[PROPOSAL] To retire autoconf for 4.1

[Michael Adam]

Hi Volker,

On 2013-05-23 at 12:24 +0200, Volker Lendecke wrote:
> On Thu, May 23, 2013 at 12:00:14PM +0200, David Disseldorp wrote:
> > On Tue, 21 May 2013 19:08:06 +1000
> > Andrew Bartlett <abartlet at samba.org> wrote:
> > > ...
> > > The autoconf build system has supported us for a long time, but just as
> > > it is time that we retire SWAT, Samba 4.1 is the appropriate time to
> > > retire autoconf, and move to a single way to build a unified Samba
> > > project. 
> > 
> > +1 from me. Continuing to maintain both build systems is a heavy and
> > unnecessary burden.
> 
> Please hold back the patch until the issues I raised are
> addressed. I am certainly willing to hack things up, but the
> points are raised are really important to me.

I am not entirely certain what you mean by "addressed".

If this is a clear veto to removing the autoconf build
until you can work exactly the same way as with the autoconf
build (speed-wise and so on), I think this is difficult to sustain.
At least not without yourself willing to put a considerable
effort into improving waf.

Remember that the proposal to remove the autoconf build does not
come as a surprise: When we decided to use waf for our central
build it was agreed that we would keep the autoconf source3 build
for some time but that it should be removed at least not too far
after releasing 4.0 to reduce the overhead in maintaining two
build systems. (At that time it was not quite clear whether there
would be a 3.7, but 4.0 was clearly the border we had to cross to
be open for removal of the source3 build.)

All of your points are valid, and I think many of us can
learn from your workflow that uncovers these issues.
These are things that would be very valuable improvements
to our waf build.

But for me _personally_, they are not severe enough to block
the removal of autoconf. (I have not used our autoconf except
when I had to since 2 years or so.) And I notice the tendency
(also by me) to not port things done in the waf build to
autoconf, partly because it is sometimes so much more difficult
to do (It is not that autoconf itself is broken, we only use
it in a very suboptimal way). This is one of the reasons for me
to really want to have only one build system rather than keeping
a second possibly faster build system for a subtree.

Now with the very recent rise of this discussion, you see that many
people are very willing to pick up the issues you raised and improve
the situation, and some improvement has even already been achieved.

I would take this as proof enough that your concerns are being addressed.
What do you think?

So if there are not more fundamental concerns, i.e. things that
do not porperly work with waf, regressions from autoconf-built
fileservers or so, I think we should move forward with removing
the autoconf build.

We don't need to exaggeratedly rush the removal, but we should
IMHO do it before branching for 4.1.

These are my personal views, and I am happy to discuss this
further, but we need clear statements and positions.

Cheers - Michael


PS: Apologies, I wanted to have this shorter but the words won.