LD.
Music, software, life… and stuff.
[ Twitter ] [ GitHub ] [ Linked In ]
In response to Rob’s post on functional testing.
Rob, a Geb committer and all around cool dude, recently posted some thoughts on function testing.
The following is what I intended to be a comment on Rob’s blog, but it became too long so I decided to post it here. You’ll notice that the language is a bit weird as in some places I am addressing Rob directly. It should make sense (no guarantees) if you pretend you are reading it as a comment on Rob’s blog.
This is not a flame war, though that would be fun. Rob’s a friend of mine and a respected colleague. This is just a discussion and exchange of ideas.
Without further ado, the comment…
I’ll try to be as objective as I can, but obviously as someone who is invested in Geb I come with a bit of a bias.
You seem to be mixing up several different issues, which is understandable as the post seems born out of frustration.
For my mind there are three discrete parts:
- The development environment (the language/framework/constructs you use to express the test)
- The execution environment
- The debugging environment
They lines blur, but I think it’s helpful to think about each part separately. Geb itself only lives in the first arena strictly speaking. How it works in the execution and debugging environment are largely aspects of the associated tooling, with the exception of static diagnostic information.
So a couple of points:
“I agree that as much testing as possible should be done at as low a level as possible”
What’s usually missing from this sentiment is the why. Too many people take this stance because testing at a low level is easier. This, in isolation, is not the right way to think about it. The end goal is always working software, not easy tests. Easy tests work towards working software by making it more likely that tests will be created in the first place and maintained. They are low cost/low value. So for me, low level testing is fine until the accuracy compromist becomes too high. If it doesn’t create a lot of confidence that the system will work, either because it assumes to much or is not backed up by higher level tests, then the developer is fooling themselves. Passing tests is not the end goal, working software is.
It’s with that in mind that I usually say to people that we need to “suck it up” to a certain extent with functional tests. You should identify if there is value in having automated tests that verify the end to end behaviour of your system, get a feel for what that value is and act accordingly. Then you can make cost/benefit judgements on how much effort to put into functional testing. Avoiding it because it’s not as easy as unit testing while not taking the value into account is irresponsible, in my opinion. However, we definitely should harness the frustration and desire for simplicity to make it easier.
“Even assuming you can optimise so that the application is running and you can run the test from an IDE then Selenium has to start up a browser instance, queries execute, views need to render, etc”
Selenium doesn’t have to start up a browser instance, it’s possible to use an running browser instance. Also, on modern development machines we should be talking about seconds at most to run queries and render views etc.
As for the Grails specific issues, you’re going to face the slow application startup when working with Grails no matter what your development environment. In theory, there’s no difference between different development environments here, but the difference comes in in execution environments. However, it’s possible to create an execution environment for Geb tests (run in IDE against a running application) that is equivalent to the Selenium IDE experience.
“Geb’s output when a waitFor condition fails is just Condition did not pass in x seconds.”
This is indeed painful. Even with implicit power asserts it doesn’t get much better. Providing static diagnostic information for functional tests (i.e. non dynamic like an interactive debugging session) is one of the unsolved problems in this area. I’m not aware of any tool that has a good approach to this. With Geb our approach is to just dump the browser state, but that information is generally not rich enough. Hopefully someone comes up with a good solution.
“Selenese is by no means great in this regard (a humble Condition timed out isn’t much help) but at least you can step back with the Selenium IDE and watch the damn thing not working much more easily.”
I think much more easily is a stretch. You can do the same thing with a Geb/WebDriver test. The only difference being is that the test method is your smallest level of granularity for re-running a part of the test. That’s a clear advantage that Selenium IDE has; the ability to run a selected portion of the story at any granularity. It seems like this would break down for most stories though as they require context that needs to be run as well.
“The most productive I’ve ever been when writing functional tests has been when using Selenium IDE.”
(and the rest of this paragraph and the next)
There’s only one point you make here that can’t be replicated with Geb/WebDriver in an IDE: selecting a portion of the test at any granularity and just running that. Everything else is also possible in an IDE, and is the exact same process you would use to debug other code. These are all points on the execution/debugging environment not the development environment.
“Despite these significant failings writing tests in Selenium IDE is very effective. Maintaining a suite of such tests is another matter. Working on a long-running project the failings of Selenese tests start to increase logarithmically.”
This is the point for me. Experience has shown that the majority of the cost of a functional test suite comes in maintenance, not initial development. Therefore the sensible thing to do is to optimise for this first and development speed second. Maintenance should be foremost in the developer’s mind when developing functional test. The PageObject pattern and inline test data are the two best techniquest that I am aware of that help here.
“I’m convinced that the goal of writing tests in the same language as the application is a pretty vapid one. Working inside one’s comfort zone is all very well but too many times I’ve seen things tested using Selenium or Geb that would be better tested with a unit-level JavaScript test.”
It’s not necessarily about using the same language, it’s about integration and sharing data structures. Equating functional testing to black box testing is incorrect I think. I have no problem with stimulating the system as a user, but poking inside to verify internal state. To me, that’s more productive and accurate.
JavaScript unit tests aren’t a replacement for functional tests. You shouldn’t be using functional tests to unit test JavaScript, but having JavaScript unit tests doesn’t relieve you from having functional tests. It can mean you need less though. If a test doesn’t emulate what a user does (including the user environment) to a reasonable degree of accuracy it is not a functional test.
“As a Grails developer I’ve looked enviously at how fast tests run in Rails apps but that’s nothing compared to watching a couple of hundred Jasmine tests run in under a second.”
I think you’re comparing apples and oranges here. I’d have little faith in a system that only has JavaScript unit tests.
“I was at one time convinced that the ability to use loops and conditional statements in Groovy made it a more suitable testing language than Selenese but honestly, how often are such constructs really required for tests?”
In my experience, these constructs are more useful when modelling the concept with appropriate abstractions. As interaction with web pages becomes richer (gestures, dragging etc.), I think we’ll see it become more useful there. That is, one logical interaction may have looping/branching logic internally but we want to model it as one abstraction.
“Building Geb content definitions with deeply nested layers of Module types is time consuming & difficult.”
I agree. However, when you put the effort in and do this right you get maintainable tests plus a wealth of predefined abstractions to draw on in future tests. That is, high cost/high value. With my Geb hat on, I think we have the best story in this area in terms of provided constructs. What we lack is the shared knowledge and experience to deploy these constructs effectively. I think every tool that tries to do this has the same problem though. In a word, inexperience.
“I can’t help thinking the page object model approach is coming at the problem from the wrong angle. Instead of abstracting the UI shouldn’t we be abstracting the behaviour?”
I think you need both. You can’t do away with abstracting the UI.
Geb’s in a tight spot here in that it has no execution model. This makes it awkward to introduce behaviour abstractions as we’d have to wrestle the execution control away from the execution framework we are integrating with. Ultimately, this abstraction belongs in JUnit/Spock/whatever. Spock is missing this capability, and that’s a known problem.
“The most impressive Selenium extension I’ve seen is Steve Cresswell’s Natural Language Extensions that layers something like JBehave’s feature definition language on top of Selenese.”
Not being a Selenese expert, this seems like it would give you the the behaviour abstractions you desire but you’d have to either largely abandon UI abstraction or maintain that separately.
I’d love to see someone start a project adapting something like Fitnesse or some other natural language based tool to Geb. You’d get the same behaviour abstraction benefits, but get to use Geb’s page modelling to maintain the UI abstractions. Hopefully someone starts this.
As for cucumber, if you want natural language authoring I think it’s the choice. There are posts out there about people layering it on top of Geb.
I’m very skeptical of FuncUnit, but not well educated about it. It’s not clear to me how it fits into an automated build and it seems like it would have a very high mental context switching cost. You also give up on data structure sharing entirely.
That’s it.
So the take home for me is that we need to spend more effort on the execution and debugging environments for Geb. Perhaps this will be the focus of the 0.7.x series which will be starting soon.
Posted: Nov 25th, 2011 @ 12:49 am
Geb Modules and isDisplayed()
A question recently came up on the Geb user mailing list about checking to see if a module is displayed on a page. It turns out the concept of displayed could do with some exploration.
Say we have the following html…
<p class="a" style="display:none"/>
<p class="b"/>
The following assertions will pass…
def a = $("p.a")
def b = $("p.b")
def c = $("p.c")
assert a && !a.displayed
assert b && b.displayed
assert !c && !c.displayed
So isDisplayed() means that the navigator matches some content and the first thing that it has matched is actually visible on the page. $("p.c").displayed returns false because it doesn’t match anything.
Navigator objects implement the Groovy Truth, so those assertions could be written as…
assert a.size() != 0 && !a.displayed
Modules also implement the Groovy Truth using the same test, so…
class ParagraphA extends Module {
static base = { $("p.a") }
}
class ParagraphB extends Module {
static base = { $("p.b") }
}
class ParagraphB extends Module {
static base = { $("p.c") }
}
Now assume we have moduleA, moduleB and moduleC vars that are instances of the corresponding modules, and we can write the following assertions…
assert moduleA && !moduleA.$().displayed
assert moduleB && moduleB.$().displayed
assert !moduleC && !moduleC.$().displayed
You can get the “base” of a module at anytime with «module».$(), just like you can do relative lookups (i.e. someModule.$(“p.d”))
Posted: Jun 16th, 2011 @ 9:56 am
Painless Page Identification with Geb & Grails
This will be a discussion of a handy pattern to use to make page identification simple and easy when working with Geb and Grails.
