The creeping bug is gone
I spent the day rewriting the unification algorithm for rows to solve the problem mentioned in the last post: there was a fundamental error in the concept. The good news is that the rewrite means that one gets (slightly) more understandable type error messages for row unification, but the program is now 150 lines longer. For now, all test cases I have work with the current version of the program, I hope they cover all situations, this time, and that I didn't miss some tricky type error somewhere.
I can now get back to the business of adding more basic types and basic operators to obtain a language that is actually usable. I hope to get a better idea about what kind of lists to implement tomorrow when I meet Philip Wadler.