Procedural language implementation questions

[Sorry, my ISP managed to get itself on the RBL, so hub.org rejects my mail,
and if I send mail to "hackers" mailing list from a different account, it
gets rejected because that account isn't a subscriber to the list. So, I'm
sending this individually.]

I'm looking at implementing a procedural language for postgresql, and I
have a few questions I'd like to get feedback on before I start.

1. First of all, is anybody already working on this? I would like to
avoid duplication of effort, if possible.

2. Looking at the code, it seems that I basically need to do three things:
write a parser (using gram.y as a model), implement code to handle flow-
control nodes in the parse tree for the executor, and write a small bit of
glue for a handler (starting with the stub in "man create_language").
Is this basically correct? Is there anything in these areas that is
particularly fragile that I need to watch out for?

3. I want to use Python as a model, because Python syntax is very simple,
and the semantics are a very good fit with the postgres backend. One
problem I see, though, is that Python is dynamically typed, and looking
at the backend code, it looks like postgres is statically typed. If you
have a Datum, there doesn't seem to be any way to know what's in it.
Comments? Suggestions?

Thanks.

-Michael Robinson