Norman,
>Overall, and despite Steve's pessimism, I think we're in a good
>position with this work, and ready to start distributing new-style
>packages quite soon.
Compliments will get you everywhere. I am only trying to get a handle on
what has happened, and what I need to do. I came back from A/L with nearly
200 E-mails in my stardev box!!
Steve.
-----Original Message-----
From: Starlink development [mailto:[log in to unmask]] On Behalf Of
Norman Gray
Sent: 13 July 2004 13:56
To: [log in to unmask]
Subject: Re: All cvs.
Steve,
On 2004 Jul 13 , at 10.29, Rankin, SE (Stephen) wrote:
> I thought that one of the main points of doing all of the
> autoconf/automake
> stuff was to allow people to build everything for them selves, so that
> we do
> not have to support directly any more platforms (encourage community
> and
> self support). I have already had requests for Tru64 stuff, and they
> are
> quite willing to build everything for themselves, how well they will
> get on
> with fixing stuff is another problem.
>
> If the above is not true, then we should have stayed with the old
> system
> which worked perfectly for us, and moved everything to CVS as is.
1. The distribution source tarballs are for folk who want the most
up-to-date _released_ versions of an application, where `released'
might mean anything from a full new version to this week's beta. The
idea is that this will build on any reasonable platform, and if it
doesn't, or they have to tweak it, that's a bug and should be fixed by
someone. It's autoconf that makes it possible for things to build on
platforms we don't formally support.
Anyone who wants to work on the software will have to work on the
repository, which means checking out the application, reading the
README, and typing `make world', or at least `make .../manifests/foo'
to configure and build foo and all its dependencies. So more
sophisticated users will need to use anonymous CVS to get the stuff,
and will have to liase with a committer (at present, anyone with an
account on cvs.starlink.ac.uk) to get any fixes back in. Anyone who
just wants the most recent functionality should steer clear of the
repository, and get and build a distribution tarball.
If someone fixes a bug in a distribution tarball, or has a build tweak
that works on a particular platform, they should get in touch with a
maintainer (whoever that is) to get it included in the CVS sources.
Perhaps we should make sure that at least one such person is named in
the component.xml file distributed with the tarball. They wouldn't
have to be a real bug-fixing maintainer, just someone with a
cvs.starlink id, who could check that code looked sane and was
documented, before applying a patch and, perhaps, adding a tag.
2. The distribution tarballs will have dependencies between them -- in
order to build one tarball you might need to have built and installed
another. The `use' dependencies in the component.xml files are for
working these out, and they'll be useful when the actual distribution
system is worked out. At present, the GenerateDependencies.java
program just works out build dependencies (which includes link
dependencies), but it's perfectly capable in principle of working out
the network of use dependencies, too.
This doesn't concern itself with component version numbers at present
(`"foo" needs component "bar" version x.y.z or better'). Perhaps it
ought to, if it's to be usable for distribution tarballs.
I thought I'd fixed this program, but David suggests that `make world'
isn't working in some circumstances. David: can you say where it isn't
working? Are the relevant component.xml files declaring correct
dependencies -- that is, are they wrong, or is
GenerateDependencies.java failing to process them correctly?
3. Mark is correct in saying that the distribution targets have not
been much tested, though I think David is releasing AST tarballs this
way (correct?). There's not an awful lot to test -- if you do "make
dist" you should get a tarball which is mostly correct, and fixing
which is just a matter of adding a couple of dist_ or nodist_ prefixes
here and there, or adding EXTRA_DIST values.
Overall, and despite Steve's pessimism, I think we're in a good
position with this work, and ready to start distributing new-style
packages quite soon.
Norman
--
----------------------------------------------------------------------
Norman Gray : Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/ : www.starlink.ac.uk
|