How do other node developers who use sinon mock out constructor calls within their unit tests? For example, suppose I have some function foo
function foo() {
var dependency = new Dependency(args);
// do stuff with dependency
}
exports.module.foo = foo;
and in a separate test file I have some test in which I want to verify what the Dependency constructor is called with (args), and I need to control what it returns
it('should call Dependency constructor with bar', function() {
var foo = require('myModule').foo
var DependencyMock; //code to make the mock
foo();
expect(DependencyMock.calledWith(bar)).to.equal(true);
});
The problem is that sinon can only mock functions attached to an object, so we have to attach the constructor to an object before it can be mocked.
What I've been doing is just making an object to attach the constructor to in the module making the constructor call, calling the constructor as a method of that object, then exporting the object to use it in tests:
var Dependency = require('path/to/dependency');
var namespace = {
Dependency: Dependency
}
function foo() {
var dependency = new namespace.Dependency(args);
// do stuff with dependency
}
exports.moduole.foo = foo;
exports.module.namespace = namespace;
testfile:
it('should call Dependency constructor with bar', function() {
var foo = require('myModule').foo;
var namespace = require('myModule').namespace;
var DependencyMock = sinon.mock(namespace, 'Dependency').returns(0);
foo();
expect(DependencyMock.calledWith(bar)).to.equal(true);
});
This works, but it feels really clunky to expose an object on my module just for the sake of testing it.
Any tips?
I think it's worth asking why you'd want to mock a constructor of a dependency instead of injecting that dependency?
Consider your example code:
// in "foo.js"
function foo() {
var dependency = new Dependency(args);
// do stuff with dependency
}
exports.module.foo = foo;
If Dependency is required for foo to work you can inject it as an argument of foo:
// in "foo.js"
function foo(dependency) {
// do stuff with dependency
}
exports.module.foo = foo;
// in "bar.js"
var foo = require('./foo.js')(new Dependency(args));
With this change it's now trivial to inject any Test Double in your tests (to find out more about JavaScript Test Doubles have a look at my article on the subject).
This approach makes the dependencies of your function/module explicit, but requires you to wire them up at some point (here: require('./foo.js')(new Dependency(args));).
If you didn't want to wire things up manually there's another approach you can take using rewire and replacing constructor with factory method:
// in "dependency.js"
module.exports= function(args) {
return new Dependency(args);
}
// in "foo.js"
var dependency = require('./dependency');
function foo() {
var dep = dependency(args);
// do stuff with dependency
}
exports.module.foo = foo;
and in your test:
var rewire = require("rewire"),
foo = rewire("../lib/foo.js");
it('should call dependency... ', function() {
foo.__set__("dependency", /* some spy */ );
foo();
});
Hope this helps!
Jan
I used a workaround for this:
// Keep a reference to your dependancy class.
const dependencyClass = Dependency;
let item = new Dependency();
// Replace class with a stub.(optionally return an item.
Dependency = sinon.stub(Dependency, 'constructor').returns(item);
// Call your code that you expect the class constructor to be called.
foo();
assert.isTrue(Dependency.calledWithNew());
assert.equal(1, Dependency.callCount);
// Restore class reference.
Dependency = dependencyClass;
Additionally, in the above case, an item is returned, so the user can have access to the dependency for further testing.
eg.
assert.equal(item.someValue, 10);
You can have other problems using this, eg defined properties will no longer be available for the class.
I agree with Jan Molek, use this in case you cannot change the code.
Not tried, but this could work: stub the constructor function of Dependency and let it return a mock.
var constructor = sinon.stub(Dependency.prototype, "constructor");
constructor.returns(sinon.mock(Dependency));
I encountered a similar problem with modules that expose functions rather than objects - this was my solution. Just putting this here as another way to go about tackling the issue but I must say solving the problem via better design, as Jan Molak suggested, seems like a better approach.
Related
I've been working with nodejs lately and still getting to grips with the module system, so apologies if this is an obvious question. I want code roughly like the below:
a.js (the main file run with node)
var ClassB = require("./b");
var ClassA = function() {
this.thing = new ClassB();
this.property = 5;
}
var a = new ClassA();
module.exports = a;
b.js
var a = require("./a");
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
util.log(a.property);
}
module.exports = ClassB;
My problem seems to be that I can't access the instance of ClassA from within an instance of ClassB.
Is there any correct / better way to structure modules to achieve what I want?
Is there a better way to share variables across modules?
Try to set properties on module.exports, instead of replacing it completely. E.g., module.exports.instance = new ClassA() in a.js, module.exports.ClassB = ClassB in b.js. When you make circular module dependencies, the requiring module will get a reference to an incomplete module.exports from the required module, which you can add other properties latter on, but when you set the entire module.exports, you actually create a new object which the requiring module has no way to access.
While node.js does allow circular require dependencies, as you've found it can be pretty messy and you're probably better off restructuring your code to not need it. Maybe create a third class that uses the other two to accomplish what you need.
[EDIT] it's not 2015 and most libraries (i.e. express) have made updates with better patterns so circular dependencies are no longer necessary. I recommend simply not using them.
I know I'm digging up an old answer here...
The issue here is that module.exports is defined after you require ClassB.
(which JohnnyHK's link shows)
Circular dependencies work great in Node, they're just defined synchronously.
When used properly, they actually solve a lot of common node issues (like accessing express.js app from other files)
Just make sure your necessary exports are defined before you require a file with a circular dependency.
This will break:
var ClassA = function(){};
var ClassB = require('classB'); //will require ClassA, which has no exports yet
module.exports = ClassA;
This will work:
var ClassA = module.exports = function(){};
var ClassB = require('classB');
I use this pattern all the time for accessing the express.js app in other files:
var express = require('express');
var app = module.exports = express();
// load in other dependencies, which can now require this file and use app
Sometimes it is really artificial to introduce a third class (as JohnnyHK advises), so in addition to Ianzz:
If you do want to replace the module.exports, for example if you're creating a class (like the b.js file in the above example), this is possible as well, just make sure that in the file that is starting the circular require, the 'module.exports = ...' statement happens before the require statement.
a.js (the main file run with node)
var ClassB = require("./b");
var ClassA = function() {
this.thing = new ClassB();
this.property = 5;
}
var a = new ClassA();
module.exports = a;
b.js
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
util.log(a.property);
}
module.exports = ClassB;
var a = require("./a"); // <------ this is the only necessary change
The solution is to 'forward declare' your exports object before requiring any other controller. So if you structure all your modules like this and you won't run into any issues like that:
// Module exports forward declaration:
module.exports = {
};
// Controllers:
var other_module = require('./other_module');
// Functions:
var foo = function () {
};
// Module exports injects:
module.exports.foo = foo;
What about lazy requiring only when you need to? So your b.js looks as follows
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
var a = require("./a"); //a.js has finished by now
util.log(a.property);
}
module.exports = ClassB;
Of course it is good practice to put all require statements on top of the file. But there are occasions, where I forgive myself for picking something out of an otherwise unrelated module. Call it a hack, but sometimes this is better than introducing a further dependency, or adding an extra module or adding new structures (EventEmitter, etc)
You can solve this easily: just export your data before you require anything else in modules where you use module.exports:
classA.js
class ClassA {
constructor(){
ClassB.someMethod();
ClassB.anotherMethod();
};
static someMethod () {
console.log( 'Class A Doing someMethod' );
};
static anotherMethod () {
console.log( 'Class A Doing anotherMethod' );
};
};
module.exports = ClassA;
var ClassB = require( "./classB.js" );
let classX = new ClassA();
classB.js
class ClassB {
constructor(){
ClassA.someMethod();
ClassA.anotherMethod();
};
static someMethod () {
console.log( 'Class B Doing someMethod' );
};
static anotherMethod () {
console.log( 'Class A Doing anotherMethod' );
};
};
module.exports = ClassB;
var ClassA = require( "./classA.js" );
let classX = new ClassB();
A solution which require minimal change is extending module.exports instead of overriding it.
