synchronous-promise

synchronous-promise

TL;DR: A prototypical animal which looks like an A+ Promise but doesn't defer immediately, so can run synchronously, for testing. Technically, this makes it not A+ compliant, since part of the A+ spec is that resolution be asynchronous.

This means that I unfortunately can't run the official tests at https://github.com/promises-aplus/promises-tests. As such, I rely on issue reports from users and welcome contributions.

Why?

The standard ES6 Promise (and any others which are A+ compliant) push the promise logic to the background immediately, departing from the mechanisms employed in years past by promise implementations in libraries such as jQuery and Q.

This is a good thing -- for production code. But it can make testing more convoluted than it really needs to be.

Often, in a test, we're stubbing out a function which would return a promise (eg http call, show a modal dialog requiring user interaction) with a promise that resolves immediately, eg (using, mocha/sinon/chai):

describe('the thing', () => {
  it('will do some stuff', () => {
    // Arrange
    const asyncLibraryFake = {
      someMethod: sinon.stub().returns(Promise.resolve('happy value!'))
    },
    sut = createSystemUnderTestWith(asyncLibraryFake);
    // Act
    sut.doSomethingInteresting();
    // Assert
    //  [*]
  })
});

[*] Ideally, we'd just have assertions here, but the code above has backgrounded, so we're not going to get our expected results unless we employ async testing strategies ourselves.

One strategy would be to return the promise from asyncLibraryFake.someMethod from the doSomethingInteresting function and perform our asserts in there:

describe('the thing', () => {
  it('will do some stuff', done => {
    // Arrange
    const asyncLibraryFake = {
      someMethod: sinon.stub().returns(Promise.resolve('happy value!'))
    },
    sut = createSystemUnderTestWith(asyncLibraryFake);
    // Act
    sut.doSomethingInteresting().then(() => {
      // Assert
      done()
    });
  })
});

And there's nothing wrong with this strategy.

I need to put that out there before anyone takes offense or thinks that I'm suggesting that they're "doing it wrong". If you're doing this (or something very similar), great; async/await, if available, can make this code quite clean and linear too.

However, when we're working on more complex interactions, eg when we're not testing the final result of a promise chain, but rather testing a side-effect at some step during that promise chain, this can become more effort to test (and, imo, make your testing more unclear).

Many moons ago, using, for example, Q, we could create a deferred object with Q.defer() and then resolve or reject ith with deferred.resolve() and deferred.reject(). Since there was no initial backgrounding, we could set up a test with an unresolved promise, make some pre-assertions, then resolve and make assertions about "after resolution" state, without making our tests async at all. It made testing a little easier (imo) and the idea has been propagated into frameworks like angular-mocks

Usage

SynchronousPromise looks (from the outside) a lot like an ES6 promise. We construct the same:

var promise = new SynchronousPromise((resolve, reject) => {
  if (Math.random() < 0.1) {
    reject('unlucky!');
  } else {
    resolve('lucky!');
  }
});

They can, of course, be chained:

var initial = new SynchronousPromise((resolve, reject) => {
  resolve('happy!');
});
initial.then(message => {
  console.log(message);
})

And have error handling, either from the basic A+ spec:

initial.then(message => {
  console.log(message);
}, error => {
  console.error(error);
});

Or using the more familiar catch():

initial.then(message => {
  console.log(message);
}).catch(error => {
  console.error(error);
})

.catch() starts a new promise chain, so you can pick up with new logic if you want to. .then() can deal with returning raw values or promises (as per A+)

Statics

.all(), .resolve() and .reject() are available on the SynchronousPromise object itself:

SynchronousPromise.all([p1, p2]).then(results => {
  // results is an array of results from all promises
}).catch(err => {
  // err is any single error thrown by a promise in the array
});

SynchronousPromise.resolve('foo');  // creates an already-resolved promise

SynchronousPromise.reject('bar'); // creats an already-rejected promise

(race() isn't because I haven't determined a good strategy for that yet, considering the synchronous design goal -- but it's unlikely you'll need race() from a test).

