Pull-streams based on generators
Switch branches/tags
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.



Generator pull streams

This is a possible implementation of how pull-streams would work with generators (arrived with Node 6).

Value flows through the stream as they are being pulled out from the end.


npm install generator-pull-stream

How to use

Here's a basic example. A "puts" values as they get back from the api, and B "takes" them, one by one, and does its thing to them before passing them on. tap (which is documented later) basically will pull out as many values as it can and console.logs them. Once A's id reaches 11 it returns, and the entire stream ends.

const {pipe, pull} = require('generator-pull-streams')
const {tap} = require('generator-pull-streams/util')

const stream = pipe(
  function * A ({cps, put}) {
    let id = 0
    while (true) {
      const data = yield cps(api, `/some/url/${id}`)
      yield put(data.thing)
      if (id > 10) return
  function * B ({take, call}) {
    while (true) {
      const thing = yield take()
      const otherThing = yield call(toOtherThing, thing)
      yield put(otherThing)


Pull will start pulling values through. If any of them ends then the whole stream stops.

Above you see four different helper functions being used: put, take, call, and cps. There are tow more: resolve, and wait.

  • yield put(value) is used to pass a value on to the next generator in the pipe (in this case, from A to B).
    • You can also do yield value as a shorthand.
  • yield take([value]) is used to fetch the next value. Notice that the very first stream can not yield a take since it has nowhere to take a value from.
    • Shorthand is to just do yield without anything.
  • yield call(fn, arg1, arg2, ...) is to call functions. Functions here are expected to be synchronous.
  • yield cps(fn, arg1, arg2, ...) is to call a node style "error first" type function that takes a yield callback as its last parameter.
    • To catch errors wrap the yield in a try/catch.
  • yield resolve(fn, arg1, arg2, ...) expects a promise to call and resolve for you.
  • yield wait(2000) pauses for two seconds, then resumes.

More information

  • Values flow from left to right (from A to B to tap in the above example).
  • If any sub-stream returns, the entire stream stops.
  • If any sub-stream throws, the entire stream stops.
  • pipe takes any number of streams, not just two.
  • The first argument to pipe may be any type of iterator. Including arrays and strings.
  • The streams coming out of pipe should compose quite well. So you can create a "sub stream" and pass it around to be included elsewhere.
  • You can communicate with a source when a stream takes a value, by passing it into the take: const data = yield take('give me something great!').
  • Streams will not start running until some calls pull on them. As such, they are pure (calling pipe with the same arguments will always give you the same value back. No side effects or randomness).

Some words

There is nothing wrong with pull-streams. Theoretically it just seemed to be something generators would be suited for; so as part of really digging into pull-streams, I decided to implement them with generators. While they aren't as mathematically beautiful, I have to say that I like the interface of using generators much more.

I should mention the amazing redux-saga library as inspiration for the interface and async model.


I've included some helper functions in /util.


Will call fn with every value coming through. fn defaults to console.log if nothing is passed in.

filter(fn, options)

const {pipe, pull} = require('generator-pull-stream')
const {filter, tap} = require('generator-pull-stream/util')

const source = [1, 2, 3, 4, 5, 6]
const filterEven = filter(x => x % 2 === 0)
const log = tap()
const stream = pipe(source, filterEven, log)
pull(stream) // logs 2, 4, 6

Pass in {cps: true} as options if the filter function is a node callback style function. Pass in {promise: true} if the function returns a promise.

reject(fn, options)

The reverse of filter above.

find(fn, options)

Same as find, except it cancels the stream once the first matching item has been found.


By default this will flatten a stream one level (it assumes arrays coming in). If you pass {deep: true} if you want it to flatten every array that it encounters.

map(fn, options)

Maps your function over every value in the stream. Pass {cps: true} if the function is a node callback style function, or {promise: true} if it returns a promise.


Takes the first n values, then cancels the stream.

until(fn, options)

Takes values until the function passes. For instance

const source = [1, 2, 3, 4, 5]
const gt3 = x => x > 3
const stream = pipe(source, until(gt3), tap())
pull(stream) // logs 1, 2, 3

Pass in {last: true} if you want the last value also (above that would be 4).

Pass {cps: true} if the function is a node callback style function, or {promise: true} if it returns a promise.

scan(fn, options)

Much like reduce, except it gives you every intermediary value and not just the end result.

const source = [1, 2, 3, 4]
const add = (a, b) => a + b
const stream = pipe(source, scan(add), tap())
pull(stream) // logs 1, 3, 6, 10

Pass {cps: true} if the function is a node callback style function, or {promise: true} if it returns a promise.


Makes sure value do not pass through faster than the ms you passed in.

const source = [1, 2, 3]
const stream = pipe(source, throttle(500), tap())
pull(stream) // logs 1, 2, 3 with 500ms interval

unique(fn, options)

Gives you only unique values. The fn is for camparing on something specific, for instance you can pass in x => x.id to compare on the ids.

Pass {cps: true} if the function is a node callback style function, or {promise: true} if it returns a promise.

notUnique(fn, options)

Reverse of unique.


Takes any (I hope) pull-stream through and converts into a generator pull stream.

const flatten = require('pull-stream/throughs/flatten')
const {convertPullStream, tap} = require('generator-pull-stream/util')
const {pipe, pull} = require('generator-pull-stream')

const source = [[1, 2, 3], [4, 5, 6]]
const generatorFlatten = convertPullStream(flatten())

const stream = pipe(source, generatorFlatten, tap())
pull(stream) // logs 1, 2, 3, 4, 5, 6,

This function should be regarded as semi experimental. Any help with testing and updating it would be helpful.

Unit testing your streams

Testing your streams is really easy. Let's have a look at A from above and see how we might go about testing it.

Notice that we are effectively testing that we are calling the api with the correct url, without actually calling it.

function * A ({cps, put}) {
  let id = 0
  while (true) {
    const data = yield cps(api, `/some/url/${id}`)
    yield put(data.thing)

// in our test file
const tape = require('tape')
const api = require('./api')
const {pull} = require('generator-pull-stream')

tape('stream A', t => {
  const a = A(pull.effects)

  const actual1 = a.next().value
  const expected1 = pull.effects.cps(api, '/some/url/0')
  t.deepEqual(actual1, expected1)

  const actual2 = a.next({thing: 'foo'}).value
  const expected2 = pull.effects.put('foo')
  t.deepEqual(actual2, expected2)

  const actual3 = a.next().value
  const expected3 = pull.effects.cps(api, '/some/url/1')
  t.deepEqual(actual3, expected3)


The full stream from above can be tested in the same manner.