a.js - app entry point and module which use method do from b.js*
_ = require('underscore'); //underscore provides extend() for shallow extend
b = require('./b'); //module `a` uses module `b`
_.extend(module.exports, {
do: function () {
console.log('doing a');
}
});
b.do();//call `b.do()` which in turn will circularly call `a.do()`
b.js - module which use method do from a.js
_ = require('underscore');
a = require('./a');
_.extend(module.exports, {
do: function(){
console.log('doing b');
a.do();//Call `b.do()` from `a.do()` when `a` just initalized
}
})
It will work and produce:
doing b
doing a
While this code will not work:
a.js
b = require('./b');
module.exports = {
do: function () {
console.log('doing a');
}
};
b.do();
b.js
a = require('./a');
module.exports = {
do: function () {
console.log('doing b');
}
};
a.do();
Output:
node a.js
b.js:7
a.do();
^
TypeError: a.do is not a function
The important thing is not to re-assign the module.exports object that you have been given, because that object may have already been given to other modules in the cycle! Just assign properties inside module.exports and other modules will see them appear.
So a simple solution is:
module.exports.firstMember = ___;
module.exports.secondMember = ___;
The only real downside is the need to repeat module.exports. many times.
Similar to lanzz and setec's answers, I have been using the following pattern, which feels more declarative:
module.exports = Object.assign(module.exports, {
firstMember: ___,
secondMember: ___,
});
The Object.assign() copies the members into the exports object that has already been given to other modules.
The = assignment is logically redundant, since it is just setting module.exports to itself, but I am using it because it helps my IDE (WebStorm) to recognise that firstMember is a property of this module, so "Go To -> Declaration" (Cmd-B) and other tooling will work from other files.
This pattern is not very pretty, so I only use it when a cyclic dependency issue needs to be resolved.
It is fairly well suited to the reveal pattern, because you can easily add and remove exports from the object, especially when using ES6's property shorthand.
Object.assign(module.exports, {
firstMember,
//secondMember,
});
the extremely simple solution is often:
usually you'd have the require at the top of the file ...
var script = require('./script')
function stuff() {
script.farfunction()
}
instead, just require it "in the function"
function stuff() {
var _script = require('./script')
_script.farfunction()
}
An other method I've seen people do is exporting at the first line and saving it as a local variable like this:
let self = module.exports = {};
const a = require('./a');
// Exporting the necessary functions
self.func = function() { ... }
I tend to use this method, do you know about any downsides of it?
TL;DR
Just use exports.someMember = someMember instead of module.exports = { // new object }.
Extended Answer
After reading lanzz's response I could finally figure it out what is happening here, so I'll give my two cents on the subject, extending his answer.
Let's see this example:
a.js
console.log("a starting");
console.log("a requires b");
const b = require("./b");
console.log("a gets b =", b);
function functionA() {
console.log("function a");
}
console.log("a done");
exports.functionA = functionA;
b.js
console.log("b starting");
console.log("b requires a");
const a = require("./a");
console.log("b gets a =", a);
function functionB() {
console.log("On b, a =", a)
}
console.log("b done");
exports.functionB = functionB;
main.js
const a = require("./a");
const b = require("./b");
b.functionB()
Output
a starting
a requires b
b starting
b requires a
b gets a = {}
b done
a gets b = { functionB: [Function: functionB] }
a done
On b, a = { functionA: [Function: functionA] }
Here we can see that at first b receives an empty object as a, and then once a is fully loaded, that reference is updated through exports.functionA = functionA. If you instead replace the entire module with another object, through module.exports, then b will lose the reference from a, since it will point out to the same empty object from the beginning, instead of pointing to the new one.
So if you export a like this: module.exports = { functionA: functionA }, then the output will be:
a starting
a requires b
b starting
b requires a
b gets a = {}
b done
a gets b = { functionB: [Function: functionB] }
a done
On b, a = {} // same empty object
Actually I ended up requiring my dependency with
var a = null;
process.nextTick(()=>a=require("./a")); //Circular reference!
not pretty, but it works. It is more understandable and honest than changing b.js (for example only augmenting modules.export), which otherwise is perfect as is.
Here is a quick workaround that I've found use full.
On file 'a.js'
let B;
class A{
constructor(){
process.nextTick(()=>{
B = require('./b')
})
}
}
module.exports = new A();
On the file 'b.js' write the following
let A;
class B{
constructor(){
process.nextTick(()=>{
A = require('./a')
})
}
}
module.exports = new B();
This way on the next iteration of the event loop classes will be defined correctly and those require statements will work as expected.
One way to avoid it is to don't require one file in other just pass it as an argument to a function what ever you need in an another file.
By this way circular dependency will never arise.
If you just can't eliminate circular dependencies (e.g useraccount <---> userlogin), there's one more option...
Its as simple as using setTimeout()
//useraccount.js
let UserLogin = {};
setTimeout(()=>UserLogin=require('./userlogin.js'), 10);
class UserAccount{
getLogin(){
return new UserLogin(this.email);
}
}
//userlogin.js
let UserAccount ={};
setTimeout(()=>UserAccount=require('./useraccount.js'), 15);
class UserLogin{
getUser(){
return new User(this.token);
}
}
My node app has a fairly large test suite that depends on ShouldJS module. The problem is that the library blocks assigning .should properties on objects, but my test suite needs to spin up a server that needs to be mockable (i.e. run in the same process as my test suite) and dynamically build ElasticSearch queries, that need to set .should property on Objects.
I tried to un-require ShouldJS using an approach described here. But that didn't help - see example script below. Is there a workaround or an alternative approach?
Example:
require('should');
var objBefore = {};
objBefore.should = 'should1'; // doesn't get assigned!
objBefore['should'] = 'should2'; // doesn't get assigned!
console.log('Before:', objBefore);
Object.keys(require.cache).forEach(function(path) {
if (path.indexOf('/node_modules/should/') === -1) return;
console.log('Un-requiring path', path);
delete require.cache[path];
});
var objAfter = {};
objAfter.should = 'should1'; // doesn't get assigned!
objAfter['should'] = 'should2'; // doesn't get assigned!
console.log('After:', objAfter);
// still has shouldJS stuff
console.log('objAfter.should:', objAfter.should);
which gives this:
Before: {}
Un-requiring path /my/path/node_modules/should/index.js
Un-requiring path /my/path/node_modules/should/lib/should.js
...
Un-requiring path /my/path/node_modules/should/lib/ext/contain.js
Un-requiring path /my/path/node_modules/should/lib/ext/promise.js
After: {}
objAfter.should: Assertion { obj: {}, anyOne: false, negate: false, params: { actual: {} } }
None of the .should properties are getting assigned.
There's a way to turn ShouldJS should getter on and off:
require('should');
(1).should.be.equal(1);
// turn off ShouldJS should getter
// and use should as a function
var should = require('should').noConflict();
should(0).be.equal(0);
// turn on ShouldJS should getter
should.extend();
require('should');
(1).should.be.equal(1);
Here's what the official Readme says:
It is also possible to use should.js without getter (it will not even try to extend Object.prototype), just require('should/as-function'). Or if you already use version that auto add getter, you can call .noConflict function.
Results of (something).should getter and should(something) in most situations are the same
I am currently testing a module in isolation using proxquire to overwrite a require of this module.
Overwriting a path of a require works fine with proxyquire. For example:
var bar = require('./bar');
But can you use proxyquire also to overwrite just a specific function of a module which is required in the module to test? So something like:
var bar = require('./foo').bar();
I need to stay at proxyquire for this since I am using it for mocking a http-request happening in another layer of the architecture. But in case of the test I need to mock the time for "now" in the module as well.
So currently I have this:
var uraStub = sendMockRequest(paramListOfCheckin, queryList);
var setNowStub = function(){ return 1425998221000; };
var checkin = proxyquire('../src/logic/logicHandlerModules/checkin', {
'../../persistence/ura' : uraStub,
'./checkin.setNow' : setNowStub
});
checkin.checkin(...)
