[GHC] #12949: Pattern coverage checker ignores dictionary arguments

[GHC]

#12949: Pattern coverage checker ignores dictionary arguments
-------------------------------------+-------------------------------------
        Reporter:  dfeuer            |                Owner:  gkaracha
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.0.1
      Resolution:                    |             Keywords:
                                     |  PatternMatchWarnings
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Incorrect         |  Unknown/Multiple
  error/warning at compile-time      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by sighingnow):

 Hello,

 I'm working on this ticket and now I can tell what's going wrong with the
 case in comment:11. I think I need some help to go further and submit the
 patch.

 Look at the example in comment:11:

 {{{#!hs
 q :: forall a b . (Foo a, Foo b) => Result a b
 q = case foo :: Maybe a of
       Nothing -> case foo :: Maybe b of
                    Nothing -> Neither
                    Just c -> That c
 }}}

 When do `checkMatches` on inner `case` statement, we first call
 `mkInitialUncovered` to set up a NameEnv, in this function we use
 `getTmCsDs` to obtain the initial name env, we have

 {{{#!hs
 [(ds_d14i, foo), (ds_d14h, Nothing), (ds_d14h, foo)]
 }}}

 Nothing that these two `foo` have different `Id` (type), but the same
 `Name`. We then translate them as `ComplexEq`, only keeping the `Name` in
 `PmExprVar`. Then we get the `initTmState`.

 In `ConVar` case of `pmcheckHd`, we have ComplexEq `(ds_d14i, Just
 $pm_d14m)` and use `solveOneEq` to decide whether it is a valid candidate
 `PmCon`. In `solveOneEq`, we first do deep substitution by looking up the
 `Name` of `PmExprVar` from env. Now `ds_d14i` is replaced with `foo` and
 further replaced with `Nothing`, then `solveOneEq` fails and the `Just`
 pattern is redundant.

 ---------------------------------------------

 To solve this, we need to consider both name and type of variables when do
 substitution. In comment:7 gkaracha proposed that

 > Turning Name back into Id in the checker

 Should we move from `PmExprVar Name` to `PmExprVar Id` ? In
 `varDeepLookup`, when find the `PmExpr` from PrVarEnv, we could first
 check it's type then do further substitution. Do that make sense ?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12949#comment:16>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler