- Notifications
You must be signed in to change notification settings - Fork10
A prototypical animal which looks like an A+ Promise but doesn't defer immediately, so can run synchronously, for testing
License
fluffynuts/synchronous-promise
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
TL;DR: A prototypical animal which looks like an A+ Promise but doesn't deferimmediately, so can run synchronously, for testing. Technically, this makes itnot A+ compliant, since part of the A+ spec is that resolution be asynchronous.
This means that I unfortunately can't run the official tests athttps://github.com/promises-aplus/promises-tests. As such, I rely on issue reports from users and welcome contributions.
Ifsynchronous-promise
has made something easier for you and you'd like to say thanks,check out my sponsors page.
The standard ES6 Promise (and any others whichare A+ compliant) push the promise logic to the backgroundimmediately, departing from the mechanisms employed in years past by promiseimplementations in libraries such as jQuery and Q.
This is a good thing -- for production code. But it can make testing moreconvoluted 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 promisethat resolves immediately, eg (using, mocha/sinon/chai):
describe('the thing',()=>{it('will do some stuff',()=>{// ArrangeconstasyncLibraryFake={someMethod:sinon.stub().returns(Promise.resolve('happy value!'))},sut=createSystemUnderTestWith(asyncLibraryFake);// Actsut.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 testingstrategies ourselves.
One strategy would be to return the promise fromasyncLibraryFake.someMethodfrom thedoSomethingInterestingfunction and perform our asserts in there:
describe('the thing',()=>{it('will do some stuff',done=>{// ArrangeconstasyncLibraryFake={someMethod:sinon.stub().returns(Promise.resolve('happy value!'))},sut=createSystemUnderTestWith(asyncLibraryFake);// Actsut.doSomethingInteresting().then(()=>{// Assertdone()});})});
And there's nothing wrong with this strategy.
I need to put that out there before anyone takes offense or thinks that I'm suggestingthat 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 nottesting the final result of a promise chain, but rather testing a side-effectat 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 withQ.defer()
and then resolve or reject ith withdeferred.resolve()
anddeferred.reject()
. Since there was no initial backgrounding, we could setup a test with an unresolved promise, make some pre-assertions, then resolveand make assertions about "after resolution" state, without making our testsasync at all. It made testing a little easier (imo) and the idea has beenpropagated into frameworks likeangular-mocks
SynchronousPromise looks (from the outside) a lot like an ES6 promise. We constructthe same:
varpromise=newSynchronousPromise((resolve,reject)=>{if(Math.random()<0.1){reject('unlucky!');}else{resolve('lucky!');}});
They can, of course, be chained:
varinitial=newSynchronousPromise((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 familiarcatch()
:
initial.then(message=>{console.log(message);}).catch(error=>{console.error(error);})
.catch()
starts a new promise chain, so you can pick up with new logicif you want to..then()
can deal with returning raw values or promises(as per A+)
SynchronousPromise
also supports.finally()
as of version 2.0.8.
.all()
,.resolve()
and.reject()
are available on theSynchronousPromise
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 promiseSynchronousPromise.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'sunlikely you'll needrace()
from a test).
SynchronousPromise
also provides two extra functions to make testing a littleeasier:
Theunresolved()
method returns a new, unresolvedSynchronousPromise
withthe constructor-function-providedresolve
andreject
functions attached as properties.Use this when you have no intention of resolving or rejecting the promise or when youwant to resolve or reject at some later date.
varresolvedValue,rejectedValue,promise=SynchronousPromise.unresolved().then(function(data){resolvedValue=data;}).catch(function(data){rejectedValue=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}
pause()
pauses the promise chain at the point at which it is called:
SynchronousPromise.resolve('abc').then(data=>{// this will be runreturn'123';}).pause().then(data2=>{// we don't get here without resuming});
andresume()
resumes operations:
varpromise=SynchronousPromise.resolve('123').pause(),captured=null;promise.then(data=>{captured=data;});expect(captured).to.be.null;// because we paused...promise.resume();expect(captured).to.equal('123');// because we resumed...
You can usepause()
andresume()
to test the state of your system undertest at defined points in a series of promise chains
SynchronousPromise is purposefully written with prototypical, ES5 syntax so youcan use it from ES5 if you like. Use thesynchronous-promise.js
file from thedist
folder if you'd like to include it in a browser environment (eg karma).
Thesynchronous-promise
package includes anindex.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 promisesequally, such thatawait
will work just fine against a SynchronousPromise -- it just won't be backgrounded.
HOWEVER: there is avery specific way that SynchronousPromisecan interfere with TypeScript: if
- SynchronousPromise is installed globally (ie, overriding thenative
Promise
implementation) and - You create a SynchronousPromise which is resolved asynchronously,eg:
global.Promise=SynchronousPromise;awaitnewSynchronousPromise((resolve,reject)=>{setTimeout(()=>resolve(),0);});// this will hang
This is due to how TypeScript generates the__awaiter
functionwhich isyielded
to provideasync
/await
functionality, inparticular that the emitted code assumes that the globalPromise
willalways 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 upin production code.SynchronousPromise
therefor also provides two methods:
installGlobally
uninstallGlobally
which can be used if your testing would be suited to havingPromise
globallyoverridden bySynchronousPromise
. This needs to get the locally-available__awaiter
and the result (enclosed with a reference to the realPromise
)must override that__awaiter
, eg:
declarevar__awaiter:Function;beforeEach(()=>{__awaiter=SynchronousPromise.installGlobally(__awaiter);});afterEach(()=>{SynchronousPromise.uninstallGlobally();});
It's not elegant that client code needs to know about the transpiledcode, but this works.
I have an issue open on GitHubmicrosoft/TypeScript#19909but discussion so far has not been particularly convincing thatTypeScript emission will be altered to (imo) a more robustimplementation which wraps the emitted__awaiter
in a closure.
The main aim of SynchronousPromise is to facilitate easier testing. That beingsaid, it appears to conform to expectedPromise
behaviour, barring thealways-backgrounded behaviour. One might be tempted to just use it everywhere.
However: I'd highly recommend usingany of the more venerable promise implementationsinstead of SynchronousPromise in your production code -- preferably the vanillaES6 Promise, where possible (or the shim, where you're in ES5). Or Q.Or jQuery.Deferred(), Bluebird or any of the implementations athttps://promisesaplus.com/implementations.
Basically, this seems to work quite well for testing andI've tried to implement every behaviour I'd expect from a promise -- but I'mpretty sure that a nativePromise
will be better for production code any day.
About
A prototypical animal which looks like an A+ Promise but doesn't defer immediately, so can run synchronously, for testing
Topics
Resources
License
Uh oh!
There was an error while loading.Please reload this page.
Stars
Watchers
Forks
Packages0
Uh oh!
There was an error while loading.Please reload this page.
Contributors9
Uh oh!
There was an error while loading.Please reload this page.