Excerpts from Brandon Allbery's message of 2015-06-04 13:06:52 -0700:
Looks to me like it's confused about whether a ~ is part of an equality constraint or is a laziness annotation. The former would be illegal at that point, though, I'd think? Somewhere it believes a constraint might be possible there, via btype.
As ezyang says in the message I see just came in, you'll need extra production rules to distinguish that top level. Although I'd wonder why it believes an equality constraint is acceptable there in the first place; is that a lurking bug in the parser?
In general, the parser is more liberal than is actually required; we can give better error messages this way. Second, here is where it's ambiguous: T a ~b where T is a TyCon. Does this parse as T a (~b) or T a ~ b Parser doesn't currently distinguish these cases. Edward