kmymoney/developer-doc/phb/unit-test.docbook

<chapter id="unit-test">
<title>Unit Testing</title>
<para>
If this project handbook would have been for a professional project (with
professional I mean, a project that people make money with), I would have
written
</para>

<caution>
<para>
<emphasis>Unit tests must be supplied by the developer
with the classes/source-code he checks into the repository!</emphasis>.
</para>
</caution>

<para>
Since this is
the handbook for a voluntary work (which is not less professional than any
other project), I replace the above sentence with
</para>

<note>
<para>
<emphasis>Each developer in this project is strongly encouraged to develop
unit tests for the code he or she develops and make them available to
 the project team!</emphasis>.
</para>
</note>

<sect1 id="why-unit-testing">
<title>Why unit testing?</title>

<para>
Before I can give an answer to this question, I should explain what unit
testing is about. I do not cover all relevant aspects here nor do I start a
discussion of the various aspects of unit testing. If you want to read more
about the details of unit testing, the philosophy behind it and about the
various tools available, please visit the project pages of JUnit and
<ulink url="http://cppunit.sourceforge.net/">CPPUNIT</ulink>.
The following explanation describes what unit testing is:
</para>

<para>
For each class developed in a project, an accompanying test container
is developed when the interface of
the class is defined but before the implementation of the class starts. The
test container consists out of testcases that perform all necessary tests
on the class while verifying the results. One or more of these test
containers (for more than one class) form a test suite.
</para>

<para>
Your might think, that it is strange to first define the interface, then
develop the tests and then start the development of the actual code, but it
has shown, that this approach has a couple of interesting side-effects:

<itemizedlist>
<listitem>
<para>
The developer spends time to think about how to test his implementation
before he actually works on the implementation. This leads to the fact,
that while working on the implementation, he already knows how his code
will be tested.
</para>
</listitem>

<listitem>
<para>
A clear definition of the <emphasis>end of implementation</emphasis> exists
due to the fact, that the testcases will all fail before the beginning of
the implementation phase. Once implementation proceeds, more and more
testcases will pass. When they all pass, the development is finished.
</para>
</listitem>

<listitem>
<para>
Since the tests will run automated and will be re-run very often during the
development cycle, a lot of problems will be caught very early on. This
reduces the number of problems found during integration of the project.
Believe me, there will be plenty left!
</para>
</listitem>
</itemizedlist>
</para>

<para>
Now, the list of all these side-effects is actually the answer to the
question <emphasis>Why unit testing?</emphasis> or does anyone have a
argument against it? I agree, that in some cases automated unit testing is
hard to achieve (e.g. for GUI related code) but I found, that whenever it
is possible to introduce automated unit tests, the benefit is huge.
</para>
</sect1>

<sect1 id="unit-testing-in-kmm">
<title>Unit testing in &app;</title>
<para>
Just about the time, when the &app; project underwent a radical change of
it's inner business logic (the KMyMoney engine), I read an article about
the existance of a unit test container for C++ projects named 
<ulink url="http://cppunit.sourceforge.net/">CPPUNIT</ulink>.
In
discussions with my colleagues at work, I got the impression, that this
would be something worth to look into. So I sat down and wrote the first
test cases for existing code to get a feeling for what is required.
</para>

<para>
I found it annoying to write test cases for code that existed and was
believed to work (version 0.4 of the project). When the decission was made
to start with the 0.5 development branch, I started working on the new
engine code that should introduce a clear interface between the business
logic and the user interface. Another design goal was to write the engine
in such a way, that it is not based on any KDE code which the old one was.
The second goal to free it from Qt based code was not that easy and was
skipped by the project team at that time.
</para>

<para>
Even if it was hard for me at first to follow the above laid out principle
to design the interface, write the test code and then start with the
implementation, I followed this trail. It has proven to be very valuable.
Once the interface was designed, I started thinking in a different manner:
How can I get this class to fail? What strange things could I do to the
code from the outside? Those were the design drivers for the test code. And
in fact, this thinking changed the way I actually implemented the code, as
I knew there was something that would check all these things over and over
again automatically.
</para>

<para>
A lot of code was implemented and when I was almost done with the first
shot of the implementation, discussion came up on the developers mailing
list about a feature called <emphasis>double entry accounting</emphasis>
that was requested for &app; by a few people. The engine I wrote up to that
point in time did not cover the aspects of double entry accounting at all,
though a few things matched. After some time of discussions, we became a
better understanding of the matter and I changed the code to cover double
entry accounting. Some of the classes remained as they were, others had to
be adopted and yet others rewritten entirely. The testcode had to be
changed as well due to the change in the interfaces, but not the logic
of the tests. Most of the thoughts how to uncover flaws remained.
</para>

