foldl

Composable, streaming, and efficient left folds

Version on this page:1.4.17
LTS Haskell 23.1:1.4.17
Stackage Nightly 2024-12-22:1.4.18
Latest on Hackage:1.4.18

See all snapshots foldl appears in

BSD-3-Clause licensed by Gabriella Gonzalez
Maintained by [email protected]
This version can be pinned in stack with:foldl-1.4.17@sha256:1d3e219819e471d55c11e651c5c8721845ef240a2adc15a578e0a492004f9edf,2632

foldl

Use this foldl library when you want to compute multiple folds over a collection in one pass over the data without space leaks.

For example, suppose that you want to simultaneously compute the sum of the list and the length of the list. Many Haskell beginners might write something like this:

sumAndLength :: Num a => [a] -> (a, Int)
sumAndLength xs = (sum xs, length xs)

However, this solution will leak space because it goes over the list in two passes. If you demand the result of sum the Haskell runtime will materialize the entire list. However, the runtime cannot garbage collect the list because the list is still required for the call to length.

Usually people work around this by hand-writing a strict left fold that looks something like this:

{-# LANGUAGE BangPatterns #-}

import Data.List (foldl')

sumAndLength :: Num a => [a] -> (a, Int)
sumAndLength xs = foldl' step (0, 0) xs
  where
    step (x, y) n = (x + n, y + 1)

That now goes over the list in one pass, but will still leak space because the tuple is not strict in both fields! You have to define a strict Pair type to fix this:

{-# LANGUAGE BangPatterns #-}

import Data.List (foldl')

data Pair a b = Pair !a !b

sumAndLength :: Num a => [a] -> (a, Int)
sumAndLength xs = done (foldl' step (Pair 0 0) xs)
  where
    step (Pair x y) n = Pair (x + n) (y + 1)

    done (Pair x y) = (x, y)

However, this is not satisfactory because you have to reimplement the guts of every fold that you care about and also define a custom strict data type for your fold. Hand-writing the step function, accumulator, and strict data type for every fold that you want to use gets tedious fast. For example, implementing something like reservoir sampling over and over is very error prone.

What if you just stored the step function and accumulator for each individual fold and let some high-level library do the combining for you? That’s exactly what this library does! Using this library you can instead write:

import qualified Control.Foldl as Fold

sumAndLength :: Num a => [a] -> (a, Int)
sumAndLength xs = Fold.fold ((,) <$> Fold.sum <*> Fold.length) xs

-- or, more concisely:
sumAndLength = Fold.fold ((,) <$> Fold.sum <*> Fold.length)

To see how this works, the Fold.sum value is just a datatype storing the step function and the starting state (and a final extraction function):

sum :: Num a => Fold a a
sum = Fold (+) 0 id

Same thing for the Fold.length value:

length :: Fold a Int
length = Fold (\n _ -> n + 1) 0 id

… and the Applicative operators combine them into a new datatype storing the composite step function and starting state:

(,) <$> Fold.sum <*> Fold.length = Fold step (Pair 0 0) done
  where
    step (Pair x y) n = Pair (x + n) (y + 1)

    done (Pair x y) = (x, y)

… and then fold just transforms that to a strict left fold:

fold (Fold step begin done) = done (foldl' step begin)

Since we preserve the step function and accumulator, we can use the Fold type to fold things other than pure collections. For example, we can fold a Producer from pipes using the same Fold:

Fold.purely Pipes.Prelude.fold ((,) <$> sum <*> length)
    :: (Monad m, Num a) => Producer a m () -> m (a, Int)

To learn more about this library, read the documentation in the main Control.Foldl module.

Quick start

Install the stack tool and then run:

$ stack setup
$ stack ghci foldl
Prelude> import qualified Control.Foldl as Fold
Prelude Fold> Fold.fold ((,) <$> Fold.sum <*> Fold.length) [1..1000000]
(500000500000,1000000)

How to contribute

Contribute a pull request if you have a Fold that you believe other people would find useful.

Development Status

Build Status

The foldl library is pretty stable at this point. I don’t expect there to be breaking changes to the API from this point forward unless people discover new bugs.

License (BSD 3-clause)

Copyright (c) 2016 Gabriella Gonzalez All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

  • Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.

  • Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

  • Neither the name of Gabriella Gonzalez nor the names of other contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Changes

1.4.17

  • Add Fold1 utilities: purely, purely_, premap, handles, foldOver, folded1
  • Add pattern synonym Fold1_ that makes the initial, step and extraction functions explicit.

1.4.16

1.4.15

  • Add Cosieve and Costrong instances

1.4.14

1.4.13

  • New “Control.Foldl.NonEmpty” module for folding non-empty containers

1.4.12

  • Data.Functor.Extend.Extended instances for Fold / FoldM
  • Remove dependency on mwc-random

1.4.11

  • Fix doctest failure when built against newer versions of the hashable package

1.4.10

  • Fix space leaks in scan / scanM

1.4.9

  • Implement vector utility more efficiently

1.4.8

  • Only depend on semigroups for older GHC versions

1.4.7

  • Add foldByKey{,Hash}Map functions

1.4.6

  • Add nest/predropWhile/drop/dropM

1.4.5

  • Increase upper bound on containers
  • Add either/eitherM

1.4.4

  • Increase lower bound on base
  • Change mean to be more numerically stable

1.4.3

  • Add Control.Scanl.scanr
  • Increase upper bound on mwc-random

1.4.2

  • Add Semigroupoid instance for Fold
  • Increase upper bound on contravariant and profunctors

1.4.1

  • Add Control.Scanl
  • Drop support for GHC 7.8 and older

1.4.0

  • BREAKING CHANGE: Change type of premapM to accept a monadic function

1.3.7

  • Add groupBy

1.3.6

  • Documentation improvements

1.3.5

  • Add Choice instance for Fold

1.3.4

  • Add prefilter and prefilterM

1.3.3

  • Add back the old vector as vectorM

1.3.2

  • Compatibility with Semigroup becoming a super-class of Monoid
  • Fix asin for Fold

1.3.1

  • Fix asin for FoldM

1.3.0

  • BREAKING CHANGE: Change vector to be a pure Fold (which is faster, too!)

1.2.5

  • Add support for folding new containers: hashSet, map, and hashMap
  • Add prescan/postscan which generalize scan to Traversable types

1.2.4

  • Add lazy folds for Text and ByteString
  • Documentation fixes and improvements

1.2.3

  • Add lookup

1.2.2

  • Add numerically stable mean, variance, and std folds
  • Add Control.Foldl.{Text,ByteString}.foldM
  • Add foldOver/foldOverM

1.2.1

  • Performance improvements
  • Re-export filtered

1.2.0

  • Breaking change: Fix handles to fold things in the correct order (was previously folding things backwards and also leaking space as a result). No change to behavior of handlesM, which was folding things in the right order
  • Breaking change: Change the Monoid used by Handler/HandlerM
  • Add folded

1.1.6

  • Add maximumBy and minimumBy

1.1.5

  • Increase lower bound on base from < 4 to < 4.5

1.1.4

  • Increase upper bound on comonad from < 5 to < 6

1.1.3

  • Increase upper bound on profunctors from < 5.2 to < 5.3
  • Add mapM_, hoists, purely, and impurely

1.1.2

  • Add lastN, randomN, sink, and duplicateM
  • Add Comonad instance for Fold
  • Add Profunctor instance for FoldM

1.1.1

  • Increase upper bound on vector from < 0.11 to < 0.12

1.1.0

  • Breaking change: Rename pretraverse/pretraverseM to handles/handlesM
  • Add Handler
  • Export EndoM

1.0.11

  • Add Profunctor instance for Fold

1.0.10

  • Add random and _Fold1

1.0.9

  • Increase upper bound on primitive from < 0.6 to < 0.7

1.0.8

  • Add revList

1.0.7

  • Add Num and Fractional instances for Fold/FoldM
  • Add count fold for Text and ByteString

1.0.6

  • Add pretraverse and pretraverseM

1.0.5

  • Add lastDef

1.0.4

  • Increase upper bounds on transformers from < 0.4 to < 0.6
  • Add nub, eqNub, and set

1.0.3

  • Add scan, generalize, simplify, and premapM

1.0.2

  • Add list and vector folds
  • Add fold function for Text and ByteString

1.0.1

  • Add support for ByteString and Text folds
  • Add Monoid instance for Fold/FoldM

1.0.0

  • Initial release