Make foldr2 a bit more strict

in order to make its RULES semantics preserving. This fixes #9495.

docs/users_guide/bugs.xml

@@ -291,7 +291,14 @@ checking for duplicates.  The reason for this is efficiency, pure and simple.
             if you get stuck on it.</para>
 	    </listitem>
 	  </varlistentry>
-
+          <varlistentry>
+            <term><literal>zip</literal> and <literal>zipWith</literal> semantics</term>
+            <para><literal>zip</literal> and <literal>zipWith</literal> can give
+            less defined results than the Report specifies in certain cases. This deviation
+            is needed to allow more opportunities for list fusion. In particular,
+            termination of the left list cannot be used to avoid hitting bottom in the
+            right list. See the documentation for details.</para>
+          </varlistentry>
 	  <varlistentry>
 	    <term><literal>Read</literal>ing integers</term>
 	    <listitem>

libraries/base/GHC/List.lhs

@@ -646,7 +646,7 @@ xs !! (I# n0) | isTrue# (n0 <# 0#) =  error "Prelude.(!!): negative index\n"
 foldr2 :: (a -> b -> c -> c) -> c -> [a] -> [b] -> c
 foldr2 k z = go
   where
-        go []    _ys     = z
+        go []    ys      = ys `seq` z -- see #9495 for the seq
         go _xs   []      = z
         go (x:xs) (y:ys) = k x y (go xs ys)
 {-# INLINE [0] foldr2 #-}
@@ -670,16 +670,6 @@ foldr2_right  k _z  y  r (x:xs) = k x y (r xs)
  #-}
 \end{code}
 
-The foldr2/right rule isn't exactly right, because it changes
-the strictness of foldr2 (and thereby zip)
-
-E.g. main = print (null (zip nonobviousNil (build undefined)))
-          where   nonobviousNil = f 3
-                  f n = if n == 0 then [] else f (n-1)
-
-I'm going to leave it though.
-
-
 Zips for larger tuples are in the List module.
 
 \begin{code}
@@ -687,10 +677,22 @@ Zips for larger tuples are in the List module.
 -- | 'zip' takes two lists and returns a list of corresponding pairs.
 -- If one input list is short, excess elements of the longer list are
 -- discarded.
+--
+-- NOTE: GHC's implementation of @zip@ deviates slightly from the
+-- standard. In particular, Haskell 98 and Haskell 2010 require that
+-- @zip [x1,x2,...,xn] (y1:y2:...:yn:_|_) = [(x1,y1),(x2,y2),...,(xn,yn)]@
+-- In GHC, however,
+-- @zip [x1,x2,...,xn] (y1:y2:...:yn:_|_) = (x1,y1):(x2,y2):...:(xn,yn):_|_@
+-- That is, you cannot use termination of the left list to avoid hitting
+-- bottom in the right list.
+
+-- This deviation is necessary to make fusion with 'build' in the right
+-- list preserve semantics.
 {-# NOINLINE [1] zip #-}
 zip :: [a] -> [b] -> [(a,b)]
+zip []     bs     = bs `seq` [] -- see #9495 for the seq
+zip _as    []     = []
 zip (a:as) (b:bs) = (a,b) : zip as bs
-zip _      _      = []
 
 {-# INLINE [0] zipFB #-}
 zipFB :: ((a, b) -> c -> d) -> a -> b -> c -> d
@@ -723,10 +725,23 @@ zip3 _      _      _      = []
 -- as the first argument, instead of a tupling function.
 -- For example, @'zipWith' (+)@ is applied to two lists to produce the
 -- list of corresponding sums.
+--
+-- NOTE: GHC's implementation of @zipWith@ deviates slightly from the
+-- standard. In particular, Haskell 98 and Haskell 2010 require that
+-- @zipWith (,) [x1,x2,...,xn] (y1:y2:...:yn:_|_) = [(x1,y1),(x2,y2),...,(xn,yn)]@
+-- In GHC, however,
+-- @zipWith (,) [x1,x2,...,xn] (y1:y2:...:yn:_|_) = (x1,y1):(x2,y2):...:(xn,yn):_|_@
+-- That is, you cannot use termination of the left list to avoid hitting
+-- bottom in the right list.
+
+-- This deviation is necessary to make fusion with 'build' in the right
+-- list preserve semantics.
+
 {-# NOINLINE [1] zipWith #-}
 zipWith :: (a->b->c) -> [a]->[b]->[c]
-zipWith f (a:as) (b:bs) = f a b : zipWith f as bs
-zipWith _ _      _      = []
+zipWith _f []     bs     = bs `seq` [] -- see #9495 for the seq
+zipWith _f _as    []     = []
+zipWith f  (a:as) (b:bs) = f a b : zipWith f as bs
 
 -- zipWithFB must have arity 2 since it gets two arguments in the "zipWith"
 -- rule; it might not get inlined otherwise

libraries/base/changelog.md

@@ -73,6 +73,10 @@
     the functions from `Data.List` (in other words, `Data.OldList` corresponds
     to `base-4.7.0.1`'s `Data.List`)
 
+  * `foldr2` (together with `zip` and `zipWith`) is made a bit stricter in the
+    second argument, so that the fusion RULES for it do not change the
+    semantics. (#9596)
+
 ## 4.7.0.1  *Jul 2014*
 
   * Bundled with GHC 7.8.3