Vidar started out by looking at the original genesis code/choices (like using Coral) and pretty rapidly did a good research on what else is available. He found some good tools, mostly based around eclipse and started hacking like there's no tomorrow. Just before my holiday (I just returned) I strongly suggested him not to add any more code but to package it all up nicely (as installing eclipse has been hell for me). Ease of installation is more important than adding one or two more features

The result blew me away. See the article on his weblog. Just one call to easy_install and it is installed. And it actually uses paster and its existing plone3 product skeleton.

Vidar worked hard and didn't need much outside encouragement. Smart, likes to dive in deep. Asks for help/advice when needed and listens to the sensible parts of the advice that people give him :-) He was a good fit with my style of mentoring (I've mentored a few students while I was a PhD student).

One possible snag regarding the original goal: the generator uses a special text format. Though, if you install eclipse, you can get a graphical editor working. And he's indicated that a transformation from UML is possible. So it is not 100% aligned with the original goal of genesis. What Vidar's done, though, looks like a way better basis than either Coral or the existing agx. And it bloody well works! Personally, I like the textual format, as it is versionable in subversion, for instance.

So: I'm real happy with how it all turned out! Especially the "easy_install" bit blew me away when I saw it last night. Vidar'll make more documentation in the days to come, so watch his blog when interested. And as he's done this gsoc project as part of his MSc study, he's finishing off a full report this week so there'll be enough documentation :-)

I'm mentoring Vidar Svansson for the Google summer of code. He's going to look at "genesis" again. Genesis is the name we use for a rewrite-archgenxml-from-scratch effort that hasn't gotten off the ground yet. One sprint, in which we selected coral as the underlying UML library, that's it. No real work done since.

Vidar's going to take a fresh look at it from a research perspective: he's doing it for his MSc thesis.

The gsoc is supposed to start next week, but Vidar has already started. Mostly reading up on zope3 and grok, trying to install coral and going to a code generation conference. I asked him a few days ago if he planned to be finished before the project started :-)

Coral is a pain to install. It is possible on linux if you use a pre-made rpm. Compiling on a mac: forget it. There's no public svn repository, so you have to use the tgz and figure out the changes since the last versions yourself. It sounds a bit risky.

At the conference, he discovered some new eclipse-based tool that looked OK to him, so he'll try things out with it. He says he'll be able to generate code for a simple case in a week or so...

End results that he/we aims for? A combination of actual code and scientific research. A theory on how to best model a plone web application. Code that shows it can be done :-)