Skip to content

Latest commit

 

History

History
183 lines (138 loc) · 6.51 KB

README.md

File metadata and controls

183 lines (138 loc) · 6.51 KB

Chain

Dedication

all day long they work so hard / 'til the sun is going down / working on the highways and byways / and wearing, wearing a frown / you hear them moaning their lives away / then you hear somebody say: /

that's the sound of the men / working on the chain gang / that's the sound of the men / working on the chain gang /

--Sam Cooke, "Chain Gang" (1960)

Overview

Lots of small collections got you down? Tired of paying O(n) to concatenate lists, or generating a lot of garbage prepending to a vector? If so, Chain is for you!

Chain is a small library that supports efficient concatenation across many collection types, as well as efficient iteration across the results.

import chain.Chain

val xs: Vector[Long] = ...
val ys: Vector[Long] = ...

// copies the entire contents of xs and ys
// before performing the summation
(xs ++ ys).foldLeft(0L)(_ + _)

// does not copy anything, just iterates over
// xs and ys in turn.
(Chain(xs) ++ Chain(ys)).foldLeft(0L)(_ + _)

This example is somewhat contrived, but I bet you have lots of code that builds intermediate collections solely to iterate over them. Chain can help make that code more efficient.

Quick Start

Chain supports Scala 2.10, 2.11, and 2.12.

To include Chain in your projects, you can use the following build.sbt snippet:

libraryDependencies += "org.spire-math" %% "chain" % "0.3.1"

Chain also supports Scala.js. To use Chain in your Scala.js projects, include the following build.sbt snippet:

libraryDependencies += "org.spire-math" %%% "chain" % "0.3.1"

Details

Chain can wrap any Iterable[A] values, and supports concatenation between mixed collection types. Here's an example that shows off a number of Chain's capabilities:

import chain.Chain

val ws: Iterable[Int] = List(1,2,3)
val xs: List[Int] = List(4,5,6)
val ys: Vector[Int] = Vector(7,8,9,10,11)
val zs: Option[Int] = Some(12)

val a = Chain(ws) ++ Chain(xs) // no copying
val b = Chain.all(ys, zs)      // same as ys ++ zs, but no copying
val c = a ++ b                 // still no copying
val d = 9 +: c :+ 100          // supports prepend/append

val ns: Iterable[Int] = c.toIterable // thin wrapper for scala compat

c.toVector           // Vector(1,2,3,4,5,6,7,8,9,10,11,12)
c.iterator.toList    // List(1,2,3,4,5,6,7,8,9,10,11,12)
c.foreach(println)   // prints 1-12
c.find(_ > 6)        // Some(7)
c.forall(_ >= 0)     // true
c.exists(_ > 100)    // false
c.map(_ * 2)         // Chain(2,4,6,8,10,12,14,16,18,20,22,24)
c.filter(_ % 3 == 0) // Chain(3,6,9,12)

(Note that .toString evaluates the entire contents of the Chain, so displaying a chain value in the REPL will force iteration over the contents of the chain.)

Chain is sealed and consists of two concrete case classes:

  • Chain.Elems wraps a single collection.
  • Chain.Concat represents a single ++ invocation.

Together these types create a tree. (Since we do not need to support arbitrary insertion into the tree, there is no need to balance it.) Iteration over the tree takes advantage of an in-memory stack to efficiently walk the contents in O(n) time.

Concatenating chains is always O(1), and iteration is always O(n).

Empty Chains can be obtained by Chain.empty[A] and are represented as a singleton Chain.Empty which is a Chain(Nil). This value is immutable and can be shared safely. Chains with a single element are constructed by Chain.single(x) which constructs Chain(x :: Nil) instances. This is done transparently in the case of +: and :+. These encoding are relatively efficient although if you are working entirely with single elements a more efficient data structure is possible.

Some operations that transform the Chain will need to allocate a new collection (either directly, or wrapped in a new Chain[A]). The comments explain the exact performance characteristics of each method, but here is a quick list of the methods which will allocate a new collection:

  • .map: always allocates a new collection.
  • .flatMap: always allocates a new collection.
  • .filter: always allocates a new collection.
  • .compress: when not already compressed, allocates a new collection.
  • .toVector: usually allocates a new Vector[A].

(If your chain is a Chain.Elems wrapping a Vector[A], then .toVector will just return that vector directly.)

Caveats

To avoid inheriting inefficient methods (such as .size), Chain itself does not extend any Scala collection interfaces. However .toIterable uses a very thin wrapper to support Iterable[A], so if you need to interoperate with the Scala collections hierarchy you can use that method.

Currently Chain supports the most commonly-used collection methods. Most of the rest should be easy to implement using .iterator, .foldLeft, or .find. Pull requests to add more of these methods will be gladly accepted.

The design of Chain assumes that the (relatively small) overhead of using Iterator[A] internally is acceptable. In the case of a large number of very small (or empty) collections this could be less efficient than simply accumulating those values into a single collection. The .compress method can be used in these situations.

Chain can be thought of as a limited kind of rope that is specialized to Scala collections (specifically Iterable[A]). You can imagine a similar (but more principled) data structure that is based around a type class like Foldable instead.

In general calling .iterator should be relatively low cost. In cases where the chain is right-associated (e.g. Chain(xs) ++ (...)), almost no work will take place. In cases where a chain is deeply left-associated, the call will need to descend until it finds the leftmost concrete collection that is not empty.

Future Work

Additional benchmarking and profiling would be great. Almost any chain method implemented with .iterator could be specialized if it proved to be a hotspot.

It might be nice to have a few different types to support various expected work loads and collection shapes. The current approach leans towards supporting large collections.

As mentioned above, it would be great to have a story for using type classes instead of Iterable[A] (either via an abstraction or a new type). It could also be nice to have a version which supported lazy filtering/mapping (although in many cases this can be emulated with things like .iterator.filter).

Copyright and License

All code is available to you under the MIT license, available at http://opensource.org/licenses/mit-license.php.

Copyright Erik Osheim, 2016-2018.