Patch to remove zlib.

[Ira Cooper]

On Fri, Jul 11, 2014 at 4:54 AM, Volker Lendecke <Volker.Lendecke at sernet.de>
wrote:

> On Mon, Jul 07, 2014 at 12:46:47PM -0400, Ira Cooper wrote:
> > Note: To apply it  unxz it then use git am --ignore-whitespace ,
> otherwise
> > you may have issues.  (xz was used to save the list a good bit of
> space...
> > it's over 500k gzipped.)
> >
> > As far as why: I listened to metze/vl.  I disagree with them.
>
> If we decide to remove zlib and make it a fixed external dependency we
> should go the full path and apply the same to at least popt and iniparser
> immediately. Next step then would be the wrappers, which are externally
> maintained as well. Then after that comes our own 3 libs tdb, talloc,
> tevent and possibly ldb. There's one big one that comes after that then,
> but which should go as well: Heimdal. It's a slight change in policy,
> but a valid one I guess.


Actually, I have some of the code needed for the zlib removal done.  I
think I'm missing a bit to make it conditional, for AD and the fileserver,
but it I'm not far off.

Then I have to do the waf parts etc, but still.. I'm hoping I can get it
done today.

As far as the rest of your points, I'm going to break the libraries into
"groups" to be considered.  I personally have a line where I'll stop,
because I think we have experience that shows where that line should be.

1. zlib, popt, iniparser, Heimdal - These libraries should be removed as
soon as technically feasible.  The wording chosen IS intentional here.
 Heimdal isn't possible today due to the AD DC, but when it IS possible to
build a working AD DC without Heimdal, it no longer belongs in tree.
2. socket_wrapper - Given that we just put it in, we should stop briefly
and think here.  I'm far from against it.  But to me, this is "more
optional", than the first pass.
3. ntdb+ccan - I believe these should go.  But they are a separate issue.
4. talloc - Talloc is stable, we aren't making changes to it.  If there is
a library we maintain, that we can break out.  I personally think talloc is
it.
5. ctdb, tdb, tevent* - Given the active development by the Samba Team on
these libraries, and that we are the main consumer, I think breaking these
out is not a good idea.
6. Everything else.  Likely it is not ready to be broken out.  There may be
technologies we wish to consider breaking out, but I don't consider that
part of this section of the project.

People have convinced me personally that talloc CAN be broken out
successfully, or that the experiment may be worth trying.  The rest, given
the recent re-importing of ctdb, I think breaking out ctdb is not real
likely to work.  Libraries should have a good number of external users that
are willing to contribute to them, or they should be so stable, that we are
surprised when we see a commit in that area, IMHO.

talloc passes.  The rest do not.

And honestly, there may be technologies we wish to break out, like pidl,
etc.  This pass, does not deal with those.  Let us break the work down into
doable chunks.

I'm going to submit a patchset that conditionalizes zlib, as you requested,
and goes a step further and lets the AD DC build with it conditionalized.

Then iniparser.  (Which if someone wants to replace with code that does not
suck, I'll let that person do.  Frankly, anything that moves us closer to
not using it is a win.)
Then popt.

Then I think we have a moment of contemplation, on socket_wrapper and
ntdb+ccan.

Another on talloc, (which I may not be the person who removes.  Because
that is really about setting up the talloc project, which I may be the
wrong person to do.)

The rest, I have real concrete objections to and I recommend the team not
to do at this time.  But, I realize my voice is one among many.

Thanks,

-Ira

PS: From here on, I will be pushing to a git repo for these patches, I like
all your e-mail boxes too much to submit another patch this size to the
list.