skaller writes:
 > I am getting an undefined label error. 
 > I know why. I do not know how to fix the problem:
 > 
 > 	(Unix.fstat argument).st_kind

You want (Unix.fstat argument).Unix.st_kind

(Is this in the docs?  I found it once when looking at the camlp4 parser.)

This is definitely one of those cases where type inference doesn't
seem such a good idea ... it feels so like K&R C, sometimes having to
fall back on giving labels little prefixes ... I wonder if some kind
of inference would be possible, not going the whole way to the
tag-based system used for classes, but at least something that
maintains a set of hypotheses consistent with the labels used in
connection with a value, which could then be narrowed down
contextually.