday 02 December 2013 14:33:55 Jeremy Allison wrote:
> On Mon, Dec 02, 2013 at 10:27:32PM +0000, Jelmer Vernooij wrote:
> > On Mon, Dec 02, 2013 at 02:23:43PM -0800, Jeremy Allison wrote:
> > > On Mon, Dec 02, 2013 at 09:43:22PM +0000, Jelmer Vernooij wrote:
> > > > On Mon, Dec 02, 2013 at 01:21:04PM -0800, Jeremy Allison wrote:
> > > > > On Fri, Nov 29, 2013 at 12:15:08PM -0500, Simo wrote:
> > > > > > The point is that we are not good at maintaining external code
> > > > > > once you
> > > > > > suck it in the tree.
> > > > > 
> > > > > IMHO The point Simo is making above is the most important
> > > > > one in this discussion.
> > > > > 
> > > > > We have pulled in external code, Heimdal, popt, zlib
> > > > > and WE DON'T KEEP IT UP TO DATE.
> > > > 
> > > > That is a different class of libraries. These three libraries were
> > > > internal to Samba; they're more akin to e.g. tdb, talloc. tevent or
> > > > ldb.
> > > 
> > > Heimdal I'm willing to concede as we needed to make
> > > changes to make the AD-DC. But are we done there ?
> > > Shouldn't we be pushing upstream ?
> > 
> > The three libraries I'm talking about are the ones that Andreas has
> > split out of the Samba tree - socket_wrapper, uid_wrapper and
> > nss_wrapper. They're the same category as tdb, talloc, tevent and ldb
> > as they were originally internal to Samba.
> 
> Doh, sorry - EREADERTOOSTUPID :-).
> 
> But I'd still argue that if they're going to be
> made generic and used by other projects that
> the ultimate goal should be to spin them off
> into a separate tree. That way the interfaces
> get more review (due to the needs of other
> projects) and tend to improve and become
> more generic over time.
> 
> I think it does lead ultimately to better
> code - it forces us to think more about
> how to interface with things rather than
> just being able to make arbitrary changes.
> 
> The only question I have is are these
> libraries ready to make that step ? We
> tried with ctdb and this eventually had
> to come back. I'd argue we've been better
> with talloc, tdb and tevent.

Yes, if the there are enough unit tests. nss_wrapper is at 75%, uid_wrapper is 
at 69%, the rest is easy.

The big step is to get socket_wrapper form 1% to at least 75%.

Testings ensures that they are working as expected and we don't regress in 
future. The more people are using it, the better testing and testcases we get.

And that's the goal. At least for me!