iter-tools
provides a comprehensive suite of utility methods for working with javascript iterables and async iterables. Iterables offer an abstraction for describing sequences of values. If you're not sure if iter-tools
is the right library for you, check out our features.
// If your environment supports es2018:
import { filter } from 'iter-tools-es';
const { filter } = require('iter-tools-es');
// Otherwise you can use es5:
import { filter } from 'iter-tools';
const { filter } = require('iter-tools');
es5 consumers are responsible for loading core-js
if their environment does not define Symbol.iterator
.
Please read our API docs!
Historical docs are markdown files on github. For 6.x docs look at tags on the history of README.md. For 7.x versions look at tags on the history of API.md.
Some information is present also in the github wiki, including a cookbook providing author and user-submitted examples of ways to use the library to accomplish common tasks.
For more high level information, read on!
Iterables will never replace arrays, but they have several advantages which make them the future for high-level work in Javascript.
-
Defining an API in terms of iterables frees you from needing to worry about how data is stored. If you only expect iterables your code will not break when the data you expect changes from being an
Array
to being aSet
or a computed representation of values stored elsewhere (or not stored at all). When only the iterable contract is used you will also be freed from other worries like a callee unexpectedly mutating a caller's data. -
Iterables are lazy, which means they do work only when the results of that work are needed. Sometimes this can save you having to do any work! This also allows iterables to definite infinite sequences which are not possible to store in an array.
range()
for example returns the sequence of all positive integers. -
The use of small, short-lived objects with predictable shapes (like the
{ value, done }
object returned byiterator.next()
) is preferable in situations where responsiveness is a bigger concern than raw speed, most notably web applications (and especially inside of tight loops such as mouse event handlers). This is because generational garbage collectors are optimized for such objects, and can free their memory in imperceptably-fast minor GCs.
All javascript data types designed for the storage of data (Array, Map, and Set) are iterable, and any object or class can be made iterable by implementing a Symbol.iterator method which returns an iterator. For more reading on what is iterable and how to define iterables, see the docs for isIterable.
Iterables are usually consumed with the for..of loop syntax or the destructuring assignment syntax (e.g. const [a, b] = iterable
). They can also be consumed by iter-tools
methods.
Some iterables are also iterators, earning them the name IterableIterator
. Any iterables returned by iter-tools
methods (e.g. filter) are iterable iterators because they are implemented using generator functions. This means they can be used in one of two ways:
import { filter, notUndefined } from 'iter-tools-es';
// as an iterable:
for (const value of filter(notUndefined, iterable)) {
/* ... */
}
// or as an iterator:
class SparseArray extends Array {
[Symbol.iterator]() {
return filter(notUndefined, this);
}
}
iter-tools
is one of a handful of libraries that offer some common functionality like map
and filter
methods. It offers these features in addition to the common functionality:
- It can eliminate a whole class of null pointer errors by treating
null
andundefined
as empty iterables. - Any
Iterable
returned is anIterableIterator
. - Methods support currying, making them ideal for usage with provided methods like pipe or compose, or in a variety of other situations.
- It provides a low-level
__
API for when speed is paramount. - It provides a full API over both sync and async iterables. These are implemented separately (for performance), but parity is guaranteed through the use of a custom build.
- It has type definitions checked in for Typescript (
>3.8.2
). These are validated on every PR. - It is semver compliant, keeps a changelog, and has no runtime dependencies.
iter-tools
supports tree-shaking, and is a commonjs/esmodule dual package, meaning it can be loaded usingrequire()
as well asimport
.- The library's sources are valid es module code. Node versions
>12.17.0
can execute them and web build systems that transpile code innode_modules
can consume them as well. This makes it possible to use a github branch as a package, e.g. by putting"dependencies": { "iter-tools-es": "gh-username/iter-tools" }
in yourpackage.json
. Correctly configured bundlers can use abrowserslist
to avoid shipping code that is slower, more cryptic, and harder to debug than it needs to be.
It is possible to import individual methods from the library. This will make the resulting program or application faster to initialize as it avoids the need to parse the whole library (including parts of it which may not be used). Note that this will not be nececssary for code which will be bundled prior to shipping, as modern versions of rollup and webpack will remove the unused code. Importing a single method looks like this:
const notUndefined = require('iter-tools-es/methods/not-undefined');
If you happen to be transpiling the code and have use of the fantastic babel-plugin-macros
, you can generate single-method imports effortlessly like so:
import { filter, map } from 'iter-tools-es/explode.macro';
// which transpiles to:
import filter from 'iter-tools-es/methods/filter';
import map from 'iter-tools-es/methods/map';
Methods whose names begin with __
(double underscore) are safe to use – they are part of the library's public API. They are intended for use by extenders of the library and in tight loops where performance is critical.
There are some differences in the order in which arguments are passed, but these are documented along with the method's other overloads. Other less visible differences are:
__
methods are not curried. You must pass all their arguments in a single call
__
methods (aside from __wrap
) do not treat null
and undefined
as iterables.
__
methods do not permit sync iterables to be used in place of async iterables (this could change in the future).
__
methods return singleton iterable iterators. You can only loop over the results once.
__
methods may expose implementation internals which are not part of the documented public API. Code using undocumented APIs is subject to breakage in any release, in accordance with the semver specification.
__
methods do not have type definitions.
Some major improvements are still to come. They are:
- UMD support
@iter-tools/regex
for evaluating regular expressions against iterables@iter-tools/unicode
for turning strings into iterables of graphemes or extended grapheme clusters- Flow types
Some methods in iter-tools consume an entire iterable, such as arrayFrom
, last
, or cycle
. These methods will not terminate if you pass them an infinite iterable such as range()
. Eventually we may have a better system for warning you in the circumstances when we know you've done something obviously wrong like cycle(range())
, but this does not exist yet.
iter-tools
does not aim to provide all possible operations. If a method only makes sense when its input or output is an array, iter-tools
probably does not have an implementation. groupBy
is probably the most common example of this. I may in the future create an @iter-tools/array
package for such methods if there is sufficient demand.
I give a lot of credit to the great itertools Python library. This package doesn't want to be a mere port, but rather aims to solve the same problems as Python's library.
Many thanks to Maurizio Lupo (sithmel) for doing the initial porting and development from Python's itertools, for being accepting of new contributors (and patient with us), and for eventually passing maintainership along.