The implementation of setNow is:
var setNow = function(){
return new Date().getTime();
};
var checkin = function (...) {
var now = require('./checkin').setNow();
Obviousley './checkin.setNow' : setNowStub in proxyquire doesn't work, since this is the wrong path. But using './checkin'.setNow() : setNowStub also doesn't work because of wrong syntaxis in the object-definition.
Any suggestions?
Thanks in advance!
What you are looking for is the noCallThru() and callThru() methods. https://github.com/thlorenz/proxyquire#preventing-call-thru-to-original-dependency
By default proxyRequire will call through to the mocked dependency which will allow you to pick the methods that you want to overwrite with your own custom function.
So if a dependency in the path '../foo' has a method bar() and fooBar() you would be able to mock out just bar by doing.
proxyquire.callThru();
var fooFunc = proxyquire('../foo', {
bar: () => return 'bar'
})
Now bar() will hit your custom overwritten funtion while fooBar() will be called as normal.
I'm trying to spy on dependencies in a Node.js module from within a Jasmine spec, and I'm encountering some weird behaviour.
Jasmine spies on a method that resides in an object exposed by the exports object in the module. When I call the spied-on method in the module using
exports.umap()
the spy detects it. But when I call it in the module using a variable name I assigned to the same method
var umap = exports.umap
umap()
the spy doesn't detect it.
This seems weird to me given that var umap refers to the same method as exports.umap. Anyone have any ideas why this is? Code below.
(The whole reason I'm doing this is that I want to stub my module's dependencies without the need to design the module as a class and use dependency injection.)
// mapper.js
exports.umap = require('underscore').map;
var umap = exports.umap;
function map () {
return exports.umap([1,2,3], function(el) {console.log(el);return el * 2});
// umap() goes unnoticed by the spy
// return umap([1,2,3], function(el) {console.log(el);return el * 2});
}
exports.map = map;
.
// mapper_spec.js
var mapper = require ('../../lib/mapper');
var map = mapper.map;
describe('mapper', function() {
describe('map', function() {
it('calls umap', function() {
var mapped;
spyOn(mapper, 'umap').and.callThrough();
mapped = mapper.map();
expect(mapper.umap).toHaveBeenCalled();
expect(mapped).toEqual([2,4,6]);
});
});
});
UPDATE
I'm still curious about this issue, however https://www.npmjs.org/package/proxyquire made my whole "spying-on-things-attached-to-export-object" strategy moot.
I've been working with nodejs lately and still getting to grips with the module system, so apologies if this is an obvious question. I want code roughly like the below:
a.js (the main file run with node)
var ClassB = require("./b");
var ClassA = function() {
this.thing = new ClassB();
this.property = 5;
}
var a = new ClassA();
module.exports = a;
b.js
var a = require("./a");
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
util.log(a.property);
}
module.exports = ClassB;
My problem seems to be that I can't access the instance of ClassA from within an instance of ClassB.
Is there any correct / better way to structure modules to achieve what I want?
Is there a better way to share variables across modules?
Try to set properties on module.exports, instead of replacing it completely. E.g., module.exports.instance = new ClassA() in a.js, module.exports.ClassB = ClassB in b.js. When you make circular module dependencies, the requiring module will get a reference to an incomplete module.exports from the required module, which you can add other properties latter on, but when you set the entire module.exports, you actually create a new object which the requiring module has no way to access.
While node.js does allow circular require dependencies, as you've found it can be pretty messy and you're probably better off restructuring your code to not need it. Maybe create a third class that uses the other two to accomplish what you need.
[EDIT] it's not 2015 and most libraries (i.e. express) have made updates with better patterns so circular dependencies are no longer necessary. I recommend simply not using them.
I know I'm digging up an old answer here...
The issue here is that module.exports is defined after you require ClassB.
(which JohnnyHK's link shows)
Circular dependencies work great in Node, they're just defined synchronously.
