Hacker News Reader

Hacker News

Comment by mduerksen | Hacker News Reader

Parent Full thread

mduerksen

The benchmark is wrong atm, the compared functions do not yield the same result.

(comp (map inc) (filter even?)) means filtering first, then incrementing.

The opposite for (->> data (map inc) (filter even?) ...

- which is not the same. And of course, it also means that the latter has to increment the double amount of numbers.

EDIT: It was me who was wrong, thanks for the corrections. Pitfall successfully identified :)

View on HN

Nope, that's the funny feature of transducers. Like lenses in Haskell, transducers compose in reversed order.

    (sequence (comp (map inc) (filter even?)) (range 10))
    ;;=> (2 4 6 8 10)
    (->> (range 10) (map inc) (filter even?))
    ;;=> (2 4 6 8 10)

Both test functions in transduce-bench return 250000500000.

It seems counterintuitive, but composing transducers yields a reducing function that runs the transformation steps left->right. What you are composing is the reducing function transformations (right to left, ordinary composition), but their result, having been built inside-out, runs the steps outside-in. So (comp tx1 tx2...) runs in same order as (->> xs tx1 tx2...)

Reply on news.ycombinator.com