Sep 82008
Confidently unit testing code that relies on a database is troublesome. There are many different approaches and many opinions on the matter. I've previously written about using transactions to safely rollback the database after every test. This worked well for me at the time, but there are pitfalls in using that approach. Recently, I've done some more research on the subject and even attempted, without success, to integrate DBUnit into my database reliant unit tests. After some trial and error, I landed on the following approach that is a culmination of a few ideas I'd previously heard of, but never tried. The goal of this approach is to:
- Run tests against a known data set.
- Reset the entire database prior to each test.
The example code that follows requires a test database, test database user and a test datasource targeting the test database.
Read more...
Aug 282008
Like the title says, this post will include code samples for setting up an
MXUnit test case using
ColdSpring and
ColdMock. The goal of any test case are to isolate the component under test as much as possible. I find ColdMock to be a simple and powerful tool for achieving test isolation. For this example, I created a test case for my ConfigFactory which I mentioned in a
previous post. TheConfigFactory component has a constructor dependency on the
Environment Config component developed by Rolando Lopez which makes this a good case for mocking.
Here's a snapshot of the files used in this sample test setup,

A common convention for test cases is to add the suffix "Test" to match the test case to the component being tested. I have my own convention for setting up tests withColdSpring bean definitions which I keep in a similarly named xml file. All test cases extend the BaseTestCase which contains a couple of methods to simplify test configuration with ColdSpring. The full code of the of the BaseTestCase follows.
<cfcomponent displayname="tests.BaseTestCase" extends="mxunit.framework.TestCase" output="false">
<cfset variables.beansXML = "">
<cffunction name="setBeanFactory" access="private" output="false" returntype="void">
<cfargument name="beansXML" type="string" required="true">
<cfargument name="params" type="struct" required="false" default="#structnew()#">
<cfscript>
if ((not structkeyExists(request,"beanFactory")) or (comparenocase(variables.beansXML,arguments.beansXML) neq 0))
{
variables.beansXML = arguments.beansXML;
request.beanFactory = createObject("component" ,"coldspring.beans.DefaultXmlBeanFactory").init(StructNew(),arguments.params);
request.beanFactory.loadBeans(variables.beansXML);
}
</cfscript>
</cffunction>
<cffunction name="getBeanFactory" access="private" output="false" returntype="any">
<cfreturn request.beanFactory>
</cffunction>
</cfcomponent>
In the past I kept my ColdSpring reference in variables scope, but I found that when testing components based that rely on Transfer ORM, I can shave the time of tests considerably when I have multiple tests in the same test case. Whether you run the MXUnit HttpAntRunner, the eclipse plugin or a manually configured test suite within a browser they all run as a single request, so as you define more tests it helps to speed things up.
Here's what the ConfigFactoryTest.xml looks like.
<beans>
<bean id="mockFactory" class="tests.MockFactory" singleton="true" />
<bean id="EnvironmentConfig" factory-bean="MockFactory" factory-method="createMock">
<constructor-arg name="objectToMock">
<value>model.Environment</value>
</constructor-arg>
</bean>
<bean id="ConfigFactory" class="model.ConfigFactory">
<constructor-arg name="hostName">
<value>${hostName}</value>
</constructor-arg>
<constructor-arg name="environmentConfig">
<ref bean="EnvironmentConfig" />
</constructor-arg>
</bean>
</beans>
You can see how easy it is to mock the dependency using the ColdMock MockFactory. Below is the full code for my ConfigFacotry test case. The important thing to note is that the component display name is used to resolve the path to the ColdSpring bean definitions used in the test suite.
<cfcomponent displayname="tests.ConfigFactoryTest" extends="tests.BaseTestCase">
<cffunction name="setUp" access="public" returntype="void">
<cfset var beanDefFileLocation = expandPath('/' & Replace(GetMetadata(this).displayname,'.','/','all') & '.xml')>
<cfset var params = Structnew()>
<cfset params.hostName = "www.somedomain.com">
<cfset setBeanFactory(beanDefFileLocation,params)>
</cffunction>
<cffunction name="testGetSetting" access="public" returntype="void">
<cfscript>
var configFactory = "";
var env = getBeanFactory().getBean("EnvironmentConfig");
var settings = structNew();
settings["MyString"] = "my string";
env.mockMethod('getEnvironmentByUrl').returns(settings);
configFactory = getBeanFactory().getBean("ConfigFactory");
assertTrue(configFactory.getSetting("MyString") eq settings["MyString"]);
</cfscript>
</cffunction>
<cffunction name="testGetAllSettings" access="public" returntype="void">
<cfscript>
var configFactory = "";
var env = getBeanFactory().getBean("EnvironmentConfig");
var settings = structNew();
settings["MyString"] = "my string";
env.mockMethod('getEnvironmentByUrl').returns(settings);
configFactory = getBeanFactory().getBean("ConfigFactory");
assertTrue(StructCount(configFactory.getAllSettings()) eq 1);
</cfscript>
</cffunction>
<cffunction name="testOnMissingMethod" access="public" returntype="void">
<cfscript>
var configFactory = "";
var env = getBeanFactory().getBean("EnvironmentConfig");
var settings = structNew();
settings["MyString"] = "my string";
env.mockMethod('getEnvironmentByUrl').returns(settings);
configFactory = getBeanFactory().getBean("ConfigFactory");
assertTrue(configFactory.getMyString() eq settings["MyString"]);
</cfscript>
</cffunction>
</cfcomponent>
Most of my tests aren't this involved. Since my ConfigFactory has a constructor dependency on EnvironmentConfig, I pull it out of ColdSpring first, mock the getEnvironmentByUrl method to return a known structure before requesting the ConfigFactory. Truly powerful stuff!
I have attached a zip file of this sample test bed as an enclosure for folks to try out.
Aug 272008
As I delve deeper into Test Driven Development (TDD), I have refined my development setup in order to feel confident that my tests are relevant and that they provide immediate feedback about the status of my working code. The two most important goals that I have tried to accomplish with my revisions are to follow the credo of never committing broken code and to isolate and reset the database to ensure the integrity of test data. The following examples are not solely related to TDD, but represent an accumulation of development "best practices" gathered from many sources. I'm going to admit straight up that I am not a TDD purist. I almost always write model components before tests. But I always test before running code on the client side. One of the biggest selling points about unit testing and for me is that I can debug faster should a problem arise and I can do it repeatedly and confidently. Working this way makes me feel somewhat pragmatic. I'm willing to spend a known amount of time setting up and running tests so that I can avoid spending an unknown amount of time debugging. First up, my application directory structure and Apache virtual host settings.
Read more...
Feb 22008
If you wonder about
Unit Testing and
Test Driven Development, but are afraid of commitment, do yourself a favour, read Terrance Ryan's thoughts on "
How to convince yourself to Unit Test". A couple of months ago, I made a commitment to start unit testing. I'll admit that it took a little while to wrap my head around the terminology and process. A good practice point for me was to run some of my first CFCs though Seth Petry-Johnson's
var scope checker. When I discovered a plethora of un-scoped vars, I used the opportunity to write tests against the methods I was going to change. After only a few tests, I started to get this warm feeling of security. By the time my updates and tests were complete, I had a nice little suite that I could return to any time a change was required.
Read more...
Recent Comments