Mar 28 2011

Continuous Testing Plug-ins verses Continuous Integration

Category: Continuous Integration,Eclipse,TestingPhil @ 12:01 am

This is an extremely interesting concept, but I really wonder how it works in the real world. Toy projects are one thing, but real production size applications are a different story! The principle is very simple; just modify a piece of code in your project and the IDE automatically executes all of the unit tests. Pretty cool concept. But, what if your unit test suite runs for several hours? I believe this is an all too common problem. I have seen many projects dedicate time to building their unit tests, and over time, end up with test suites that can run for multiple hours. There is a very negative side effect associated with this scenario, it can change a developer’s attitude towards checking in code, it actually discourages it. If the developer knows their commits will not be validated for several hours, they tend to check in very infrequently and only when they feel it will not impact other developers. Many factors can contribute to this scenario. The primary factor is that many of these tests are not really unit tests; they typically do not provide the proper isolation (via mocking) and take too long to execute. I’m not going to address those issues directly, but the concept does support and encourage the idea of quick test execution.

Wouldn’t it be cool to run all of the unit tests, in your IDE before checking in the your changes? Obviously, this action is expected by the developer and is completely possible without any special tools. However, it is not automatic and requires extra effort by the developer. Usually, it is easier to let the Continuous Integration server run the tests and fix the fallout later. It seems like a simple problem to solve; create a plug-in that automatically executes all of the unit tests saving a file, requiring no additional effort or interaction from the developer. If you are an Eclipse user, you have two good options, JUnit Max and Infinitest. I would really like to try Kent Beck’s JUnit Max, but you need to purchase a subscription to use (and even try)  the plug-in. That is a show stopper for me, as there are free alternatives such as Infinitest. I evaluated the Infinitest plug-in last year, after it was open-sourced. I did encounter some problems with the plug-in, my Spring-based projects did not work; I was quickly discouraged and abandoned the tool. Months later, I was still intrigued by the idea, so I gave it a try last week on a different project and achieved success!

Unfortunately, Infinitest is not currently installable from the Eclipse Marketplace. It can be quickly installed the old fashioned way from the Eclipse update site.  There is only one setting you may want to tweak, the Slow Test Warning threshold. If  you have unit tests that execute for longer than half a second, Infinitest will add a warning to your Problems View for each one; this could be problematic if you have a large number of long running unit tests. You should also notice a huge new “Continuous Test” status bar in your Eclipse footer. It will update automatically based on your coding activity. JUnit Max is a little more subtle, but provides the same type of feedback mechanism. That’s it, you are now ready to write code and auto test!

I’m not exactly sure how intelligent these plug-ins are, related to determining which unit tests to execute based on what code was changed. Hopefully that is the next level of continuous testing evolution. However, if you modify a single unit test, Infinitest will only run that single test; beyond that, I don’t believe there is much intelligence. If your code modification causes any unit tests to fail, the status bar will turn red. Unfortunately, it does not tell you how many test failed, it only provides the number of tests which it executed. If you mouse over the status bar, it will display the names of the tests which failed, but you can not click on them, they are just there for feedback. You will need to look in the Problems View to discover which test broke, the view will now contain markers for each failed test.

The only issue I have with the plugin, is around user interaction. It does not provide its own View for test results, it simply adds the failures to the existing items in the Problems View. I assumed that the Unit Test View would be updated with the test results, similar to manually running a unit test; however, this was not the case.  It appears that both Infinitest and JUnit Max use the same strategy, providing minimal feedback to the developer. Maybe this a good approach, but I was expecting more visual feedback and confirmation. I associate static problems with the Problems View, such as compiler and code quality errors, not dynamic problems such as test failures. The developer should be able to click on a marker (problem) and be navigated to the exact line of code indicated by that marker, giving them the opportunity to resolve the issue .  Here is my problem with this approach, if the developer clicks on the failed JUnit marker, where will they be navigated? The only possible location is the unit test which failed. Generally speaking, this is not where the problem would be found, it would lie in the code that was just changed. Given this issue, I would rather use a different View for continuously executed tests;   maybe even better, find a way to reuse the existing Unit Test View.

This does bring up an interesting question, do these plug-ins minimize or eliminate the need for Continuous Integration? I don’t believe this is the case, but would seem to eliminate the need for some of the more advanced features the vendors are implementing. One such example is TeamCity’s remote build, pre-commit test feature. There are many factors to consider, but if unit tests execute quickly, I do not see the need for this functionality to be provided by the CI server. If the tests execute for multiple hours, we could be adding more risk into the process, by not committing changes when they are ready to integrate. Remote build, pre-commit testing sounds valuable on paper, but I am struggling to see the real benefits. I tend to think this functionality over complicates the Continuous Integration process. One of the main benefits of Continuous Integration is that it is seamless and automatic. Nothing is required by the development team to reap the CI benefits, simply check in your code early and often.  As more education and process overhead are required by the CI process, it is no longer simple nor free.  The focus should be on doing what makes sense, at the right time, in the right place. These IDE plug-ins seem to have correct perspective, pre-check-in on the developers desk, simple, automatic, and efficient.

I hope that you will try to integrate one of these plug-ins into your personal development process. I believe they will help create a more robust development process, creating fewer Continuous Integration failures and ultimately, a more efficient, faster executing test suite. If you have real world experience with one of these tools, please provide some comments. I am still curious, does this approach work on real projects?

3 Responses to “Continuous Testing Plug-ins verses Continuous Integration”

  1. Continuous Testing Plug-ins verses Continuous Integration says:

    […] are a different story! The principle is very simple; just modify a piece of code in your… [full post] Phil Soapbox Rants and Raves continuous integrationeclipsetesting 0 0 […]

  2. roseate says:

    In my opinion Infinitest is not precise enough: when you modify a method, Infinitest runs all the tests that have a reference to the class containing this method, even though it isn’t called anywhere. For me this would be the most relevant: to execute the minimum set of tests. I would prefer a basic block level solution where only those test methods run which affect the modified basic block. This solution would even decrease the number of executed tests. This could be the next step of continuous testing evolution.
    Another thing that gets me thinking: in case of a failure where should the developer be navigated by clicking on the marker of the failure? I don’t see better way than navigating to the failed assert. The other opportunity could be the modified source, but the location is not evident because a failure can come from several modifications as well.

  3. mrityunjoy says:

    Just using the reference or immediate calls, is ineffective. the most feasible alternative is to build code, running all the tests. the only alternative if you really want to do smart continuous testing, is using Dynamic Program slicing, based on Symbolic execution, that is referenced by the modified code. this is pretty advanced for the tools in market.

Leave a Reply