ignore RealWorld in size_expr; flag to keep w/w from creating sharing

size_expr now ignores RealWorld lambdas, arguments, and applications.

Worker-wrapper previously removed all lambdas from a function, if they
were all unused. Removing *all* value lambdas is no longer
allowed. Instead (\_ -> E) will become (\_void -> E), where it used to
become E. The previous behavior can be recovered via the new
-ffun-to-thunk flag.

Nofib notables:

----------------------------------------------------------------
        Program               O2          O2 newly ignoring RealWorld
                                          and not turning function
                                          closures into thunks
----------------------------------------------------------------

 Allocations

  comp_lab_zift            333090392%           -5.0%
reverse-complem            155188304%           -3.2%

        rewrite             15380888%           +4.0%
         boyer2              3901064%           +7.5%

rewrite previously benefited from fortunate LoopBreaker choice that is
now disrupted.

A function in boyer2 goes from $wonewayunify1 size 700 to size 650,
thus gets inlined into rewritelemmas, thus exposing a parameter
scrutinisation, thus allowing SpecConstr, which unfortunately involves
reboxing.

Run Time

 fannkuch-redux                 7.89%          -15.9%

            hpg                 0.25%           +5.6%
           wang                 0.21%           +5.8%

/shrug

1 files changed, 20 insertions, 4 deletions

diff --git a/compiler/stranal/WwLib.lhs b/compiler/stranal/WwLib.lhs
index fb9396e5ea..810db2069b 100644
--- a/compiler/stranal/WwLib.lhs
+++ b/compiler/stranal/WwLib.lhs
@@ -140,7 +140,7 @@ mkWwBodies dflags fun_ty demands res_info one_shots
         -- Do CPR w/w.  See Note [Always do CPR w/w]
 	; (wrap_fn_cpr, work_fn_cpr,  cpr_res_ty) <- mkWWcpr res_ty res_info
 
-	; let (work_lam_args, work_call_args) = mkWorkerArgs work_args all_one_shots cpr_res_ty
+	; let (work_lam_args, work_call_args) = mkWorkerArgs dflags work_args all_one_shots cpr_res_ty
 	; return ([idDemandInfo v | v <- work_call_args, isId v],
                   wrap_fn_args . wrap_fn_cpr . wrap_fn_str . applyToVars work_call_args . Var,
                   mkLams work_lam_args. work_fn_str . work_fn_cpr . work_fn_args) }
@@ -184,23 +184,39 @@ add a void argument.  E.g.
 We use the state-token type which generates no code.
 
 \begin{code}
-mkWorkerArgs :: [Var]
+mkWorkerArgs :: DynFlags -> [Var]
              -> Bool    -- Whether all arguments are one-shot
 	     -> Type	-- Type of body
 	     -> ([Var],	-- Lambda bound args
 		 [Var])	-- Args at call site
-mkWorkerArgs args all_one_shot res_ty
-    | any isId args || not (isUnLiftedType res_ty)
+mkWorkerArgs dflags args all_one_shot res_ty
+    | any isId args || not needsAValueLambda
     = (args, args)
     | otherwise	
     = (args ++ [newArg], args ++ [realWorldPrimId])
     where
+      needsAValueLambda =
+        isUnLiftedType res_ty
+        || not (gopt Opt_FunToThunk dflags)
+           -- see Note [Protecting the last value argument]
+
       -- see Note [All One-Shot Arguments of a Worker]
       newArg = if all_one_shot 
                then setOneShotLambda voidArgId
                else voidArgId     
 \end{code}
 
+Note [Protecting the last value argument]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+If the user writes (\_ -> E), they might be intentionally disallowing
+the sharing of E. Since absence analysis and worker-wrapper are keen
+to remove such unused arguments, we add in a void argument to prevent
+the function from becoming a thunk.
+
+The user can avoid that argument with the -ffun-to-thunk
+flag. However, removing all the value argus may introduce space leaks.
+
 Note [All One-Shot Arguments of a Worker]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~