Source Code Explanation of Co for Generator Based Flow-Control

08 November, 2016

Since the release of ES6(ES2015), which introduced a new feature called Generator, which is used for async control flow to prevent the callback hell. Generator allows us to do async tasks with sync coding, which makes our code more elegant and straight forward. Generator is like a normal function, after you call the function, it returns an Iterator, which then you can call next method to control the code running all by yourself. Even though we have the full control with generator, which is really cool, but sometimes it is also a little bit overhead, when you have lots of generators nested together, which makes you stuck in the nested hell:(. Fortunately we already had a tool called Co to help us run the whole generator(s) flow automatically out of the box. The source code of Co is simple and elegant with only about 200 lines of code, below we will go through the main function code line by line to show you how the small tool makes your life easier.


Code is based on on Co@4.6.0. At the end of the article, I will show a demo which uses all the yieldables that co supports

var slice = Array.prototype.slice;

This line is simple, which is used to convert the arguments to array and pass the needed parameters to fn.apply call.

co.wrap = function (fn) {
  createPromise.__generatorFunction__ = fn;
  return createPromise;
  function createPromise() {
    //just pass the returned Iterator(Generator) to co call
    return, fn.apply(this, arguments));

The comments in the source code shows that it wraps the generator function into a function to prevent unnecessary closure with multiple co() calls, the wrap fun is also used internally in Koa, Inside the createPromise fn, it calls the co fn while also passing the consistent this, and the called generator function.

In the main function co(gen){} fun:

var ctx = this;
var args =, 1);

First reference this to ctx, and extract the arguments except the first generator parameter, and pass them to gen.apply() . Co wraps everything into a Promise, and will return the Promise.

return new Promise(function(resolve, reject) {
    if (typeof gen === 'function') gen = gen.apply(ctx, args);
    if (!gen || typeof !== 'function') return resolve(gen);


Inside the promise executor, it checks if gen is a function type, if yes, just run the function, and set the result to gen again, which will be a generator iterator if it was a GeneratorFunction, and then we can call next to start the code execution. And if the result is not a Iterator, just resolve the promise with the result of gen itself. Run onFulfilled() to start the recursive execution of the nested yieldables.

function onFulfilled(res) {
      var ret;
      try {
        ret =;
      } catch (e) {
        return reject(e);
      return null;

With to pass res from the previous Promise result to previously paused yield expression result, and also get the next yield expression result to set to ret, if exception occurs, reject the current promise. Then call next(ret) to check the yielded result and create Promise based on the result object type to continue the next nested recursive flow. onRejected function is almost the same as above.
The last function inside the main co is next(), which is called as many times as needed to go through the whole execution flow, until it reaches the end.

function next(ret) {
      if (ret.done) return resolve(ret.value);
      var value =, ret.value);
      if (value && isPromise(value)) return value.then(onFulfilled, onRejected);
      return onRejected(new TypeError('You may only yield a function, promise, generator, array, or object, '
        + 'but the following object was passed: "' + String(ret.value) + '"'));

First it checks if the current ret generator iterator is done, if yes, return and resolve the promise with the final value of the generator. Otherwise, convert ret.value to another Promise, and add the resolve/reject handler for the promise, if it's not promise, just reject the original promise with the type not supported Error. Code above are the main logic of the Co library, and rest of the utility functions are used to convert different objects to Promises, object checking... All supported yieldables are:

  • thunk(function with a node-style callback, e.g function(err, res){}), may be deprecated soon.
  • Promise
  • Generator, GeneratorFunction
  • Array, concurrently run multi async tasks
  • Object, same as above, but with a key for every task


Let's go through a demo to have a better understanding of the co control flow:

//control flow test
function testCo() {
  co(function *(a){
    console.log('passed args:', a);
    let temp = 1;
    temp = yield Promise.resolve(++temp);
    temp = yield foo();
    console.log('foo done:', temp);
    return 'all done';
  }, 11).then(function(res){
    console.log('final result:', res);

function *foo() {
  const one = yield [
    new Promise(function (resolveFn, rejectFn){
        resolveFn('array timeout done!');
      }, 500);
  console.log('array: ', one);
  const two = yield bar();
  console.log('from bar:', two);
  return 'foo-done';

function *bar() {
  const obj = yield {
    objA: Promise.resolve('bar-1'),
    objB: new Promise(function (resolveFn, rejectFn) {
        resolveFn('obj timeout done!');
      }, 500);
  console.log('object:', obj);
  //yield a thunk
  const cc = yield function(cb){
    cb(null, 'thunk result');
  console.log('bar thunk:', cc);
  return 'bar done';

The result will be:

passed args: 11
array:  [ 'foo-1', 'array timeout done!' ]
object: { objA: 'bar-1', objB: 'obj timeout done!' }
bar thunk: thunk result
from bar: bar done
foo done: foo-done
final result: all done

You can go through the flow to understand the whole idea of Co.


Co is a nice solution for simpler generator control flow, we can yield any supported async tasks in our business logic, and leave everything else to Co to finish all the tasks with the respect of our sync and async sequence needs. Feel free to try it out, your life will become easier :).