Duncan,

On 2005 Apr 8 , at 12.27, Malcolm J. Currie wrote:

> Duncan asked Steve and myself to either come along to the Site
> Managers'
> Meeting next week or give him an update on the status of the build
> system and software news.

Have a good time at the SMM.  There won't be anyone from Glasgow there
this time, as Graeme has now moved on to a different job within the
department, and his replacement, as of this week, is looking after four
groups within the department (and looking a bit shell-shocked), and
thus unlikely to feel able to spare the time for the SMM.  Perhaps next
time, though....

> [...] Anyway, would you like to comment
> on the current status of the GNU build system, and even dare to
> estimate when the CVS builds will all be working on a regular basis?

I think the build system can be said to be complete, in the sense that
there are no outstanding problems remaining to be addressed, though
there are one or two small tweaks I'd like to apply when I get the
chance and inclination.  Plus, of course (relax, Tim), submitting the
autoconf FPP patches to the autoconf mainline when CLRC decide that
this would be permissable, though that shouldn't hold up anyone else.

So either over the weekend or on Monday, I'll make SSN/78 version 1,
and put together the platform-specific install instructions and brief
migration document that I promised at the programmers' meeting.

When will it all be working on a nightly basis?  That's difficult to
say, but we do seem to be close now, after a number of weeks of
iterations finding wrinkle after wrinkle.  Current problems as of last
night's builds are:

*  starjava15 and rpms seem to have failed on a few systems;

*  sgmlkit is listed as failing on sparc, alpha and debian, though this
is a lie;

*  htx has failed on sparc, for the `bad file number' reason (Peter:
weird, this seems to be sporadic, as I'm sure that's a third component
that this has spontaneously appeared on -- I wonder if I should just
change the FD being used to a different one, on the off-chance that
it's being hijacked by something);

*  extreme has failed on alpha, though I can't see why;

*  starjava has failed on osx

So, if we exclude the starjava* problems (not classic), and exclude the
rpms target (important, but more than a developer would care about),
and ignore the sgmlkit `failures', then we can say that the classic
build is completely working on x86_64, i386, debian and osx, as of last
night.  And the sole sparc problem is this weird migrating-filenumber
problem, and the alpha one is this `extreme' buglet.  That sounds like
a whole lot of exclusions, but if we're talking about the perceptions
of someone who just wants to build the CVS version of the classic apps,
I think these do count as `everything working'.  So that's not bad!

Let's call it a beta release of the build system.  I'll put up a page
describing this, pointing to SSN/78 and the migration documents,
pointing folk at <http://www.starlink.ac.uk/Download/getcvs.html> (is
that a stable URL?), and inviting folk to try it out and tell us what
breaks on their systems.  David's experience with Dennis Kelly today
suggests that this will be a non-null set.  Do I encourage folk to
subscribe to stardev?

When is the SMM?

Norman


--
----------------------------------------------------------------------
Norman Gray  :  Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/  :  www.starlink.ac.uk