[GHC] #8827: Inferring Safe mode with GeneralizedNewtypeDeriving is wrong

[GHC]

#8827: Inferring Safe mode with GeneralizedNewtypeDeriving is wrong
-------------------------------------+------------------------------------
        Reporter:  goldfire          |            Owner:
            Type:  bug               |           Status:  new
        Priority:  normal            |        Milestone:  7.8.1
       Component:  Compiler          |          Version:  7.8.1-rc2
      Resolution:                    |         Keywords:
Operating System:  Unknown/Multiple  |     Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown      |       Difficulty:  Unknown
       Test Case:                    |       Blocked By:
        Blocking:                    |  Related Tickets:  8226, 8745
-------------------------------------+------------------------------------

Comment (by nomeata):

 > I'll admit to liking the general flavor of comment:27, but the details
 elude me somewhat. What, precisely, would go after the word deriving?

 Well, either `data MyList a = [a] deriving Coercible`, which will give you
 the same instance as described in the paper, but independent of the scope
 of the constructor.

 Or `deriving instance Coercible a b => Coercible (Map k a) (Map k b)`,
 which allows you to control the coercing

 This is even strictly more powerful that our current system: It would
 allow the author of `Map` to export the instance above, while still using
 `coerce` in his code (with constructors in scope and no abstraction to
 follow) to coerce the key, if he wishes to do so.

 > What's the interaction between derived instances and role annotations?
 Could a derived instance be more restrictive than the assigned roles? (It
 certainly couldn't be less restrictive!) How would a user specify that?

 As you say: The derived instance have to adhere the roles, but may be more
 restrictive (if created using a standalone deriving instance). The non-
 standalone-deriving instance would be the one that we have currently.

 OTOH, this gets hairy once we mix tycons with exported constructors and
 tycons without, but with exported Coercible instances, in a new data
 declaration and try to find out when we can coerce under ''that''... and
 that issue is well handled by roles and role inference. *shrug*

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8827#comment:30>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler