[Rivet] Rivet issues with Genser libraries

[Alberto Ribon]

  Hi Andy,

  after some discussion with Witek and some thinking, I can 
  give you my point of view.

  First, I think that the Rivet community (both developers
  and users) is an important "client" for the LCG Generator
  Services, and therefore we consider seriously your requests.

  Second, I think that we, LCG Generator Services, should try 
  to work together with the Rivet developers to make the 
  Genser building machinery more general and robust. 
  I agree with you that autoconf is probably the right solution.
 
  The hand-written shell scripts, that are currently used in
  Genser for those generators that do not come with their own
  building machinery, were, in my opinion, the right solution
  when the new structure of Genser was put in place and the main
  priority was to deliver the libraries as soon as possible to 
  the LHC experiments. These scripts have achieved their goal
  and the experiments were satisfied. 
  However, for a more general usage, as Rivet showed, these
  scripts are not enough, and keep improving them seems to me
  a difficult and likely too ambitious activity.

  Moving to a new system based on autoconf seems to me the
  right thing to do at this stage, especially because we do
  not need to start from scratch, taking advantage of the
  experience, and assistence/help that you, Rivet developers,
  are willing to give us.

  To be concrete, I would propose, as Witek already suggested, 
  to start with the two main Fortran generators, Pythia and 
  Herwig. If you can send us your building structure based on
  autoconf for these two generators we can start to have a look
  and to play with.

  Let's keep in touch,
                       Alberto



On Fri, 25 Jan 2008, Andy Buckley wrote:

->Hi Alberto,
->
->Quite a long time ago, we (Rivet developers) decided together with
->Genser that we would stop maintaining our own parallel generator
->bundles, which used autoconf and libtool, and move to using Genser's
->libraries. Genser, particularly Witek, have been very supportive and
->helpful. We finally got round to properly migrating to the Genser
->distributions in the last few weeks, and it's not been a simple process,
->with a lot of problems remaining.
->
->In summary, our major problem is that the Genser libraries aren't
->portably buildable on platforms other than Scientific Linux with the
->CERN AFS area mounted in the usual place. We need a more generally
->useful set of generators, since our users and developers are using
->laptops, Apple Macs and various versions of Linux other than SL(C). We
->also need some portable and reliable way of automatically assembling a
->local build of the whole Genser AFS area, with a documented and
->programmatic directory structure.
->
->More specifically, we've set up a wiki page for AGILe (our generator
->interface library) which catalogues our problems:
->http://projects.hepforge.org/agile/trac/wiki/GenserBuildProblems
->We can translate these problems into bugs on the Genser Savannah
->tracker, if it's being used --- my impression at the moment is that it
->isn't.
->
->We're very happy to help Genser to fix these problems --- it's in our
->interest to do so and we already have experience of maintaining portable
->builds of Fortran generators via autotools: this could be easily
->extended to provide the LCG features that Genser needs.
->My impression is that Rivet is a testbed for Genser usage which can help
->to improve the overal robustness of the system against changes in the
->way that LCG platforms work. So, despite being lower priority than the
->experiments, I think that our needs are far from contradictory to
->theirs. Please let me know what you think and if we can be of assistance...
->
->Cheers,
->Andy
->
->