I maintain a complex Angular (1.5.x) application that is being E2E tested using Protractor (2.5.x). I am experiencing a problem with this approach, which presents primarily in the way the tests seem flaky. Tests that worked perfectly well in one pull request fail in another. This concerns simple locators, such as by.linkTest(...). I debugged the failing tests and the app is on the correct page, the links are present and accessible.
Has anyone else experienced these consistency problems? Knows of a cause or workaround?
Just Say No to More End-to-End Tests!
That said, here are the few things you can do to tackle our mutual merciless "flakiness" enemy:
update to the latest Protractor (currently 4.0.0) which also brings latest selenium and chromedriver with it
turn off Angular animations
use dragons browser.wait() with a set of built-in or custom Expected Conditions. This is probably by far the most reliable way to approach the problem. Unfortunately, this is use-case and problem specific, you would need to modify your actual tests in the problematic places. For example, if you need to click an element, wait for it to be clickable:
var EC = protractor.ExpectedConditions;
var elm = $("#myid");
browser.wait(EC.elementToBeClickable(elm), 5000);
elm.click();
maximize the browser window (to avoid random element not visible or not clickable errors). Put this to onPrepare():
browser.driver.manage().window().maximize();
increase the Protractor and Jasmine timeouts
slow Protractor down by tweaking the Control Flow (not sure if it works for 4.0.0, please test)
manually call browser.waitForAngular(); in problematic places. I am not sure why this helps but I've seen reports where it definitely helped to fix a flaky test.
use the jasmine done() callback in your specs. This may help to, for example, not to start the it() block until done is called in beforeEach()
return a promise from the onPrepare() function. This usually helps to make sure things are prepared for the test run
use protractor-flake package that would automatically re-run failed tests. More like a quick workaround to the problem
There are also other problem-specific "tricks" like slow typing into the text box, clicking via JavaScript etc.
Yes, I think all of us experienced such flakiness issue.
Actually, the flakiness is quite common issue with any browser automation tool. However, this is supposed to be less in case of Protractor as Protractor has built-in wait consideration which performs actions only after loading the dom properly. But, in few cases you might have to use some explicit waits if you see intermittent failures.
I prefer to use few intelligent wait methods like:
function waitForElementToClickable(locator) {
var domElement = element(by.css(locator)),
isClickable = protractor.ExpectedConditions.elementToBeClickable(domElement);
return browser.wait(isClickable, 2000)
.then(function () {
return domElement;
});
}
Where 2000 ms is used as timeout, you can make it configurable using a variable.Sometimes I also go with browser.sleep() when none of my intelligent wait works.
It's been my experience that some methods (eg. sendKeys()) do not always fire at the expected time, within the controlFlow() queue, and will cause tests to be flakey. I work around this by specifically adding them to the controlFlow(). Eg:
this.enterText = function(input, text) {
return browser.controlFlow().execute(function() {
input.sendKeys(text);
});
};
A workaround that my team has been using is to re-run only failed tests using the plugin protractor-errors. Using this tool, we can identify real failures versus flakey tests within 2-3 runs. To add the plugin, just add a require statement to the bottom of the Protractor config's onPrepare function:
exports.config = {
...
onPrepare: function() {
require('protractor-errors');
}
}
You will need to pass these additional parameters when to run your tests with the plugin:
protractor config.js --params.errorsPath 'jasmineReports' --params.currentTime (timestamp) --params.errorRun (true or false)
There is also a cli tool that will handle generating the currentTime if you don't have an easy way to pass in a timestamp.
Related
I have a test that sometimes gets run with the command line flag --disable-multiple-windows and sometimes it doesn't. I would like the test to skip over a section if multiple windows are disabled. Is there an easy way to tell that within a test?
My best solution so far is something like the following
let disableMultipleWindows=true;
try{
await t.openWindow('www.google.com');
disableMultipleWindows=false
await t.closeWindow();
}
But I'm wondering if there's a better way.
At present, disabling support for multiple browser windows is possible only for tests run scenarios. There is no access to the --disable-multiple-windows option within a test body. You can learn more about this option at Disable Support for Multiple Windows.
Could you please clarify your scenario of using the --disable-multiple-windows option? I think you can split your tests into different tests suites: with/without multiple browser windows.
I am testing a large project with long scenarios (some with more than 100 interactions with webpage). I would like to break them down into shorter steps that run in sequence (like in Mocha) but I don't know how to do that.
Example: In a single test, I would like to run
fixture('test1')
test('test1', async (t) => {
...login
...createSubAccount
...modifySubAccount
...activateSubAccount
})
where each of the steps would show in console and in report. Right now, the only thing I know how to do is to put each step into its own test() context, but that means that if e.g. createSubAccount fails, modifySubAccount and activateSubAccount will still run (even though the workflow already failed). Also, there is the unhappy part that each test() clears the browser (but I can deal with that).
In short: How can I split the tests in a way that if a single substep of fixture fails, the whole fixture fails immediately? Or similar thing, but for test()?
Also, I don't want the whole pipeline to end on the first test failure, as would happen with --stopOnFirstFail flag - I want to run all the tests, to find which are failing.
test() is the smallest unit. The idea is it's an independent piece of testing code, e.i. a bunch of test steps. This doesn't change no matter what tool you use (TestCafe, Playwright, Puppeteer, Cypress, mocha, Jest, ...).
And so:
Right now, the only thing I know how to do is to put each step into its own test() context, but that means that if e.g. createSubAccount fails, modifySubAccount and activateSubAccount will still run (even though the workflow already failed).
seems like breaking one of the main principles of tests, that is they are independent. Don't split test steps that belong together between different tests.
If the only drawback now is the length of your test, why don't you do it like you hinted at in the example:
test('test1', async (t) => {
login();
createSubAccount();
modifySubAccount();
activateSubAccount();
});
you can create functions for login, createAccount etc. and then use only such function in your tests, which would make them as short as shown here. You can also easily create various scenarious:
test('activate account without modification', async (t) => {
login();
createSubAccount();
activateSubAccount();
});
test('create account', async (t) => {
login();
createSubAccount();
});
test('create account without login', async (t) => {
createSubAccount();
});
// and so on
It doesn't even look that long.
TestCafe does not support the functionality you require at the moment. The only solution I could think of is, as you proposed, to implement your test as a fixture with steps as tests, use disablePageReloads feature (NOTE: it is experimental), track the number of passed tests manually, and check it at the beginning of each test. It is a bit tedious, but it should work as you need.
Another solution that has not been implemented yet and the easiest way to split the long test into steps is to simply divide it into functions. The only issue that may arise is related to reporting. Even if you implement a custom reporter, there is no possibility to pass information about the steps into it (you can vote for the corresponding feature request).
Also, I would like to draw your attention to Page Model pattern. This can shrink your tests and make them more readable.
Please open a new feature request with a comprehensive description if you have a better idea of how this should be done.
I have a webapp that uses dojo widgets and drag-n-drop functionalities and I'm using Intern in order to test it. Now I want to test the drag-n-drop mechanism, and for this I hoped to use the Leadfoot's helper, DragAndDrop.js
As seen in the script's example, here my code:
return new DragAndDrop(remote)
.findByXpath(source)
.dragFrom()
.end()
.findByXpath(target)
.dragTo()
I have the return statement because this code is part of a promise chain.
However, it seems to be not working and I do not get any kind of errors|exceptions, neither in the browser neither in selenium neither on intern side. Honestly, I have no idea where to start from.
Any suggestion? May I provide further information?
Have you tried
return remote.findByXpath(target)
.then(function(targetNode){
return remote.findByXpath(source)
.moveMouseTo(1,1)
.pressMouseButton().sleep(500)
.moveMouseTo(targetNode).sleep(500)
.releaseMouseButton();
});
Note: sleep isn't necessary, I put it here so that you can see clearer the actions
I would like to inject a protractor instance into my tests, and then use this to perform navigation and element selection, but it appears that the functionality hanging off the Protractor object is all asynchronous, and the functionality on browser and element is synchronous.
Is there a way to access the synchronous behavior via the Protractor object?