Geb and Page Identification
Geb supports (and encourages) using the Page Object Pattern. An aspect of this is having someway of being able to verify that the page object actually does represent the page that the browser is at. Consider filling out a form. When you submit the form you might expect to go to a confirmation page, but because of a validation error you get returned to the form page. In your test code after submitting the form you go on testing for things on the page and the results can be confusing. The best thing to do before testing parts of the page is to test that the page is indeed what you expect it to be. There are other reasons for page identification, but this is the main one.
Geb implements page identification through it’s “at checking”. In short, pages declare an “at” closure that contains some checks of the page structure, and returns true if the actual browser page is what the page object represents.
import geb.*
class ManualPage extends Page {
static at = { $("h1", text: contains("Manual")) }
}
This “at check” is used implicitly when using the browser at(«PageClass») method…
Browser.driver {
to ManualPage
assert at(ManualPage)
}
The at(«PageClass») method does two things:
- Checks the browser’s page object is an instance of the given class
- Checks the browser’s page object’s at checker returns
true
The use of a Class with the at() method may seem strange. After using Geb for a short amount of time the reasons do become clear. Primarily, your test code is likely to not to declare page changes (such as when content declares a to property). This means that asserting the type of the current page object becomes important.
Writing a good at checker can be hard at times. You want to test enough so that you are sure the page is a match, yet you don’t want to test too much to make your checks fragile.
Grails and Conventions
Fortunately, Grails’ controller/action convention makes identifying pages cost free and definitive (for almost all cases). The trick lies in the controllerName and actionName variables being available in each action’s view implicitly. This allows you to make your Grails GSP templates look something like this…
<html>
<head>
<meta name="pageId" content="${controllerName}.${actionName}" />
…
</head>
<body>
…
</body>
</html>
Every view generated from a controller action in your application now has a unique identifier embedded into the page in an unobtrusive manner.
Action Page Objects
Geb goes out of it’s way to support using inheritance with page objects. This means that you can define an at checker in a common base class that reads our meta pageId property and have that used for each subclass.
import geb.*
abstract class GrailsPage extends Page {
// To be overridden by subclasses
static controller = null
static action = null
static at = {
// delegate here is the original page _instance_ (i.e. the subclass)
def expectedPageControllerName = delegate.class.controller
if (expectedPageControllerName == null) {
throw new IllegalStateException("${delegate.class} forgot to declare which controller it belongs to")
}
def expectedPageActionName = delegate.class.action
if (expectedPageActionName == null) {
throw new IllegalStateException("${delegate.class} forgot to declare which action it is")
}
def actualPageControllerName = controllerName
def actualPageActionName = actionName
assert actualPageControllerName == expectedPageControllerName
assert actualPageActionName == expectedPageActionName
true // at checkers must return true
}
static content = {
pageId { $("meta", name: "pageId").@content }
controllerName { pageId.split('\\.')[0] }
actionName { pageId.split('\\.')[1] }
}
}
Here we do three things:
- declare
static‘controller’ and ‘action’ properties that subclasses will populate (more on this later) - declare an “at checker” that compares the declared controller/action against those in the page
- declare what the
controllerNameandactionNamebits of content on the page are (consult the Book Of Geb for more on this content DSL)
We now have a rock solid page identification mechanism that from here on in in the development and testing of our application costs us almost nothing. The only thing you need to do when writing new page objects is set the controller and action static properties…
class ViewCartPage extends GrailsPage {
static controller = "cart"
static action = "view"
}
Or in the case where all your actions for a particular controller share content, you may want to have an abstract base class for all actions of that controller…
abstract class CartPage extends GrailsPage {
static controller = "cart"
}
class ViewCartPage extends CartPage {
static action = "view"
}
Geb and Inheritance (and Groovy)
This is made possible by Geb supporting inheritance of content definitions (content definitions are merged) and Groovy supporting a dynamic resolution of static methods and properties.
Given the above classes, the following works…
assert ViewCartPage.controller == "cart"
However, if we were calling this from Java, it would not work. In Java, static methods/properties are never inherited like they are for all intents and purposes in Groovy. This means that when requesting a page class’s “at checker” it will search the inheritance hierarchy for one (e.g. assert ViewCartPage.at == GrailsPage.at). This also means that at any point in the inheritance hierarchy you can override any static method/property defined in a super class (but it only works if calling from Groovy).
Summary
We discussed a useful pattern for page identification (“at checking” in Geb parlance) with Geb and Grails. I have used this pattern in anger and have found it to be very useful. Give it a try.
Posted: Aug 26th, 2010 @ 7:22 pm