<para>
And that is another reason, why unit testing is so useful: You can change
your internal implementation and still get a feeling, if your code is
working or not. And believe me: even if some changes are small, one usually
oversees a little side-effect here and there. If one has good unit tests
this is not a problem anymore, as those side-effects will be uncovered and
tested.
</para>

<para>
During the course of implementing the engine, I wrote more than 100
testcases. Each testcase sets up a testenvironment for the class and tests
various parameters against the class' methods in this environment in so
called test steps.
Exceptions are also tested to be thrown. The testcases handle unexpected
exceptions as well as expected exceptions that do not occur.
</para>
</sect1>

<sect1 id="unit-testing-howto">
<title>Unit testing HOWTO</title>
<para>
This section of the developer handbook should give some examples on how to
organize test cases and how to setup a test environment.
</para>

<para>
My examples will all be based on the code of the &app; engine found in the
subdirectory <command>kmymoney2/kmymoney2/mymoney</command> and it's
 subdirectory <command>storage</command>. A
single executable exists that contains all the test code for the engine.
It's called <command>autotest</command> and resides in the mymoney
subdirectory.
</para>

<sect2 id="unit-test-integration">
<title>Integration of CPPUNIT into the &app; project</title>
<para>
The information included in the following is based on version 1.8.0 of
CPPUNIT. The &app; build system has been enhanced to check for it's
presence. Certain definitions setup by
<emphasis>automake/configure</emphasis> allow to compile the project
without unit testing support.
<caution>
<para>
This is not the recommended way for developers!
</para>
</caution>
</para>

<para>
If code within test environments is specific to the presence of CPPUNIT it
can be included in the following #ifdef primitive:

<screen>

#ifdef HAVE_LIBCPPUNIT
// specific code that should only be compiled,
// if CPPUNIT >= 1.8.0 is present
#endif


</screen>
For an example see the
<link linkend="test-container-example">Unit Test Container Source File
Example</link>.
</para>

<para>
The same applies for directives that are used in
<command>Makefile.am</command> files. The primitive to be used there is as
follows:

<screen>

if CPPUNIT

# include automake-directives here, that should be evaluated
# only, when CPPUNIT is present

else

# include automake directives here, that should be evaluated
# only, when CPPUNIT is not present.

endif


</screen>
For an example see <command>kmymoney2/mymoney/Makefile.am</command>.
</para>
</sect2>

<sect2 id="unit-test-naming">
<title>Naming conventions</title>
<para>
The test containers are also classes. Throughout CPPUNIT, the test
containers are referred to as <emphasis>test fixtures</emphasis>. In the
following, I use both terms.
For a given class <emphasis>MyMoneyAbc</emphasis>, which
resides in the files <command>mymoneyabc.h</command> and
<command>mymoneyabc.cpp</command>, 
the test container is named <emphasis>MyMoneyAbcTest</emphasis> and resides
in the files 
<command>mymoneyabctest.h</command> and
<command>mymoneyabctest.cpp</command> in the same directory.
The test container must be derived
publicaly from <command>CppUnit::TestFixture</command>.
Each testcase is given a
descriptive name (e.g. EmptyConstructor) and I found it useful to prefix
this name with the literal 'test' resulting into something like
testEmptyConstructor.
</para>

</sect2>
<sect2 id="unit-test-includes">
<title>Necessary include files</title>
<para>
In order to use the functionality provided by CPPUNIT, one has to include
some information provided with CPPUNIT in the test environment. This is
done with the following include primitive as one of the first things in the
header file of the test case container (e.g. mymoneyabctest.h):

<screen>

#include &lt;cppunit/extensions/HelperMacros.h&gt;

</screen>
</para>
</sect2>

<sect2 id="unit-test-private">
<title>Accessing private members</title>
<para>
For the verification process it is sometimes necessary to look at some
internal states of the object under test. Usually, all this information is
declared private in the class and only accessible through setter and getter
methods. Cases exist, where these methods are not implemented on purpose
and thus accessing the information from the test container is not possible. 
</para>

<para>
Various mechanism have been developed all with pros and cons. Throughout
the test containers I wrote, I used the method of redefining the specifier
<emphasis>private</emphasis> through <emphasis>public</emphasis> but only
for the time when reading the header file of the object under test. This can
easily be done by the C++ preprocessor. The following example shows how to
do this:

<screen>

#define private public
#include "mymoneyabc.h"
#undef private


</screen>

The same applies to protected members. Just add a line containing
<emphasis>#define protected public</emphasis> before including the class
definition and a line containing <emphasis>#undef protected</emphasis>
right after the inclusion line.
</para>
</sect2>

