All,
Just to clarify, I need the build to at least do the bootstrap phase, then I
can deal with every package on an individual basis even if it fails to
build, but it would be nice if it did, then I can check that I have captured
all appropriate output. I need to be able setup E-mail notifications for
each package, log its output etc and detect when it fails on a particular
target. If I can't run the individual target beyond bootstrap, then I can
not finish the setup of the nightly build properly. We could just have a
nightly build of bootstrap for now, how much use would that be?
Steve.
-----Original Message-----
From: Starlink development [mailto:[log in to unmask]] On Behalf Of
Rankin, SE (Stephen)
Sent: 12 October 2004 08:33
To: [log in to unmask]
Subject: Re: make world and nightly builds
Norman,
So it will get past bootstrap then on most platforms, is this what you are
telling me?
Steve.
-----Original Message-----
From: Starlink development [mailto:[log in to unmask]] On Behalf Of
Norman Gray
Sent: 11 October 2004 17:23
To: [log in to unmask]
Subject: Re: make world and nightly builds
Steve,
On 2004 Oct 11 , at 16.05, Rankin, SE (Stephen) wrote:
> The progress with nightly builds was that it was setup, but bootstrap
> only
> worked on Linux, and at the time CVS was still in a state of flux,
> will all
> the changes going on. Perhaps now is the time to start the nightly
> builds
> properly, but I would at least like the first build to work on all the
> base
> platforms, Solaris, Linux distributions that we support and possibly
> Alpha.
I don't think we have to -- or should -- wait until everything builds
on everything. The Linux build (and probably only one distribution,
say RH) and the OSX build are probably the first to aim for. Peter has
got all the way through the Solaris build once, but only with a little
babying, and we haven't even started on Alpha (will we ever?).
Also, we don't have to start with `make world'. We can start with,
say, `make /xxx/manifests/subpar', and thus get the nightly-build
system debugged, and only later, when we know that the whole thing
builds on a given platform as far as component X at least, get you to
change the nightly build target for that platform to /xxx/manifests/X.
That way, we will inch towards `make world'.
I _think_ there would be time for a full checkout, bootstrap and
configure each night, since the second two of those steps are now a
good deal faster than they once were. If not, then we could do the
complete thing once a week, and just do incremental rebuilds (cvs
update, then delete all the manifests except the buildsupport ones, and
then `make ???') once a night.
What this needs is for each of the platforms to be adopted by someone,
who can spend some time resolving the build problems that have recently
(surprisingly) appeared on that platform, and periodically letting you
know how you should set that platform's nightly-build target. They can
either fix the problems themself, or be resposible for badgering
whoever _can_ fix that component. Since the buildsystem is `mine' and
I have a Mac, I'll undertake to do this for the OSX build (which
currently falls over building gks, as the Mac linking problem seems to
have risen from the grave -- why!? argh!).
Volunteers? Which Linux distribution should we do this for? Do we
need to try this for Solaris?
> Do we have a tag in CVS to
> say this is a good source set to test?
Just go with the HEAD, since that's always what the nightly build
should work on (anyone disagree?). If you try `make world' it'll
currently fail on everything. But `make /xxx/manifests/subpar' will
very probably build on everything, and the folk who adopt a platform
can update you with later ones than that.
AstroGrid are doing nightly builds at present (and it's succeeding more
often than not). We ought to be really embarassed if we can't say the
same, on multiple platforms, by ADASS.
After chatting with DavidG, I've updated the
http://wiki.starlink.ac.uk/twiki/bin/view/Starlink/InterfaceWithUsers
page, noting those things that I suggest have to be done before ADASS.
Everyone: disagree and edit away! The first three tasks probably have
to be done at RAL, though we only have half of Steve currently and
Duncan's unavailable. Is there a need, and if so is there time and
access, for these to be farmed out at all?
If these can't be done, does anyone have any suggestions about what we
say to AstroGrid folk who ask about our code?
Sigh,
Norman
--
----------------------------------------------------------------------
Norman Gray : Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/ : www.starlink.ac.uk
|