Although I believe I'm following the instructions here for setting up $httpBackend to pass selected requests to the server, it's not working for me.
Here is a Plunkr with a failing test that shows what I'm doing and explains in comments what seems to be going wrong.
My spelunking suggests that, for some reason, the mock $httpBackend
doesn't have an inner copy of the real $httpBackend
so that, when it comes time to pass through the XHR request, it passes it to the mock $httpBackend
instead. That second call throws an exception because it doesn't know what to do with the request.
I remember with appreciation your post on midway testing. You identify an important range of integration testing that falls between unit- and E2E-testing. I am standing on that middle ground.
I don't think you are being snarky at all. Your answer is perfect reasonable ... or it would be reasonable if it weren't contradicted by the text of the "API reference / ngMockE2E / $httpBackend". I quote:
This implementation can be used to respond with static or dynamic responses via the
when
api and its shortcuts (whenGET
,whenPOST
, etc) and optionally pass through requests to the real$httpBackend
for specific requests (e.g. to interact with certain remote apis or to fetch templates from a webserver) ...[I]n an end-to-end testing scenario or in a scenario when an application is being developed with the real backend api replaced with a mock, it is often desirable for certain category of requests to bypass the mock and issue a real http request .... To configure the backend with this behavior use the
passThrough
request handler of when instead ofrespond
.[emphasis mine].
The documentation is silent on the matter of E2E $httpBackend
usage within a Jasmine environment. I can't think of a reason to preclude it. If there is such a reason they should state it clearly. Seriously, who reads about a mock component and doesn't anticipate using it in a test environment?
To "pass through requests to the real $httpBackend
for specific requests, e.g. to interact with certain remote apis" is precisely what I intend to do. What could they possibly mean by "real $httpBackend" except the non-mock version of that component?
I do not understand your claim that
The
ngMocksE2E
module is designed to be used on the "server" side of things where the actual angular application is executing.
The word "server" appears exactly 3 times on that page, not once suggesting that any application code would be executed on a "server". I don't know what you mean by the "actual angular application" executing on "the 'server' side of things."
The documentation is perfectly clear that the E2E $httpBackend
is not limited to E2E testing. It is also for "a scenario when an application is being developed with the real backend api replaced with a mock".
That's just a half step away from my scenario in which an application is being tested with the real backend api."
In my scenarios, the SUT is calling upon a component which fetches data from a server. My tests exist to verify that this dependent component succeeds in making such requests of the real backend and will retrieve or save data in the expected manner. This is an integration test that cannot be adequately satisfied by mocking the behavior of the backend.
Of course I can test (and do test) with mock XHR responses the component's ability to respond properly to what I predict will be the backend's behavior. That is not the same as validating that the component responds appropriately to the actual backend's behavior ... which might change as the application evolves and depart from the mocked responses in some significant way.
I would consider using your midway tester for this purpose if I understood how to swap it into the SUT's code path. I don't. I think the component making XHR requests is inaccessible to your ngMidwayTester
. But I do know how to jam a real XHR helper into the pipeline if I have to.
Here is where I stand at the moment.
Either someone can show how to make $httpBackend
pass certain requests through to the server - as the documentation proclaims it can - or I will replace the passThrough
implementation myself with a working XHR implementation.
I prefer the first option. If driven to the second, I shall offer a link to it here for the benefit of others who share my needs and my interpretation of the API documentation.
Is there a 3rd way I'm missing?
I stumbled on the same problem but instead of implementing a rich API or replacing the original angular-mocks simply added in the following helper:
angular.module('httpReal', ['ng'])
.config(['$provide', function($provide) {
$provide.decorator('$httpBackend', function() {
return angular.injector(['ng']).get('$httpBackend');
});
}])
.service('httpReal', ['$rootScope', function($rootScope) {
this.submit = function() {
$rootScope.$digest();
};
}]);
It patches two issues that prevent an HTTP request from getting passed through:
Restores original $httpBackend
;
Provides a method to fulfill the requests, as otherwise they would be in AngularJS' queue waiting for a digest loop.
describe('my service', function() {
var myService, httpReal;
beforeEach(module('myModule', 'httpReal'));
beforeEach(inject(function( _myService_, _httpReal_ ) {
myService = _myService_;
httpReal = _httpReal_;
}));
it('should return valid data', function(done) {
myService.remoteCall().then(
function(data) {
expect(data).toBeDefined();
done();
}, function(error) {
expect(false).toBeTruthy();
done();
});
httpReal.submit();
});
});
The following is an explanation of the purpose of the $httpBackend
that is in the ngMockE2E
module.
The ngMockE2E
module is simply not designed nor intended to be used from within a jasmine specification.
When doing end-to-end testing there are two sides to the test. One is the angular application that is being tested, the other is the angular-scenario code that lives in the Jasmine Specification.
Under E2E tests there are no angular module, or ng-mocks, or anything angular-related on the jasmine side of things (other than the scenario runner).
The ngMocksE2E
module is designed to be used on the "server" side of things where the actual angular application is executing. It's main purpose is to enable us to pre-can responses so that integration-level UI testing can proceed much quicker than if each page actually went to the server for JSON.
When using jasmine along with ng-mocks
, angular will always replace the $httpBackend with the mock backend. When adding the ngMocksE2E
module it will not be able to get ahold of any "real" $httpBackend
and as you have already found out, will just wrap the mock and delegate to it on the pass-through.
It would seem that the kind of test you are trying to write is a test that doesn't test the UI integration, but tests the application javascript and server integration.
This is perfectly legitimate style of testing (referred to some as 'midwayTesting' in the angular community). Your problem is you are using the wrong tool.
I would take a look at this:
https://github.com/yearofmoo/ngMidwayTester
Which you would use instead of angular-mocks and angular.module() in order to facilitate the kind of testing I'm assuming you want to do.
You can read more about it here:
http://www.yearofmoo.com/2013/01/full-spectrum-testing-with-angularjs-and-karma.html
(apologies if you have already been linked there)
EDIT: (To address additional comments in question)
You have a real beef in that the documentation is not clear that ngMockE2E
cannot be used on the client (i.e. karma/jasmine) side of an end-to-end testing setup. It is not unreasonable to interpret things like you have interpreted them, but it doesn't change the fact that the interpretation is wrong.
The ngMockE2E will pass through requests if instructed when used on the server side of an application rather than on the client side. This is meant so that you can still pass through certain requests that are hard to mock as pre-canned responses. What I mean by client and server-side is that in end-to-end testing there are two ends. You have the application to be tested which is served by a standard application server, and you have the test code that is driving the application usually executing in Karma or another test runner, which uses standard HTTP requests to communicate to the application which is executing in another process.
If you look at the documentation and how to setup ngMockE2E
you will notice there is no mention of Jasmine, and the instructions are for how to set up in a real angular application:
myAppDev = angular.module('myAppDev', ['myApp', 'ngMockE2E']);
myAppDev.run(function($httpBackend) {
phones = [{name: 'phone1'}, {name: 'phone2'}];
// returns the current list of phones
$httpBackend.whenGET('/phones').respond(phones);
// adds a new phone to the phones array
$httpBackend.whenPOST('/phones').respond(function(method, url, data) {
phones.push(angular.fromJson(data));
});
$httpBackend.whenGET(/^\/templates\//).passThrough();
//...
});
As you can see in this example, they are mocking all the JSON data instructions, while letting it still fetch the templates from the server.
In order to use it from jasmine the setup would be quite different, using angular.mock.module('ngMockE2E')
and then setting up the $httpBackend.whenGET()
in a beforeEach()
rather than in a module.run()
.
As far as ngMidwayTester
I linked you to, I believe this would, in fact, be compatible with ngMockE2E
. Essentially ngMidwayTester
replaces angular.mock.module()
and inject()
with it's own implementations. So you could use it like this:
beforeEach(function(){
tester = ngMidwayTester('app', 'ngMockE2E');
$http = tester.inject('$http');
$httpBackend = tester.inject('$httpBackend');
$rootScope = tester.inject('$rootScope');
});
This should work, because you are no longer using the ngMock module (which always gets included when you use angular.mock.module()
). Things should work exactly like you want them to using ngMidwayTester
.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With