<sect2 id="unit-test-methods">
<title>Standard methods for each testcase</title>
<para>
Three methods must exist for each test fixture. These are a default
constructor, setUp and tearDown. I think, it is not necessary to explain
the default constructor here. SetUp and tearDown have a special function
within the test cases. setUp() will be called before the execution of any
test case in the test fixture. tearDown() will be called after the execution
of the test case, no matter if the test case passes or fails. Thus setUp()
is used to perform initialization necessary for each test case in the
fixture and tearDown() is used to clean things up. setUp() and tearDown()
should be written in such a way, that all objects created
through a test case should be removed by tearDown(), i.e. the environment
is restored exactly to the state it was before the call to setUp().

<note>
<para>
This is not always the case within the testcase for &app;. Espacially when
using a database as the permanent storage things have to be overhauled for
e.g. MyMoneyFileTest.
</para>
</note>

</para>

<para>
CPPUNIT comes with a set of macros that help writing testcases. I cover
them here briefly. If you wish a more detailed description, please visit
the
<ulink url="http://cppunit.sourceforge.net/">CPPUNIT</ulink> project
homepage.
</para>
</sect2>

<sect2 id="test-macro-assert">
<title>CPPUNIT_ASSERT</title>
<para>
This is the macro used at most throughout the test cases. It checks, that a
given assumption is true. If it is not, the test case fails and a
respective message will be printed at the end of the testrun.
</para>

<para>
CPPUNIT_ASSERT has a single argument which is a boolean expression. The
expression must be true in order to pass the test. If it is false, the test
case fails and no more code of the test case is executed. The following
example shows how the macro is used:
<screen>

  int a, b;
  a = 0, b = 1;
  CPPUNIT_ASSERT(a != b);
  a = 1;
  CPPUNIT_ASSERT(a == b);


</screen>
The example shows, how two test steps are combined. One checks the
inequality of two integers, one the equality of them. If either one does
not work, the test case fails.
</para>
<para>
See the
<link linkend="test-source-example">Unit Test Source File Example</link>
for a demonstration of it's use.
</para>

</sect2>

<sect2 id="test-macro-fail">
<title>CPPUNIT_FAIL</title>
<para>
This is the macro used when the execution of a test case reaches a point it
should not. This usually happens, if exceptions are thrown or not thrown.
</para>

<para>
CPPUNIT_FAIL has a single argument which is the error message to be
displayed.  The following example shows how the macro is used:
<screen>

  int a = 1, b = 0;
  try {
    a = a / b;
    CPPUNIT_FAIL("Expected exception missing!");
  } catch (exception *e) {
    delete e;
  }

  try {
    a = a / a;
  } catch (exception *e) {
    delete e;
    CPPUNIT_FAIL("Unexpected exception!");
  }
  

</screen>
The example shows, how two test steps are combined. One checks the
occurance of an exception, the other one that no exception is thrown.
If either one does not work, the test case fails.
</para>

</sect2>

<sect2 id="test-macro-testsuite-start">
<title>CPPUNIT_TEST_SUITE</title>
<para>
This macro is used as the first thing in the declaration of the test fixture.
 A single argument is the name of the class for the test fixture. It starts
the list of test cases in this fixture defined by the 
<link linkend="test-macro-testcase">CPPUNIT_TEST</link> macro. The list must be
terminated using the <link
linkend="test-macro-testsuite-end">CPPUNIT_TEST_SUITE_END</link> macro.
</para>
<para>
See the
<link linkend="test-header-example">Unit Test Header File Example</link>
for a demonstration of it's use.
</para>
</sect2>

<sect2 id="test-macro-testsuite-end">
<title>CPPUNIT_TEST_SUITE_END</title>
<para>
This macro terminates the list of test cases in a test fixture. It has no
arguments.
</para>
<para>
See the
<link linkend="test-header-example">Unit Test Header File Example</link>
for a demonstration of it's use.
</para>
</sect2>

<sect2 id="test-macro-testcase">
<title>CPPUNIT_TEST</title>
<para>
This macro defines a new test case within a test fixture. As argument it
takes the name of the test case.
</para>
<para>
See the
<link linkend="test-header-example">Unit Test Header File Example</link>
for a demonstration of it's use.
</para>
</sect2>

<sect2 id="test-macro-registration">
<title>CPPUNIT_TEST_SUITE_REGISTRATION</title>
<para>
This macro registers a test fixture within a test suite. It takes the name
of the test fixture as argument.
</para>
<para>
See the
<link linkend="test-container-example">Unit Test Container Source File
Example</link>
for a demonstration of it's use.
</para>
</sect2>

</sect1>
</chapter>