From 3a3c5dde1a91357fa258682510f75e37e0340913 Mon Sep 17 00:00:00 2001
From: simonmar <unknown>
Date: Wed, 11 Dec 2002 15:55:17 +0000
Subject: [PATCH] [project @ 2002-12-11 15:55:17 by simonmar] Obj type must be
 forall a.a now, to avoid confusing GHC's new tail-calling
 scheme.

---
 Data/Dynamic.hs |   12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/Data/Dynamic.hs b/Data/Dynamic.hs
index c67ea20..6379d33 100644
--- a/Data/Dynamic.hs
+++ b/Data/Dynamic.hs
@@ -110,8 +110,16 @@ instance Show Dynamic where
 	  showsPrec 0 t   . 
 	  showString ">>"
 
-data Obj = Obj  
- -- dummy type to hold the dynamically typed value.
+type Obj = forall a . a
+ -- Dummy type to hold the dynamically typed value.
+ --
+ -- In GHC's new eval/apply execution model this type must
+ -- be polymorphic.  It can't be a constructor, because then
+ -- GHC will use the constructor convention when evaluating it,
+ -- and this will go wrong if the object is really a function.  On
+ -- the other hand, if we use a polymorphic type, GHC will use
+ -- a fallback convention for evaluating it that works for all types.
+ -- (using a function type here would also work).
 
 -- | A concrete representation of a (monomorphic) type.  'TypeRep'
 -- supports reasonably efficient equality.
-- 
1.7.10.4