So what next?

[Jim Gettys]

On Wed, 2004-04-14 at 05:06, Paul Anderson wrote:

> 
> Based on what I've followed in these discussions so far, I think both
> sides have valid arguments.  Using tools that are common and
> well-understood by a larger pool of open source developers is
> good for the organization in the long term.  At the same time, 
> I can understand that converting to a new build framework is 
> just something that gets in the way of the real fun of pulling
> in, evaluating, and experimenting with new features on their platform
> of choice.  It would be nice if we could find a way to balance 
> these short- and long-term needs accordingly.
> 

Just a point, Paul.

Much of the development is very hard to get properly tested
in a timely fashion if the way to build and install those
features implies building and installing all of an X 
distribution.  Examples are Xcb, new X extensions, new versions
of Mesa and DRI, and new device driver versions.  Lots
of these need serious soak time on lots of desktops.

One of the major issues with any new work right now is that
of "unintended changes" breaking your X environment, when we
need people to spend time running the new more experimental
pieces.  My experience of being able to have "make install"
work reliably in the monolithic distribution is quite poor
(at most 70%), and when things break, I then have no idea
of what caused the breakage, and whether what I wanted to
test had any part of the breakage.  This has made me, for
one, very unwilling to routinely test the entire distribution,
as I've spent too much time trying to repair unrelated damage.

And not everything reaches a point where they should start
soaking simultaneously: so harnessing everything to full up
releases to start testing seriously slows down development.
In this case, I mean time from initial coding to production
deployment, not solely the time the code is mostly written
and mostly working.  

And yes, if I'm experienced in the monolithic X distribution,
I can work around this by being
careful what I install when, but most people have no clue
about an X distribution's layout, so to do "make installs",
for individual pieces of the distribution, (e.g. a library, 
or driver, or new X server is quite complex.  
This limits the onset of serious testing to
when full up releases happen.

These interrelated issues are, in my mind, what has driving me
strongly toward modularity.

Given this, I do not see remaining with Imake has much to offer;
this is not to say that there are not issues that we need to
address with autotools a viable replacement.  Getting a handle
on remaining issues here is, of course, very important.
                             - Jim

-- 
Jim Gettys <Jim.Gettys at hp.com>
HP Labs, Cambridge Research Laboratory