[GHC] #8503: New GeneralizedNewtypeDeriving check still isn't permissive enough

[GHC]

#8503: New GeneralizedNewtypeDeriving check still isn't permissive enough
-------------------------------------+------------------------------------
        Reporter:  goldfire          |            Owner:  goldfire
            Type:  bug               |           Status:  new
        Priority:  normal            |        Milestone:
       Component:  Compiler          |          Version:  7.7
      Resolution:                    |         Keywords:
Operating System:  Unknown/Multiple  |     Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown      |       Difficulty:  Unknown
       Test Case:                    |       Blocked By:
        Blocking:                    |  Related Tickets:
-------------------------------------+------------------------------------

Comment (by nomeata):

 Using just `TyCon.checkRecTc` is too permissive for `Coercible`.
 `checkRecTc` succeeds for `Fix (Either x)` (where `newtype Fix f = MkFix
 (f (Fix f))`) for all type `x`. But then we’d generate the instances

 {{{
 instance Coercible (Either x (Fix (Either x))) b => Coercible (Fix (Either
 x)) b
 instance Coercible a (Either x (Fix (Either x))) => Coercible a (Fix
 (Either x))
 }}}

 so trying to solve the constraint `Coercible (Fix (Either Age)) (Fix
 (Either Int))` will loop. (With the current implementation, which hooks
 into the general constraint resolving, I’d expect it to not loop at
 compile time, but create a diverging witness, and crash at run time.
 Neither of these are desired).

 Now what about your example, `newtype RecNT = MkRecNT (Either RecNT
 String)`. Assume that `Coercible` would work for it, and assume there is
 also `newtype RecNT2 = MkRecNT2 (Either RecNT2 String)`. Then, similarly,
 solving `Coercible RecNT RecNT2` will loop with the current
 implementation...

 So it is not a matter of relaxing the checks; if we want to support
 recursive newtypes in `Coercible`, the algorithm needs to be generally
 overhauled.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8503#comment:15>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler