[commit: ghc] ghc-7.6: Be lazier when typechecking data type contexts (Trac #7321) (e7be164)

Repository : ssh://darcs.haskell.org//srv/darcs/ghc

On branch  : ghc-7.6

http://hackage.haskell.org/trac/ghc/changeset/e7be164fcd44821af9ffd9b8ec0ff2a0529fa6aa

>---------------------------------------------------------------

commit e7be164fcd44821af9ffd9b8ec0ff2a0529fa6aa
Author: Simon Peyton Jones <[email protected]>
Date:   Fri Oct 12 08:15:44 2012 +0100

    Be lazier when typechecking data type contexts (Trac #7321)
    
    We should be lazy when type-checking the equality-contraint
    part of a data constructor's type, to make the knot-tying
    work out right.  The fact that it's always worked before
    is a fluke: no one else wrote a GADT whose type index mentions
    itself
      data T a wher
        MkT :: T (T Int)

>---------------------------------------------------------------

 compiler/iface/TcIface.lhs |   25 +++++++++++--------------
 1 files changed, 11 insertions(+), 14 deletions(-)

diff --git a/compiler/iface/TcIface.lhs b/compiler/iface/TcIface.lhs
index 80c2029..55b9978 100644
--- a/compiler/iface/TcIface.lhs
+++ b/compiler/iface/TcIface.lhs
@@ -570,20 +570,17 @@ tcIfaceDataCons tycon_name tycon _ if_cons
      = bindIfaceTyVars univ_tvs $ \ univ_tyvars -> do
        bindIfaceTyVars ex_tvs    $ \ ex_tyvars -> do
         { name  <- lookupIfaceTop occ
-        ; eq_spec <- tcIfaceEqSpec spec
-        ; theta <- tcIfaceCtxt ctxt     -- Laziness seems not worth the bother here
-                -- At one stage I thought that this context checking *had*
-                -- to be lazy, because of possible mutual recursion between the
-                -- type and the classe: 
-                -- E.g. 
-                --      class Real a where { toRat :: a -> Ratio Integer }
-                --      data (Real a) => Ratio a = ...
-                -- But now I think that the laziness in checking class ops breaks 
-                -- the loop, so no laziness needed
-
-        -- Read the argument types, but lazily to avoid faulting in
-        -- the component types unless they are really needed
-        ; arg_tys <- forkM (mk_doc name) (mapM tcIfaceType args)
+
+        -- Read the context and argument types, but lazily for two reasons
+        -- (a) to avoid looking tugging on a recursive use of 
+        --     the type itself, which is knot-tied
+        -- (b) to avoid faulting in the component types unless 
+        --     they are really needed
+        ; ~(eq_spec, theta, arg_tys) <- forkM (mk_doc name) $
+             do { eq_spec <- tcIfaceEqSpec spec
+                ; theta   <- tcIfaceCtxt ctxt
+                ; arg_tys <- mapM tcIfaceType args
+                ; return (eq_spec, theta, arg_tys) }
         ; lbl_names <- mapM lookupIfaceTop field_lbls
 
         -- Remember, tycon is the representation tycon