Brian, but don't you think that you have to write a lot
of boilerplate code in Haskell?

I have never felt I was writing a lot of boilerplate. There are a lot of abstraction mechanisms in Haskell to avoid boilerplate.

Second, if Haskell should be more successful in the
real world there has to be a way of demonstrating
basic ideas of a big program to "customers". How
would you do this? Everybody knows UML class
diagrams, for example. In contrast, nobody knows
about termgraphs or lambda *g*.

I've never had to show a UML or ER diagram to any business people--usually they want a slideshow that is far simpler and a little prettier. The fact that "nobody knows about termgraphs or lambda" in your group means that you probably shouldn't be considering Haskell (for the same reason my bosses always asked me to document everything--"in case you get hit by a bus").

Thank you very much for contributing to the discussion.
Please assume, that you have to generate the code from
a model. Further assume, that you have no choice and
are not allowed to discuss the sense of this approach :-)
How should the code look like?

I am not sure if you are trying to solve a real problem or not. If you are solving a real problem, where you already happen to have an EMF model which you are required to generate code from, then I recommend to just do everything in Java using the existing tools built for EMF.

If you decide to still keep working in Haskell, and it works out well, please share your solution because I think many people here will be very interested. wxHaskell, OOHaskell, and O'Haskell are all starting points for this type of project.

- Brian