On Thu, Apr 10, 2014 at 2:03 PM, Brandon Allbery <allbery.b@gmail.com> wrote:
Actually, what I'm saying is that it might be difficult to reliably catch it at that point; consider that the parse leading to that interpretation may not exist within ghc at that point, only an AST that may have been rearranged by uses of ConstraintKinds so as to lose its direct relationship to the parse that initially led to it.

Or worse, this is happening inside the typechecker and the tree it's manipulating has *no* relationship to the original parse or its AST, such that it has no idea whether it was on the wrong side of an =>. Depends on the internal representations used by the typechecker, and most likely doing it right means carrying around a lot more information than it currently does.

--
brandon s allbery kf8nh                               sine nomine associates
allbery.b@gmail.com                                  ballbery@sinenomine.net
unix, openafs, kerberos, infrastructure, xmonad        http://sinenomine.net