TestPage

Zope Unit Testing

Zope Testing

If you encounter a directory named "tests" in a package within within the Zope source code, it most likely indicates that the directory contains test code used to ensure that the code owned by the package works as it was designed. Many of the test scripts contained within Zope "tests" directories will be scripts which contain "unit tests".

What Unit Tests Are

A "unit" may be defined as a piece of code with a single intended purpose. A "unit test" is defined as a piece of code which exists to codify the intended behavior of a unit and to compare its intended behavior against its actual behavior.

Unit tests are a way for developers and quality assurance engineers to quickly ascertain whether independent units of code are working as expected. Unit tests are generally written at the same time as the code they are intended to test. A unit testing framework allows a collection of unit tests to be run without human intervention, producing a minimum of output if all the tests in the collection are successful.

What Unit Tests Are Not

It's very useful to define unit tests in terms of what they are not. From the "Extreme Programming Enthusiast" website (http://c2.com/cgi/wiki?UnitTestsDefined):

   Unit tests are not:

    - Manually operated.

    - Automated screen-driver tests that simulate user input (these
      are "functional tests").

    - Interactive.  They run "no questions asked."

    - Coupled.  They run without dependencies except those native to
      the thing being tested.

    - Complicated.  Unit test code is typically straightforward
      procedural code that simulates an event.

Unit Testing Frameworks

A unit testing framework is generally employed to collect related unit tests together in order to make it easier to run them as a group. When used with a unit testing framework, unit tests live outside of the modules of code they're intended to test.

How Unit Tests Help In The Development Process

Unit tests should be written at the same time the code they test is written. A short, healthy cycle of "code/write test/run test/repeat" can help a developer code more quickly by reducing "backtracking" effort and by helping the developer focus on the actual problem at hand. Additonally, the unit tests generated at initial development time can serve as later assurance that maintenance and refactoring performed on code does not break any of its intended functionality or behavior. The results of unit tests may additionally be used as a metric by quality assurance personnel along with the results of other tests to gauge code quality before before it is "shipped."

Basic Unit Testing Philosophies

Write unit tests at the same time that you write the code. Nothing's worse than being faced with the prospect of writing tests against a huge chunk of existing code because you "have to." Writing unit tests post-facto can be boring and also robs you of the main benefits that unit testing can provide. Writing unit tests at the same time you write the code helps you focus on the task at hand. Writing unit tests in conjuction with code can be fun and satisfying, and can help you improve the quality of your code while its goals are fresh in your mind. Used properly, unit testing may also help you write code faster, because you will need to "backtrack" less. Some "Extreme Programming" enthusiasts posit that you should write a test before you write its associated code, and then develop the code until it passes the test.

Unit tests should be developed against as small and specific a subset of a system's or subsystem's functionality as possible. For instance, a one unit test may test that a unique id generator produces ids of a specific length, while another unit test in the same group may ensure that a thousand ids from the same unique id generator do not contain the same value. Writing a single unit test which tests a broad swath of a system's capabilities is counterproductive, because it does not allow for a fine enough granularity when attempting to figure out "what went wrong," requiring the developer to "backtrack". Unit testing is capable of helping to help reduce backtracking, but only if used properly.

A unit test does not produce any output unless it fails. If a unit test fails, it should print something useful, but short. A unit test should never fill the screen with output or otherwise produce output that needs to be manually examined for "clues". This is the realm of other testing methodologies. If unit tests are written at sufficiently granular level, it is often enough just to know the name of the unit test that failed.

"It is better to have tried to test and failed than to not have tried to test at all" aka "test the riskist things first." If the prospect of writing a series of unit tests for an existing system seems daunting, it's important to remember that no matter how many unit tests you write, you cannot prove that your software does not have bugs. Therefore, you cannot possibly test every case while developing. You should plan to write tests against code based on the risks involved in not testing that code. Don't feel that you need to write a test case for every "corner case" (although do try to hit the riskiest ones). In the worst case, it's better to have a test module with one lonely unit test in it than none at all.

"Test fixtures" are employed by unit tests. Test fixtures are bits of state and environment that allow the unit test to perform its job properly. An example of a test fixture might be a file, an instance of a class, or a row in a database table. Any part of the environment needed by a unit test besides the unit testing framework itself is considered a test fixture.

In general, the common fixtures required by individual tests in a testing group should be more or less identical. If the fixtures needed by a single test are radically different than the fixtures required by the rest of the tests, or if the test does not require the fixtures provided to the other tests, it probably belongs in another (or its own) group of tests.

When a unit test in a group modifies the state of a test fixture, the test fixture should be restored to a known state before the next unit test in the group is run. Often, this means "rolling back" changes to a transactional database or restoring the state of a string so the next unit test can inherit a known state of the same fixtures. The unit testing framework has capabilities which allow you to automate most of this work by providing a "set up" method which gets called before each test is run and a "tear down" method that gets called after a test is finished.

Unit tests should play nicely with the unit testing framework they employ. Unit tests should not call "sys.exit()" or do similar things which effect their ability to be run as part of a group of tests. The testing framework attempts to deal with misbehaved unit tests, but it's better just to make them behave nicely in the first place.

Unit tests should "grow" with the code that they're intended to test. For example, if a group of unit tests is intended to verify the behavior of all of the routines within a module, additional unit tests should be added to the test group when new functionality is added to that module.

Writing Unit Tests For The Zope Core

If you're writing core code, you probably don't need to listen to any more of this. :-) The rules for writing tests for Zope core code are simple:

• The testing code should make use of PyUnit? (/lib/python/unittest.py). Instructions for using PyUnit? are available at http://pyunit.sourceforge.net.
• Tests must be placed in a "tests" subdirectory of the package or directory in which the core code you're testing lives.
• Test modules should be named something which represents the functionality they test, and should begin with the prefix "test." E.g., a test module for BTree? should be named testBTree.py.
• An individual test module should take no longer than 60 seconds to complete.

Writing Unit Tests For Applications Based On Zope

Zope uses the PyUnit? unit testing framework, the documentation for which is available at http://pyunit.sourceforge.net. The lib/python/unittest.py module is the framework. You should establish your own conventions for naming and placement of test modules.

Writing unit tests against applications based on Zope can be difficult. Zope is a collection of related modules, some with non-trivial interdependencies. Running its code successfully also in some cases depends on state provided only in the context of a web request, so calling Zope methods directly may not work as you expect. If you're not intimately familiar with Zope, implementing unit tests can be fustrating. For example, for the common case, before you are able to write code which tests a Zope SQL Method, you must establish a test fixture which represents your entire Zope site.

Luckily, some tools are at your disposal to make writing unit tests against Zope components and applications easier by making the creation of these fixtures easier.

Surprisingly, one of the most effective tools for facilitating unit testing is ZEO (http://www.zope.org/Products/ZEO). ZEO is an open-source clustering solution for Zope which makes it possible to front-end a single "storage server" which manages a Zope object database with multiple Zope clients that run a "client storage". The reason ZEO is interesting for unit testing is mostly an unintended side-effect of how it works as compared to Zope without ZEO. Zope without ZEO commonly uses a "FileStorage?" to hold its object database. When Zope is started with a FileStorage?, the FileStorage? code processes an "index" file. This takes time. Zope using a ClientStorage? as with ZEO does not process an index file, making startup faster. Fast startup of Zope is critical to effective unit testing. It is recommended that you implement ZEO if you're heavy in to unit testing, as it really speeds things up. It's not strictly required, however.

Making Zope itself into a test fixture is straightforward. Your test code must:

• add the lib/python directory of your Zope installation to the PYTHONPATH (via sys.path.insert())
• import ZODB (may not be required, but just put it in for good measure)
• import Zope
• app = Zope.app()
• operate on the Zope instance space by calling methods from the root object (bound to app), e.g.:

  app.acl_users.manage_addUser()

• a transaction will not be committed to the Zope object database until you call the global function "get_transaction().commit()", so changes made to the ZODB are not visible to succeeding tests unless that function is called explicitly.
• As a part of your tearDown, make sure to call "app._p_jar.close()". This closes the database connection cleanly.

For more information on operating on Zope programatically by "importing" it, see Michel Pelletier's "The Debugger Is Your Friend" at http://www.zope.org/Members/michel/HowTos/TheDebuggerIsYourFriend

Sometimes, just importing Zope isn't enough. For example, it's often not possible to obtain the results of a DTML or Python method by simply calling it from your running code without doing lots of trickery by establishing fake REQUEST object to pass in to it, as Zope expects a REQUEST object to exist in the context of a web request (which a DTML method is somewhat logically designed to serve).

If you find yourself getting bogged down while writing unit tests by Zope's refusal to run certain methods due to missing state (like REQUEST), it's useful to know about the "debug" method of the Zope package. This method allows you to simulate a web request, which generally provides all the state necessary to run methods which depend on web requests, and returns the results of the web request as it would be seen in by a web browser. To use the Zope debug method, do the following:

• add the lib/python path to your PYTHONPATH (via sys.path.insert())
• import ZODB
• import Zope
• Zope.debug(/a/url/representing/a/method?with=a?couple=arguments', u=username:password, s=silent, e={'some':environment, 'variable':'settings'})

The "silent" option causes Zope not to print anything. You can set your python's stdout to a file or a file-like object to capture the output if you do not set the silent flag.

In Zope versions before 2.2.2, all calls to Zope.debug commit the transaction represented by the call to Zope.debug by default. This can pose problems for unit testing, as the state of the environment should not be effected by prior tests. A solution should be available by the release of Zope 2.3.

Administrivia

Unit test scripts found in the Zope source code currently make use of the PyUnit? unit testing framework, available from http://pyunit.sourceforge.net, written by Stephen Purcell (thanks Stephen!). PyUnit? is based on the JUnit? testing framework for Java (written by Kent Beck and Erich Gamma), which in turn was based on a testing framework designed for Smalltalk (also written by Kent Beck).

Unit testing is a primary tenet of "Extreme Programming", a software development methodology designed to faciliate the rapid production of high quality code with a minimum of developmental ceremony. For more information on unit tests as they relate to Extreme Programming, see http://c2.com/cgi/wiki?UnitTestsDefined. Although Digital Creations has not embraced the entire spectrum of Extreme Programming methodologies in its software development process, we've found unit tests a way to speed development and produce higher-quality code.

There is a testing support package in lib/python/Testing that might be useful to you for writing unit tests. See the README.txt in that directory for details.