Consider allowing e2e test to ignore certain timeouts

[juliemr]

Based on a discussion we had about angular/protractor#2950. Since this feature would require changes to Angular.js, moving discussion here.

The basic problem is that sometimes, apps have a long-running or repeating $timeout call for some reason. Protractor waits for all timeouts to finish before reporting stable, so this causes it fail for any tests on that page.

An example use case would be a page using a widget that they don't necessarily control the code for, which has a long running $timeout. Ideally, they'd be able to ignore this timeout without changing the code for the widget.

Mitigating, currently available solutions:

• you can always turn off all waiting for Protractor using browser.ignoreSynchronization = true or by directly using browser.driver.<method>
• If you control the offending app code, use $interval instead of $timeout

Potential solutions that would require changes to angular:

• Instead of waiting for there to be 0 timeouts, the test writer could pass in a number X, and we could wait for there to be <= X.
• We could add yet another parameter to $timeout telling angular to ignore it for reporting stability (note that this requires some changes to the app code)
• We could add some ability to name your timeouts, and let protractor pass in a list of timeouts it should ignore. Concerns here: name conflicts, names getting minified. (note that this requires some changes to the app code)

[petebacondarwin]

1: Instead of waiting for there to be 0 timeouts, the test writer could pass in a number X, and we could wait for there to be <= X.

Although this would be easy to implement, I think this would not work in enough cases that it is not worth investing time in as a complete solution. Moreover it is fragile in the case that someone adds a new unexpected timeout that basically breaks unrelated tests.

2: We could add yet another parameter to $timeout telling angular to ignore it for reporting stability

While this is a nice option in itself, we already have a heavily overloaded $timeout function signature - in fact we already allow unlimited arguments on this function.

3: We could add some ability to name your timeouts, and let protractor pass in a list of timeouts it should ignore. Concerns here: name conflicts, names getting minified.

If we could work around the minification issue then this would be neat, since it is good practice to name your functions from a debugging point of view.

[Narretz]

You could exclude the names from minification in a built-step. Would be a bit of a hassle, but not too problematic.

[petebacondarwin]

@Narretz : By "exclude" did you mean "extract"?

[juliemr]

I don't think that introducing new build steps is a reasonable default recommendation - there are lots of different build systems and we can't have a preset solution for all of them. The only way to really do this generally would be to have the names a string constants.

[gkalpak]

So, I list a few cases that we would want to support ideally (feel free to add more):

1. Ignore a specific long-running timeout.
2. (a) Ignore all timeouts (but wait for other things, such as HTTP requests).
   (b) As an extra requirement, it would be nice to be able to toggle ignoring timeouts on/off.
3. Ignore everything (e.g. timeouts, HTTP requests etc).

[gkalpak]

After looking into it a little more, I believe that (3) is covered by browser.ignoreSynchronization.

Also, (for most cases) (2a) and (2b) can also be achieved with timely use of browser.ignoreSynchronization. There is also the protractor-xhr-only-plugin (which BTW is not mentioned on the Plugins docs), which does exactly that.

I believe that many issues people are running into, stem from the improper use of browser.ignoreSynchronization, due to overlooking the asynchronicity that Protractor introduces to tests. (I'm specuating here.)

The obvious thing to do is to improve the documentation of browser.ignoreSynchronization, so that people use it correctly. BTW, it isn't mentioned at all on the API docs. Is this intentional, @juliemr ?

Another option to make this easier for people to take advantage of, would be to expose a method on protractor to allow people to run blocks of code with ignoreSynchronization enabled/disabled. E.g.:

/* Code run here waits for sync */

protractor.ignoreSynchronization(function () {
  /* Code run here does not wait for sync */
});

/* Code run here waits for sync */

And we still don't have a good story for (1).

---

FWIW, I think we could add support in Angular's testability to wait for specific asynchronous event types. This would allow more flexibility for (2a)/(2b) and not require extra plugins.
E.g. augment $$testability.whenStable() like this:

// Wait for everything. The same as what we have now (so it's backwards compatible)
$$testability.whenStable(callback);

// Wait for $http requests only. Ignore $timeouts or other stuff.
$$testability.whenStable('http', callback);

// Wait for $timeouts only. Ignore $http requests or other stuff.
$$testability.whenStable('timeout', callback);

// Wait for X and Y event types only. Ignore other stuff.
$$testability.whenStable(['X', 'Y'], callback);

Then Protractor should expose an API for the users to specify what they care to wait for at a given point.

[petebacondarwin]

$$testability.whenStable(['X', 'Y'], callback);
What are X and Y here?