[GHC] #15052: DeriveAnyClass instances may skip TypeError constraints

[GHC]

#15052: DeriveAnyClass instances may skip TypeError constraints
-------------------------------------+-------------------------------------
        Reporter:  jcpetruzza        |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
                                     |  CustomTypeErrors
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 When `deriving` infers an instance context, it collects some set of
 constraints, simplifies them as much as possible (discharging constraints
 when possible), and uses the remaining constraints as the context.

 For instance, in `data T a = MkT a Int deriving Show`, you start with the
 constraint set `(Show a, Show Int)`. GHC is able to discharge the `Show
 Int` constraint, since there's a top-level instance for that, but not the
 `Show a` constraint, so the final instance context is `Show a`.

 Returning to your `TypeError` example, you start with a constraint set
 that contains many constraints, with `TypeError (...)` sprinkled among
 them. There is no way to discharge a `TypeError (...)` constraint—if you
 could, you would be able to do lots of things you weren't supposed to be
 able to do, since `TypeError` is designed to be the context for
 unreachable code! Thus, `TypeError (...)` ends up in the final instance
 context.

 At this point, you might be thinking "well why doesn't the constraint
 solver immediately error when it encounters any constraint mentioning
 `TypeError`"? That might be feasible, but again, it would be a breaking
 change—it would explicitly go against the convention established by
 #14339.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15052#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler