Commit 302265d5 authored by's avatar
Browse files

Remove NOINLINE strictness hack

The stricteness analyser used to have a HACK which ensured that NOINLNE things
were not strictness-analysed.  The reason was unsafePerformIO. Left to itself,
the strictness analyser would discover this strictness for unsafePerformIO:
	unsafePerformIO:  C(U(AV))
But then consider this sub-expression
	unsafePerformIO (\s -> let r = f x in 
			       case writeIORef v r s of (# s1, _ #) ->
			       (# s1, r #)
The strictness analyser will now find that r is sure to be eval'd,
and may then hoist it out.  This makes tests/lib/should_run/memo002

Solving this by making all NOINLINE things have no strictness info is overkill.
In particular, it's overkill for runST, which is perfectly respectable.
	f x = runST (return x)
This should be strict in x.

So the new plan is to define unsafePerformIO using the 'lazy' combinator:

	unsafePerformIO (IO m) = lazy (case m realWorld# of (# _, r #) -> r)

Remember, 'lazy' is a wired-in identity-function Id, of type a->a, which is 
magically NON-STRICT, and is inlined after strictness analysis.  So
unsafePerformIO will look non-strict, and that's what we want.

Now we don't need the hack in the strictness analyser.
parent 36b27193
......@@ -902,22 +902,26 @@ seqId
ty = mkForAllTys [alphaTyVar,openBetaTyVar]
(mkFunTy alphaTy (mkFunTy openBetaTy openBetaTy))
[x,y] = mkTemplateLocals [alphaTy, openBetaTy]
-- gaw 2004
rhs = mkLams [alphaTyVar,openBetaTyVar,x,y] (Case (Var x) x openBetaTy [(DEFAULT, [], Var y)])
-- lazy :: forall a?. a? -> a? (i.e. works for unboxed types too)
-- Used to lazify pseq: pseq a b = a `seq` lazy b
-- No unfolding: it gets "inlined" by the worker/wrapper pass
-- Also, no strictness: by being a built-in Id, it overrides all
-- the info in PrelBase.hi. This is important, because the strictness
-- Also, no strictness: by being a built-in Id, all the info about lazyId comes from here,
-- not from GHC.Base.hi. This is important, because the strictness
-- analyser will spot it as strict!
-- Also no unfolding in lazyId: it gets "inlined" by a HACK in the worker/wrapper pass
-- (see WorkWrap.wwExpr)
-- We could use inline phases to do this, but that would be vulnerable to changes in
-- phase numbering....we must inline precisely after strictness analysis.
= pcMiscPrelId lazyIdName ty info
info = noCafIdInfo
ty = mkForAllTys [alphaTyVar] (mkFunTy alphaTy alphaTy)
lazyIdUnfolding :: CoreExpr -- Used to expand LazyOp after strictness anal
lazyIdUnfolding :: CoreExpr -- Used to expand 'lazyId' after strictness anal
lazyIdUnfolding = mkLams [openAlphaTyVar,x] (Var x)
[x] = mkTemplateLocals [openAlphaTy]
......@@ -530,34 +530,6 @@ by dmdAnalTopBind.
mk_sig_ty never_inline thunk_cpr_ok rhs (DmdType fv dmds res)
| never_inline && not (isBotRes res)
-- Don't strictness-analyse NOINLINE things. Why not? Because
-- the NOINLINE says "don't expose any of the inner workings at the call
-- site" and the strictness is certainly an inner working.
-- More concretely, the demand analyser discovers the following strictness
-- for unsafePerformIO: C(U(AV))
-- But then consider
-- unsafePerformIO (\s -> let r = f x in
-- case writeIORef v r s of (# s1, _ #) ->
-- (# s1, r #)
-- The strictness analyser will find that the binding for r is strict,
-- (becuase of uPIO's strictness sig), and so it'll evaluate it before
-- doing the writeIORef. This actually makes tests/lib/should_run/memo002
-- get a deadlock!
-- Solution: don't expose the strictness of unsafePerformIO.
-- But we do want to expose the strictness of error functions,
-- which are also often marked NOINLINE
-- {-# NOINLINE foo #-}
-- foo x = error ("wubble buggle" ++ x)
-- So (hack, hack) we only drop the strictness for non-bottom things
-- This is all very unsatisfactory.
= (deferEnv fv, topSig)
| otherwise
= (lazy_fv, mkStrictSig dmd_ty)
dmd_ty = DmdType strict_fv final_dmds res'
......@@ -137,7 +137,7 @@ wwExpr e@(Note InlineMe expr) = returnUs e
wwExpr e@(Var v)
| v `hasKey` lazyIdKey = returnUs lazyIdUnfolding
| otherwise = returnUs e
-- Inline 'lazy' after strictness analysis
-- HACK alert: Inline 'lazy' after strictness analysis
-- (but not inside InlineMe's)
wwExpr (Lam binder expr)
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment