I am new to unit testing, I understand the basic concepts, and I am able to get unit testing setup correctly in my Cocoa projects; however the thing that is giving me a hard time is what exactly I should be writing unit tests for. For example, I know that you should write tests for model objects, but is that all I should be writing tests for? Should I also be writing tests for controllers, and views? What exactly would I be testing then? Could somebody please try and clarify what you should write unit tests for and what I should be testing?
The rule I generally follow is that all public interface needs to be tested. It's always up to you what exactly to test however the bigger your test coverage the less possibility there is for some nasty bug to creep out.
BTW for testing views I suggest Google Toolbox which allows you to compare screenshots.
Related
I have two questions. My first question is: Do applications exist which measure the coverage of GUI testing for web applications (not code, but the coverage of GUI components on web page)?
My second question is:
Is GUI testing with Selenium for example necessary if we have tests for javascript as well?
Thank you in advance.
You can write your custom application to find all dom elements using http://www.w3schools.com/js/js_htmldom_elements.asp, store this in some place and after completing your test automation framework run this utility to make sure that none of the elements are missing.
GUI test is required to make sure that all your integration points b/w several backend API are working. Also we will be sure that non of the UI elements are break over UI and all your business use cases are working as expected. Mostly UI testing is done for Acceptance Testing and we can show to the customer that all there use cases are working as expected. Later in the next release you can make sure that you are not breaking any UI code. UI testing gives us confidence while releasing to end users.
My goal is to be able to write core testing that I can use within a unit testing framework as well as UI testing with selenium.
For simple test like:
Scenario: Add two numbers
Given I have entered 50 into the calculator
And I have entered 70 into the calculator
When I press add
Then the result should be 120
I would create both unit tests to prove that my core API would pass as well as a Selenium test that would prove my UI is doing the correct thing as well.
I briefly tried to find anyone doing something similar through Google, but couldn't find any examples. So I guess my question is, has anyone here done anything similar?
On approach I had thought of was simple adding the feature files to a project or directory and using the add existing item as link as the solution.
Update: Adding feature files to a common directory and adding them as a link appears to be working great. The feature bindings regenerates for each project the feature file was included in so I can run unit tests in one and Selenium UI tests in the other.
First, lets start with why you might want to do this. Its laziness of the good kind.
The quality that makes you go to great effort to reduce overall energy expenditure. It makes you write labor-saving programs that other people will find useful, and document what you wrote so you don't have to answer so many questions about it. Hence, the first great virtue of a programmer. Also hence, this book. See also impatience and hubris. (p.609)
Larry Wall, Programming Perl
Except it isn't, because we aren't going to reduce our overall energy expenditure.
When you are using SpecFlow, the easy part to keep up to date is the plain text. You will find yourself refactoring the [Binding]s again and again, but the scenarios tend to be quite easy to work with, and need very little revision once they have been agreed.
In addition the [Binding]s are global. Load them in from any assembly and they are available to the SpecFlow runner. In respect of what you are trying this actually makes things harder as you need to put effort in to keep the UI bindings from being mixed up with the non-UI bindings.
Also consider the way that SpecFlow actually runs the tests from feature files. It's a two stage process.
When you save the .feature file the SpecFlow VS plugin generates a .feature.cs file.
When you run your test engine (e.g. NUnit) it ignores the plain text and uses compiled code from .feature.cs
So if you start using linked .features I have no idea if the SpecFlow plugin will generate .feature.cs for both instances of the file. (If you try this please let us know)
Second lets consider the features themselves. I think you will constantly finding yourself compromising your tests to make them fit the other place they are used. Already in the example you have given you have on the screen. If you are working with just the core API then there won't be a screen, so do we change this to fit better in a non-UI scenario?
Finally you have another thing to consider, just how useful will your tests be. If you have already got a test that tests the Core API, then what will it mean to run same test via Selenium. All you will really test is the UI layer. In my current employment we have a great number of regression tests that perform this very kind of testing, running up a client that connects to a server and manipulating the UI to get the desired scenarios enacted. These are the most fragile tests we have due to their scale. They constantly break and we basically have to check our entire codebase to find the line that broke them. Often something like 10-100 of them break just for a one line change. If these tests weren't so important to the regression cycle then the effort in maintaining them would just be too much. In my own personal projects I tend to remove these tests completely and instead with UIs, I avoid testing the View layer. With WPF MVVM, I execute Commands and test for results in ViewModels. If somebody then decides the TextBox should be a ComboBox or that it will work better in mauve, then my testing is isolated.
In short, there is a reason you can't find anything about this on Google :-)
In general (see http://martinfowler.com/bliki/TestPyramid.html), one should limit the number of automated tests that test the UI directly, and prefer tests that start at the presentation layer (just below the view layer), or below.
SpecFlow is agnostic; the tests can be implemented using e.g. Selenium at the UI layer or just MSTest or NUnit at any of the layers below.
However having said that I appreciate that you will have situations where you are doing ATDD and want to implement SpecFlow scenarios to match each of the acceptance criteria. Some of the criteria will be perfectly fine to test at a lower architectural level, but one or two of them may be specific to the GUI-- for example testing Login and ensuring that the user is redirected to the home page after successful login. If using Angular2 or React routing (see https://en.wikipedia.org/wiki/Single-page_application), that redirect is likely done in the GUI layer itself.
I don't have a perfect answer yet, but as a certified SpecFlow trainer, I have a vested interest in this! The way I am currently leaning is to use a complementary tool like CucumberJS for the front-end specific tests (such as testing React router redirects) and SpecFlow for tests at lower architectural layers. Our front-end uses Node.JS/Express and our backend is .NET Core. The idea is that the front-end tests mostly use the front-end only with mocked out AJAX calls to the backend (see sinonjs), and the back-end tests use EF Core with the in-memory option (see docs.efproject.net/en/latest/providers/in-memory/. So the tests all run fast.
Of course, you still need a few tests that actually go all the way through, but those are different-- we should call those integration tests. I do not believe that acceptance tests need to be integration tests. That way, you have a suite of acceptance tests from doing ATDD, plus a relatively small set of integration tests that test all the way front-to-back. The integration tests run more slowly and require more maintenance, so you separate them out into a different part of the CI/CD build chain.
I hope this makes sense. It is not so much solving the problem as avoiding the problem.
I am beginning to adopt Test Driven Design (TDD) behaviors and workflow for my iOS development projects. There is at least one impediment though in the context of legacy software. I will often have to add features to a pre-existing code base that I am new to. I will typically want to refactor at the beginning of working with the code base which will often have no tests available to ensure that my refactor-ings are not altering code functionality or worse, adding bugs.
My question is how do TDD folks bootstrap the whole process when the code is not written from scratch but rather legacy code that they are brought in to work on?
Thanks,
Doug
UPDATE
For a concrete example, I am using the example from Martin Fowler's Refactoring re-coded in Objective-C as a training device for TDD (and AppCode) >>
I built the code from tests. I found I needed to add instance variables to the Customer class to ensure I didn't screw up the cost calculations in the statement method as I grew the code. This is the fundamental issue I need insight into.
To begin with, if you don't understand the legacy code you're working with, you need to fix that before you do things which you're concerned may change behavior.
In your situation, after understanding the legacy code, I would write tests that will run against that legacy code. Once you're satisfied that these tests function as you expect, you're in a much better position to test your refactored code to ensure it functions as the old code did.
Environment: (C# WinForms application in Visual Studio Professional 2008)
I've been digging around a little for guidance on NUnit best practices. As a solo programmer working in a relatively isolated environment I'm hoping that collective wisdom here can help me.
Scott White has a few good starting points here but I'm not sure I totally agree with everything he's said -- particularly point 2. My instincts tell me that the closer a test is to the code being tested the more likely you are to get complete test coverage. In the comments to Scott's blog posting is a remark that just testing the public interface is considered best practice by some, but I would argue the test framework is not a typical class consumer.
What can you recommend as best practices for NUnit?
If by point 2, you mean the "bin folder per solution" -- I can see your point. Personally, I would simply add the reference to each test project. If, on the other hand, you really mean (1b) "don't put your tests in the same assembly as your code" I heartily agree with him and disagree with you. Your tests should be distinct from your production code in order to enhance code clarity and organization. Keeping your test classes separate helps the next programmer understand it more easily. If you need access to internals in your tests -- and you might since internal methods are "public" to the assembly, you can use the InternalsVisibleTo construct in the Assembly.cs file.
I, too, would recommend that, in general, it is sufficient to unit test only the public interface of the code. Done properly (using TDD), the private methods of your code will simply be refactorings of previous public code and will have sufficient test coverage through the public methods. Of course, this is a guideline not a law so there will be times that you might want to test a private method. In those instances, you can create an accessor and use reflection to invoke the private method.
Another recommendation that I would make is to use unit testing and code coverage in tandem. Code coverage can be a useful heuristic to identify when you need more tests. Lack of coverage should be used as a guide to indicate where more testing may be needed. This isn't to say that you need 100% coverage -- some code may be simple enough not to warrant a unit test (automatic properties, for instance) and they may not be touched by your existing tests.
There were a couple of issues that I had with the article. Probably the biggest is the lack of abstraction away from the database for unit tests. There probably are some integration tests that need to go against the db -- perhaps when testing trigger or constraint functionality if you can't convince yourself of their correctness otherwise. In general, though, I'm of the opinion that you should implement your data access as interfaces, then mock out the actual implementations in your unit tests so that there is no need to actually connect to the database. I find that my tests run faster, and thus I run them more often when I do this. Building up a "fake" database interface might take a little while but can be reused as long as you stick with the same design pattern for your data access.
Lastly, I would recommend using nUnit with TestDriven.Net - a very useful plugin whether you're doing nUnit or MSTest. Makes it very handy to run or debug tests with a right-click context menu.
My instincts tell me that the closer a
test is to the code being tested the
more likely you are to get complete
test coverage. In the comments to
Scott's blog posting is a remark that
just testing the public interface is
considered best practice by some, but
I would argue the test framework is
not a typical class consumer.
If your code cannot be tested using only public entry points, then you have a design problem. You should read more about TDD and SOLID principles (especially single responsibility principle and dependency inversion). Then you will understand that following TDD approach will help you write more testable, flexible and maintainable code, without the need for using such "hacks" as testing classes' private parts.
I also highly recommend reading Google's guide to testability by Miško Hevery, it has plenty of code samples which cover these topics.
I'm in a fairly similar situation and this question describes what I do keep-your-source-close-and-your-unit-tests-closer. There weren't too many others enamoured with my approach but it works perfectly for me.
I know there's a lot of stuff on TDD and i'm trying to pick up the practice too.
But i wonder is it a good idea to to TDD your bug fix too?
I was thinking along the lines of find the bug and narrow it down.
Write unit test to ensure that it will now pass whatever problem that it previously caused.
Write more unit test for other breakable conditions.
And finally write unit test to test for Integration test, as we don't have any unit test prior to this, so whenever i'm fixing a bug i'm always worried that i might accidentally break something.
So is TDD suitable for debugging
too?
Or is
there other method or resource/book
that would be more useful for this
purpose?
I am more concerned for the
Integration Test part as i've
mentioned above. I'm looking for
anything in LAMP/Axkit/Perl related
method.
Thanks.
Write a test that fails because of
the bug.
Fix the bug in your code.
Re-run test.
Re-run all tests.
So - yes - very much so.
When a tester finds a bug I usually write a unit test for it. When the test succeeds, I know the bug is fixed and it will always be covered again in the future.
When you have a bug in your system it is good practice in TDD to write a test that spots the bug (i.e. a red test that proves the bug). When you have to fix the bug, the test should eventually turn green. It may be a good idea to ferret out any other bugs if they're close enough.
Regarding debugging, TDD should be used to leverage debugging sessions away from the programmer. You can still debug, if you have no idea where some bug is, but it's easier to pinpoint a bug if you have a regression test suite with enough granularity.
You have to remember though that TDD is more about unit testing and not about integration testing. There is nothing wrong with writing integration tests since they're a sanity check to make sure your application or system under test works.
One book that is about testing with xUnit frameworks is the xUnit Patterns book that is general enough to work with any unit testing framework (even for PerlUnit I would guess) a cookbook of interesting tricks you can do with xUnit frameworks.
UPDATE:
There is a chapter about unit testing in perl at Extreme Perl.
The answer in short is yes. The basic structure of doing that is to write a test case which would simulate the bug and fail the test case. Then fix the bug which would pass the test case.
Yes. Of course all the tests you performed during TDD of your release will have been added to a regression test suite. But, in the case of a bug, that regression suite obviously wasn't detailed enough.
The first step in fixing a bug is replicating it and this is TDD. Once you find a test case that replicates the bug, you can expand on it if you wish (to catch other obvious problems of the same class), but I don't tend to do a lot of expansion since we have specific turnaround times for fixing a single bug.
Once you have a fix for that bug, add the test case to the regression suite. The idea is to keep adding test cases to the regression suite for both releases and bug fixes to give it very good coverage.
I always wrote tests before the actual code while fixing a bug.
This way I had the example of what to expect from a working code - and I could just focus on making this test (and all others, for regression) pass.
Yes but beware, If you write as many bugs as I do you will soon have a huge set of tests to cover all the bugs you have written and then fixed.
This will mean tests runs will be slower, and intent of behaviour will become muddied by your bug tests.
You could keep these tests logically separate or revisit your original set of specified behaviour checks (read tests) to see if you really have covered all your expected behaviour.
I think its important to differentiate between the two.
I think you'd fix the bug initially and then the regression tests would be composed of TDD.