When used properly, they actually solve a lot of common node issues (like accessing express.js app from other files)
Just make sure your necessary exports are defined before you require a file with a circular dependency.
This will break:
var ClassA = function(){};
var ClassB = require('classB'); //will require ClassA, which has no exports yet
module.exports = ClassA;
This will work:
var ClassA = module.exports = function(){};
var ClassB = require('classB');
I use this pattern all the time for accessing the express.js app in other files:
var express = require('express');
var app = module.exports = express();
// load in other dependencies, which can now require this file and use app
Sometimes it is really artificial to introduce a third class (as JohnnyHK advises), so in addition to Ianzz:
If you do want to replace the module.exports, for example if you're creating a class (like the b.js file in the above example), this is possible as well, just make sure that in the file that is starting the circular require, the 'module.exports = ...' statement happens before the require statement.
a.js (the main file run with node)
var ClassB = require("./b");
var ClassA = function() {
this.thing = new ClassB();
this.property = 5;
}
var a = new ClassA();
module.exports = a;
b.js
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
util.log(a.property);
}
module.exports = ClassB;
var a = require("./a"); // <------ this is the only necessary change
The solution is to 'forward declare' your exports object before requiring any other controller. So if you structure all your modules like this and you won't run into any issues like that:
// Module exports forward declaration:
module.exports = {
};
// Controllers:
var other_module = require('./other_module');
// Functions:
var foo = function () {
};
// Module exports injects:
module.exports.foo = foo;
What about lazy requiring only when you need to? So your b.js looks as follows
var ClassB = function() {
}
ClassB.prototype.doSomethingLater() {
var a = require("./a"); //a.js has finished by now
util.log(a.property);
}
module.exports = ClassB;
Of course it is good practice to put all require statements on top of the file. But there are occasions, where I forgive myself for picking something out of an otherwise unrelated module. Call it a hack, but sometimes this is better than introducing a further dependency, or adding an extra module or adding new structures (EventEmitter, etc)
You can solve this easily: just export your data before you require anything else in modules where you use module.exports:
classA.js
class ClassA {
constructor(){
ClassB.someMethod();
ClassB.anotherMethod();
};
static someMethod () {
console.log( 'Class A Doing someMethod' );
};
static anotherMethod () {
console.log( 'Class A Doing anotherMethod' );
};
};
module.exports = ClassA;
var ClassB = require( "./classB.js" );
let classX = new ClassA();
classB.js
class ClassB {
constructor(){
ClassA.someMethod();
ClassA.anotherMethod();
};
static someMethod () {
console.log( 'Class B Doing someMethod' );
};
static anotherMethod () {
console.log( 'Class A Doing anotherMethod' );
};
};
module.exports = ClassB;
var ClassA = require( "./classA.js" );
let classX = new ClassB();
A solution which require minimal change is extending module.exports instead of overriding it.
a.js - app entry point and module which use method do from b.js*
_ = require('underscore'); //underscore provides extend() for shallow extend
b = require('./b'); //module `a` uses module `b`
_.extend(module.exports, {
do: function () {
console.log('doing a');
}
});
b.do();//call `b.do()` which in turn will circularly call `a.do()`
b.js - module which use method do from a.js
_ = require('underscore');
a = require('./a');
_.extend(module.exports, {
do: function(){
console.log('doing b');
a.do();//Call `b.do()` from `a.do()` when `a` just initalized
}
})
It will work and produce:
doing b
doing a
While this code will not work:
a.js
b = require('./b');
module.exports = {
do: function () {
console.log('doing a');
}
};
b.do();
b.js
a = require('./a');
module.exports = {
do: function () {
console.log('doing b');
}
};
a.do();
Output:
node a.js
b.js:7
a.do();
^
TypeError: a.do is not a function
The important thing is not to re-assign the module.exports object that you have been given, because that object may have already been given to other modules in the cycle! Just assign properties inside module.exports and other modules will see them appear.
So a simple solution is:
module.exports.firstMember = ___;
module.exports.secondMember = ___;
The only real downside is the need to repeat module.exports. many times.
