[Rivet] Getting Rivet and AGILe released

[Andy Buckley]

Jonathan Butterworth wrote:
> Hi Andy et al,

Hi,

I've come back from hols a bit early for the ATLAS Herwig++ session. Is
anyone else at CERN at the moment? It would be good to get all known
bugs whacked this week and decide on where to put the emphasis for the
next release (other than improving the histogramming).

> We have one more session today. I think the general perception is that
> the first day involving rivet was close to the edge. Some say it was a
> disaster, some 5/10. Many (mostly small) irritating problems. However,
> not all of them were rivet issues. Also most of the students seemed
> reasonably content anyway - I think the generator authors and practical
> organisers felt the pain most. Certainly we would not have have been
> happy if all days had been like that.

What sort of problems in particular? It was unfortunate how this
progressed - weeks of "only a couple of issues to fix", and then
everyone found problems in the last couple of days! Given that the whole
gen-AGILe-Rivet chain is now complex, and that we have increasing users,
unit testing is looking less like a luxury and more like a necessity...

> Yesterday, after much work from several people (testing code before hand
> on the actual system here, and thinking through the physics instructions
> more) I think we would call an unqualified success. As far as I know
> there were no rivet/agile issues, and we spent nearly all the time
> discussing the physics and the generators.

That's good. I didn't see many Rivet commits between the sessions,
though, and no new tickets - were they genuine logic bugs or things that
were found to be counter-intuitive? Next session, I'll try to be there, too.

> But I agree we should release the "MCnet school" version ASAP and if
> James can do this great - I am away for two weeks now too.

Yes, I'd like to get it out, too. I see it's not there yet - has there
been any activity in the last week? The "Usable by MCnet summer school"
milestone still allegedly has three bugs blocking it: the rivetgun param
ordering, the W analysis and the FastJets bug. I thought all of these
had been solved... the W analysis at least to the point where a release
is possible, even if some extra tweaking is needed to make it "perfect".
Is that right? Can we release?

> We need to define some acceptance tests for rivet/agile releases. I
> think a matrix of analyses X generators with green, yellow or red boxes
> (works, not tested, broken) would be a good start. I think for the MCnet
> school we should use the tutorial experience as the acceptance test, but
> still produce the matrix. For future releases we can think further. In
> particular we could ask some users to become testers for some parts of
> the releases.

I agree that testing is essential now. I'd like to get us using
Boost.Test in the next release: can someone (James?) get started on that
after the 1.1.1 release?

David Grellscheid may be able to help us with using the Hudson
continuous integration system that he's set up for Herwig++ development.
That seems a really good way to do it, especially if we can bypass the
learning curve via David's experience. I'll talk to him about that when
I'm back in Durham (in mid-September!)

> My feeling is that from this school onwards, assuming today also goes
> well, we are likely to have a step up in number of users. Important to
> maintain the momentum and not piss them off.

Right!

Andy