Currently I am about halfway done writing a parser for mypy. At this point in time, it can process a set of rules written in python, and match those rules to a list of tokens. I am currently working on the tokenization part of the parser (up until now, I've been using hard-coded lists of tokens to test code).
The end goal is to be able to read in a ebnf-formatted grammar file, and output optimized parsing code that can be used to read the mypy files.

Looks like this never led to anything, so closing. Note that @ddfisher is working on making mypy use a (modified version of) the Python ast module, for speed.

See #1353