Extras

SynchronousPromise also provides two extra functions to make testing a little easier:

Static methods

The unresolved() method returns a new, unresolved SynchronousPromise with the constructor-function-provided resolve and reject functions attached as properties. Use this when you have no intention of resolving or rejecting the promise or when you want to resolve or reject at some later date.

var
  resolvedValue,
  rejectedValue,
  promise = SynchronousPromise.unresolved().then(function(data) {
    resolved = data;
  }).catch(function(data) {
    rejected = data;
  });
  // at this point, resolved and rejected are both undefined

  // ... some time later ...
  if (Math.random() > 0.5) {
    promise.resolve("yay");
    // now resolvedValue is "yay" and rejectedValue is still undefined
  } else {
    promise.reject("boo");
    // now rejectedValue is "boo" and resolvedValue is still undefined
  }

Instance methods

pause() pauses the promise chain at the point at which it is called:

SynchronousPromise.resolve('abc').then(data => {
  // this will be run
  return '123';
}).pause().then(data2 => {
  // we don't get here without resuming
});

and resume() resumes operations:

var
  promise = SynchronousPromise.resolve('123').pause(),
  captured = null;
promise.then(data => {
  captured = data;
});

expect(data).to.be.null;   // because we paused...
promise.resume();
expect(data).to.equal('123'); // because we resumed...

You can use pause() and resume() to test the state of your system under test at defined points in a series of promise chains

ES5

SynchronousPromise is purposefully written with prototypical, ES5 syntax so you can use it from ES5 if you like. Use the synchronous-promise.js file from the dist folder if you'd like to include it in a browser environment (eg karma).

Typescript

The synchronous-promise package includes an index.d.ts. To install, run:

typings install npm:synchronous-promise --save

On any modern TypeScript (v2+), you shouldn't need to do this.

Also note that TypeScript does async/await cleverly, treating all promises equally, such that await will work just fine against a SynchronousPromise -- it just won't be backgrounded.

HOWEVER: there is a very specific way that SynchronousPromise can interfere with TypeScript: if

• SynchronousPromise is installed globally (ie, overriding the native Promise implementation) and
• You create a SynchronousPromise which is resolved asynchronously, eg:

global.Promise = SynchronousPromise;
await new SynchronousPromise((resolve, reject) => {
  setTimeout(() => resolve(), 0);
}); // this will hang

This is due to how TypeScript generates the __awaiter function which is yielded to provide async/await functionality, in particular that the emitted code assumes that the global Promise will always be asynchronous, which is normally a reasonable assumption.

Installing SynchronousPromise globally may be a useful testing tactic, which I've used in the past, but I've seen this exact issue crop up in production code. SynchronousPromise therefor also provides two methods:

• installGlobally
• uninstallGlobally

which can be used if your testing would be suited to having Promise globally overridden by SynchronousPromise. This needs to get the locally-available __awaiter and the result (enclosed with a reference to the real Promise) must override that __awaiter, eg:

declare var __awaiter: Function;
beforeEach(() => {
  __awaiter = SynchronousPromise.installGlobally(__awaiter);
});
afterEach(() => {
  SynchronousPromise.uninstallGlobally();
});

It's not elegant that client code needs to know about the transpiled code, but this works.

I have an issue open on GitHub https://github.com/Microsoft/TypeScript/issues/19909 but discussion so far has not beein particularly convincing that TypeScript emission will be altered to (imo) a more robust implementation which wraps the emitted __awaiter in a closure.

Production code

The main aim of SynchronousPromise is to facilitate easier testing. That being said, it appears to conform to expected Promise behaviour, barring the always-backgrounded behaviour. One might be tempted to just use it everywhere.

However: I'd highly recommend using any of the more venerable promise implementations instead of SynchronousPromise in your production code -- preferably the vanilla ES6 Promise, where possible (or the shim, where you're in ES5). Or Q. Or jQUery.Deferred(), Bluebird or any of the implementations at https://promisesaplus.com/implementations.

Basically, this seems to work quite well for testing and I've tried to implement every behaviour I'd expect from a promise -- but I'm pretty sure that a native Promise will be better for production code any day.