GitLab

{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE EmptyDataDecls #-}

newtype Succ a = Succ a deriving Show
data    Zero   = Zero
instance Show Zero where show _ = "Zero"

type family Add a b
type instance Add Zero a     = a
type instance Add (Succ n) m = Succ (Add n m)

add :: a -> b -> Add a b
add = undefined

okay = show $ add Zero Zero
bad  = add Zero Zero

{- ghci transcript:

*Main> okay
"Zero"
*Main> bad
ghc-6.7.20070828: panic! (the 'impossible' happened)
  (GHC version 6.7.20070828 for i386-unknown-linux):
    readFilledBox t_a1D9{tv} [box]

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

*Main> show bad

<interactive>:1:0:
    No instance for (Show (Add Zero Zero))
      arising from a use of `show' at <interactive>:1:0-7
    Possible fix:
      add an instance declaration for (Show (Add Zero Zero))
    In the expression: show bad
    In the definition of `it': it = show bad
-}

The panic arises as follows: tcGhciStmts calls TcMatches.tcDoStmt to type check it <- bad, which in turn evaluates

 withBox liftedTypeKind $ \ pat_ty -> 
   tcMonoExpr rhs (mkAppTy m_ty pat_ty)

The withBox executes the readfilledBox that causes the panic, as tcMonoExpr promises to fill the boxes in its second argument. This promise is not fulfilled, as tcMonoBox defers the match Add Zero Zero ~ IO t_ayP.

Further up the call chain tcGhciStmts will eventually simplify the LIE and would discover the type mismatch (and hence abandon "Plan A", of which all this is part). Unfortunately, we never get there due to withBoxs attempt to read t_avP.

Trac metadata

Trac field	Value
Version	6.7
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler (Type checker)
Test case
Differential revisions
BlockedBy
Related
Blocking
CC	chak@cse.unsw.edu.au
Operating system	Unknown
Architecture	Unknown