Also: I have seen tests that invoke the following at the start:
protractor.getInstance()
...and I have seen tests that use the globally available browser and element objects directly.
What are the important differences between these two approaches?
Nothing in protractor is synchronous. It just looks that way before jasmine has been patched (by jasminewd) to wait for the async code so that it's easier to work with.
Please read https://code.google.com/p/selenium/wiki/WebDriverJs to see how controlflow makes everything in webdriver look synchronous. Then read https://github.com/angular/protractor/blob/master/docs/control-flow.md to see how protractor did it with jasminewd.
The link shared by #glepretre (stackoverflow.com/q/25496379/3049002) tells you the difference between ptor.get and browser.get. In short, use browser.get()
I am putting together some ideas for our automated testing platform and have been looking at Selenium for the test runner.
I am wrapping the recorded Selenium C# scripts in an MbUnit test, which is being triggered via the MbUnit NAnt task. The Selenium test client is created as follows:
selenium = new DefaultSelenium("host", 4444, "*iexplore", "http://[url]/");
How can I pass the host, port and url settings into the test so their values can be controlled via the NAnt task?
For example, I may have multiple Selenium RC servers listening and I want to use the same test code passing in each server address instead of embedding the settings within the tests themselves.
I have an approach mocked up using a custom NAnt task I have written but it is not the most elegant solution at present and I wondered if there was an easier way to accomplish what I want to do.
Many thanks if anyone can help.
Thanks for the responses so far.
Environment variables could work, however, we could be running parallel tests via a single test assembly so I wouldn't want settings to be overwritten during execution, which could break another test. Interesting line of thought though, thanks, I reckon I could use that in other areas.
My current solution involves a custom NAnt task build on top of the MbUnit task, which allows me to specify the additional host, port, url settings as attributes. These are then saved as a config file within the build directory and then read in by the test assemblies. This feels a bit "clunky" to me as my tests need to inherit from a specific class. Not too bad but I'd like to have less dependencies and concentrate on the testing.
Maybe I am worrying too much!!
I have a base class for all test fixtures which has the following setup code:
[FixtureSetUp]
public virtual void TestFixtureSetup ()
{
BrowserType = (BrowserType) Enum.Parse (typeof (BrowserType),
System.Configuration.ConfigurationManager.AppSettings["BrowserType"],
true);
testMachine = System.Configuration.ConfigurationManager.AppSettings["TestMachine"];
seleniumPort = int.Parse (System.Configuration.ConfigurationManager.AppSettings["SeleniumPort"],
System.Globalization.CultureInfo.InvariantCulture);
seleniumSpeed = System.Configuration.ConfigurationManager.AppSettings["SeleniumSpeed"];
browserUrl = System.Configuration.ConfigurationManager.AppSettings["BrowserUrl"];
targetUrl = new Uri (System.Configuration.ConfigurationManager.AppSettings["TargetUrl"]);
string browserExe;
switch (BrowserType)
{
case BrowserType.InternetExplorer:
browserExe = "*iexplore";
break;
case BrowserType.Firefox:
browserExe = "*firefox";
break;
default:
throw new NotSupportedException ();
}
selenium = new DefaultSelenium (testMachine, seleniumPort, browserExe, browserUrl);
selenium.Start ();
System.Console.WriteLine ("Started Selenium session (browser type={0})",
browserType);
// sets the speed of execution of GUI commands
if (false == String.IsNullOrEmpty (seleniumSpeed))
selenium.SetSpeed (seleniumSpeed);
}
I then simply supply the test runner with a config. file:
For MSBuild I use environment variables, I create those in my CC.NET config then they would be available in the script. I think this would work for you too.
Anytime I need to integrate with an external entity using NAnt I either end up using the exec task or writing a custom task. Given the information you posted it would seem that writing your own would indeed be a good solution, However you state you're not happy with it. Can you elaborate a bit on why you don't think you current solution is an elegant one?
Update
Not knowing internal details it seems like you've solved it pretty well with a custom task. From what I've heard, that's how I would have done it.
Maybe a new solution will show itself in time, but for now be light on yourself!