From: simonpj@microsoft.com <unknown>
Date: Fri, 18 Dec 2009 10:54:34 +0000 (+0000)
Subject: Comments only
X-Git-Url: http://git.megacz.com/?p=ghc-hetmet.git;a=commitdiff_plain;h=e7dfa0c0430f77010dd41786d68615d19fc0e9fe

Comments only
---

diff --git a/compiler/typecheck/TcInstDcls.lhs b/compiler/typecheck/TcInstDcls.lhs
index c7fc8ab..37bebcf 100644
--- a/compiler/typecheck/TcInstDcls.lhs
+++ b/compiler/typecheck/TcInstDcls.lhs
@@ -205,6 +205,9 @@ So the ClassOp is just a cast; and so is the dictionary function.
 No need for fancy BuiltIn rules.  Indeed the BuiltinRule stuff does
 not work well for newtypes because it uses exprIsConApp_maybe.
 
+The INLINE on df is vital, else $cop_list occurs just once and is inlined,
+which is a disaster if $cop_list *itself* has an INLINE pragma.
+
 
 Note [Subtle interaction of recursion and overlap]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -771,6 +774,7 @@ tc_inst_decl2 dfun_id (VanillaInst monobinds uprags standalone_deriv)
 
              dfun_id_w_fun | isNewTyCon (classTyCon clas) 
                            = dfun_id   -- Just let the dfun inline; see Note [Single-method classes]
+                             `setInlinePragma` alwaysInlinePragma
                            | otherwise 
                            = dfun_id   -- Do not inline; instead give it a magic DFunFunfolding
 			     	       -- See Note [ClassOp/DFun selection]