<div dir="auto">Sorry for the confused traversal. That should've been<div dir="auto"><br></div><div dir="auto">traverse f (Leaf x) = Leaf <$> f x</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Jan 14, 2017 4:49 PM, "David Feuer" <<a href="mailto:david.feuer@gmail.com">david.feuer@gmail.com</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="auto"><div dir="auto">Right now, we define</div><div dir="auto"><br></div><div dir="auto">liftA2 :: Applicative f</div><div dir="auto"> => (a -> b -> c) -> f a -> f b -> f c<br></div><div dir="auto">liftA2 f x y = f <$> x <*> y</div><div dir="auto"><br></div>For some functors, like IO, this definition is just dandy. But for others, it's not so hot. For ZipList, for example, we get<div dir="auto"><br></div><div dir="auto">liftA2 f (ZipList xs) (ZipList ys) =</div><div dir="auto"> ZipList $ zipWith id (map f xs) ys</div><div dir="auto"><br></div><div dir="auto">In this particular case, rewrite rules will likely save the day, but for many similar types they won't. If we defined a custom liftA2, it would be the obviously-efficient</div><div dir="auto"><br></div><div dir="auto">liftA2 f (ZipList xs) (ZipList ys) =</div><div dir="auto"> ZipList $ zipWith f xs ys</div><div dir="auto"><br></div><div dir="auto">The fmap problem shows up a lot in Traversable instances. Consider a binary leaf tree:</div><div dir="auto"><br></div><div dir="auto">data Tree a = Bin (Tree a) (Tree a) | Leaf a</div><div dir="auto"><br></div><div dir="auto">The obvious way to write the Traversable instance today is</div><div dir="auto"><br></div><div dir="auto">instance Traversable Tree where</div><div dir="auto"> traverse _f Tip = pure Tip</div><div dir="auto"> traverse f (Bin p q) = Bin <$> traverse f p <*> traverse f q</div><div dir="auto"><br></div><div dir="auto">In this definition, every single internal node has an fmap! We could end up allocating a lot more intermediate structure than we need. It's possible to work around this by reassociating. But it's complicated (see Control.Lens.Traversal.<wbr>confusing[1]), it's expensive, and it can break things in the presence of infinite structures with lazy applicatives (see Dan Doel's blog post on free monoids[2] for a discussion of a somewhat related issue). With liftA2 as a method, we don't need to reassociate!</div><div dir="auto"><br></div><div dir="auto">traverse f (Bin p q) = liftA2 Bin (traverse f p) (traverse f q)</div><div dir="auto"><br></div><div dir="auto">The complication with Traversable instances boils down to an efficiency asymmetry in <*> association. According to the "composition" law,</div><div dir="auto"><div dir="auto" style="font-family:sans-serif"><br></div><div dir="auto" style="font-family:sans-serif">(.) <$> u <*> v <*> w = u <*> (v <*> w)<br></div><div dir="auto" style="font-family:sans-serif"><br></div><div dir="auto" style="font-family:sans-serif">But the version on the left has an extra fmap, which may not be cheap. With liftA2 in the class, we get a more balanced law:</div><div dir="auto" style="font-family:sans-serif"><br></div><div dir="auto" style="font-family:sans-serif">If for all x and y, p (q x y) = f x . g y, then liftA2 p (liftA2 q u v) = liftA2 f u . liftA2 g v</div><div dir="auto" style="font-family:sans-serif"><br></div></div><div dir="auto"><br></div><div dir="auto">[1] <a href="https://hackage.haskell.org/package/lens-4.15.1/docs/Control-Lens-Traversal.html#g:11" target="_blank">https://hackage.haskell.<wbr>org/package/lens-4.15.1/docs/<wbr>Control-Lens-Traversal.html#g:<wbr>11</a></div><div dir="auto"><br></div><div dir="auto">[2] <a href="http://comonad.com/reader/2015/free-monoids-in-haskell/" target="_blank">http://comonad.com/reader/<wbr>2015/free-monoids-in-haskell/</a></div></div>
</blockquote></div></div>