Similar to lanzz and setec's answers, I have been using the following pattern, which feels more declarative:
module.exports = Object.assign(module.exports, {
firstMember: ___,
secondMember: ___,
});
The Object.assign() copies the members into the exports object that has already been given to other modules.
The = assignment is logically redundant, since it is just setting module.exports to itself, but I am using it because it helps my IDE (WebStorm) to recognise that firstMember is a property of this module, so "Go To -> Declaration" (Cmd-B) and other tooling will work from other files.
This pattern is not very pretty, so I only use it when a cyclic dependency issue needs to be resolved.
It is fairly well suited to the reveal pattern, because you can easily add and remove exports from the object, especially when using ES6's property shorthand.
Object.assign(module.exports, {
firstMember,
//secondMember,
});
the extremely simple solution is often:
usually you'd have the require at the top of the file ...
var script = require('./script')
function stuff() {
script.farfunction()
}
instead, just require it "in the function"
function stuff() {
var _script = require('./script')
_script.farfunction()
}
An other method I've seen people do is exporting at the first line and saving it as a local variable like this:
let self = module.exports = {};
const a = require('./a');
// Exporting the necessary functions
self.func = function() { ... }
I tend to use this method, do you know about any downsides of it?
TL;DR
Just use exports.someMember = someMember instead of module.exports = { // new object }.
Extended Answer
After reading lanzz's response I could finally figure it out what is happening here, so I'll give my two cents on the subject, extending his answer.
Let's see this example:
a.js
console.log("a starting");
console.log("a requires b");
const b = require("./b");
console.log("a gets b =", b);
function functionA() {
console.log("function a");
}
console.log("a done");
exports.functionA = functionA;
b.js
console.log("b starting");
console.log("b requires a");
const a = require("./a");
console.log("b gets a =", a);
function functionB() {
console.log("On b, a =", a)
}
console.log("b done");
exports.functionB = functionB;
main.js
const a = require("./a");
const b = require("./b");
b.functionB()
Output
a starting
a requires b
b starting
b requires a
b gets a = {}
b done
a gets b = { functionB: [Function: functionB] }
a done
On b, a = { functionA: [Function: functionA] }
Here we can see that at first b receives an empty object as a, and then once a is fully loaded, that reference is updated through exports.functionA = functionA. If you instead replace the entire module with another object, through module.exports, then b will lose the reference from a, since it will point out to the same empty object from the beginning, instead of pointing to the new one.
So if you export a like this: module.exports = { functionA: functionA }, then the output will be:
a starting
a requires b
b starting
b requires a
b gets a = {}
b done
a gets b = { functionB: [Function: functionB] }
a done
On b, a = {} // same empty object
Actually I ended up requiring my dependency with
var a = null;
process.nextTick(()=>a=require("./a")); //Circular reference!
not pretty, but it works. It is more understandable and honest than changing b.js (for example only augmenting modules.export), which otherwise is perfect as is.
Here is a quick workaround that I've found use full.
On file 'a.js'
let B;
class A{
constructor(){
process.nextTick(()=>{
B = require('./b')
})
}
}
module.exports = new A();
On the file 'b.js' write the following
let A;
class B{
constructor(){
process.nextTick(()=>{
A = require('./a')
})
}
}
module.exports = new B();
This way on the next iteration of the event loop classes will be defined correctly and those require statements will work as expected.
One way to avoid it is to don't require one file in other just pass it as an argument to a function what ever you need in an another file.
By this way circular dependency will never arise.
If you just can't eliminate circular dependencies (e.g useraccount <---> userlogin), there's one more option...
Its as simple as using setTimeout()
//useraccount.js
let UserLogin = {};
setTimeout(()=>UserLogin=require('./userlogin.js'), 10);
class UserAccount{
getLogin(){
return new UserLogin(this.email);
}
}
//userlogin.js
let UserAccount ={};
setTimeout(()=>UserAccount=require('./useraccount.js'), 15);
class UserLogin{
getUser(){
return new User(this.token);
}
}