GitLab

As was pointed out in this SO question, GHC appears to incorrectly accept (->) as a valid "consym" in data declarations. That is, the following code compiles:

data Type
  = TBool
  | TInt
  | (->) Type Type

though attempts to use the constructor are correctly rejected as parse errors:

wontwork ((->) a b) = (->) b a  -- both "->" are parse errors

The expected behavior is that, since -> doesn't start with a colon, it ought to be rejected as a parse error.

I can duplicate this with GHC 8.6.5 and the latest HEAD.

The issue seems to be in tyConToDataCon which as per the Note "Parsing data constructors is hard" reinterprets a potential type constructor expression as a data constructor expression. It checks the validity of the constructor with isLexCon which calls isLexConSym, which specifically permits -> in addition to constructors that start with a colon (startsConSym).