CxxTest User's Guide

1 Introduction

CxxTest is a JUnit/CppUnit/xUnit-like framework for C++.

Its advantages over existing alternatives are that it:

• Doesn't require RTTI
• Doesn't require member template functions
• Doesn't require exception handling
• Doesn't require any external libraries (including memory management, file/console I/O, graphics libraries)

In addition, CxxTest is slightly easier to use than the C++ alternatives, since you don't need to "register" your tests. It also features some extras like a richer set of assertions and even support for a "to do" list (see TS_WARN()).

CxxTest is available under the GNU Lesser General Public License.

1.1 About this guide

This guide is not intended as an introduction to Extreme Progamming and/or unit testing. It describes the design and usage of CxxTest.

2 Getting started

2.1 Getting CxxTest

The homepage for CxxTest is http://cxxtest.sourceforge.net. You can always get the latest release from the SourceForge download page, here or here. The latest version of this guide is available online at http://cxxtest.sourceforge.net/guide.html. A PDF version is also available at http://cxxtest.sourceforge.net/guide.pdf.

2.2 Your first test!

Here's a simple step-by-step guide:

1. Tests are organized into "Test Suites". Test suites are written in header files.

   A test suite is a class that inherits from CxxTest::TestSuite. A test is a public void (void) member function of that class whose name starts with test, e.g. testDirectoryScanner(), test_cool_feature() and even TestImportantBugFix().

             
             // MyTestSuite.h
             #include <cxxtest/TestSuite.h>
             
             class MyTestSuite : public CxxTest::TestSuite 
             {
             public:
                void testAddition( void )
                {
                   TS_ASSERT( 1 + 1 > 1 );
                   TS_ASSERT_EQUALS( 1 + 1, 2 );
                }
             };
             
             

2. After you have your test suites, you use CxxTest to generate a "test runner" source file:

        # cxxtestgen.pl --error-printer -o runner.cpp MyTestSuite.h
        

   or, for those less fortunate:

        C:\tmp> perl -w cxxtestgen.pl --error-printer -o runner.cpp MyTestSuite.h
        

3. You would then simply compile the resulting file:

        # g++ -o runner runner.cpp
        

   or perhaps

        C:\tmp> cl -GX -o runner.exe runner.cpp
        

   or maybe even

        C:\tmp> bcc32 -erunner.exe runner.cpp
        

4. Finally, you run the tests and enjoy a well tested piece of software:

        # ./runner
        Running 1 test.OK!
        

2.3 Your second test

Now let's see what failed tests look like. We will add a failing test to the previous example:

     
     // MyTestSuite.h
     #include <cxxtest/TestSuite.h>
     
     class MyTestSuite : public CxxTest::TestSuite 
     {
     public:
        void testAddition( void )
        {
           TS_ASSERT( 1 + 1 > 1 );
           TS_ASSERT_EQUALS( 1 + 1, 2 );
        }
     
        void testMultiplication( void )
        {
           TS_ASSERT_EQUALS( 2 * 2, 5 );
        }
     };
     
     

Generate, compile and run the test runner, and you will get this:

     
     # ./runner
     Running 2 tests.
     MyTestSuite.h:15: Expected (2 * 2 == 5), found (4 != 5)
     Failed 1 of 2 tests
     Success rate: 50%
     
     

Fixing the bug is left as an excercise to the reader.

2.4 Graphical user interface

• Under Windows with Visual C++, run perl cxxtestgen.pl -o runner.cpp --gui=Win32Gui MyTestSuite.h.
• Under X-Windows, try ./cxxtestgen.pl -o runner.cpp --gui=X11Gui MyTestSuite. You may need to tell the compiler where to find X, usually something like g++ -o runner -L/usr/X11R6/lib runner.cpp -lX11.
• If you have Qt installed, try running cxxtestgen.pl with the option --gui=QtGui. As always, compile and link the Qt headers and libraries.

See Graphical user interface and Running the samples for more information.

3 Really using CxxTest

There is much more to CxxTest than seeing if two times two is four. You should probably take a look at the samples in the CxxTest distribution. Other than that, here are some more in-depth explanations.

3.1 What can you test

Here are the different "assertions" you can use in your tests:

Macro	Description	Example
TS_FAIL(message)	Fail unconditionally	TS_FAIL("Test not implemented");
TS_ASSERT(expr)	Verify (expr) is true	TS_ASSERT(messageReceived());
TS_ASSERT_EQUALS(x, y)	Verify (x==y)	TS_ASSERT_EQUALS(nodeCount(), 14);
TS_ASSERT_SAME_DATA(x, y, size)	Verify two buffers are equal	TS_ASSERT_SAME_DATA(input, output,, size);
TS_ASSERT_DELTA(x, y, d)	Verify (x==y) up to d	TS_ASSERT_DELTA(sqrt(4.0), 2.0, 0.0001);
TS_ASSERT_DIFFERS(x, y)	Verify !(x==y)	TS_ASSERT_DIFFERS(exam.numTook(), exam.numPassed());
TS_ASSERT_LESS_THAN(x, y)	Verify (x<y)	TS_ASSERT_LESS_THAN(ship.speed(), SPEED_OF_LIGHT);
TS_ASSERT_LESS_THAN_EQUALS(x, y)	Verify (x<=y)	TS_ASSERT_LESS_THAN_EQUALS(requests, items);
TS_ASSERT_PREDICATE(R, x)	Verify P(x)	TS_ASSERT_PREDICATE(SeemsReasonable, salary);
TS_ASSERT_RELATION(R, x, y)	Verify x R y	TS_ASSERT_RELATION(std::greater, salary, average);
TS_ASSERT_THROWS(expr, type)	Verify that (expr) throws a specific type of exception	TS_ASSERT_THROWS(parse(file), Parser::ReadError);
TS_ASSERT_THROWS_EQUALS(expr, arg, x, y)	Verify type and value of what (expr) throws	(See text)
TS_ASSERT_THROWS_ASSERT(expr, arg, assertion)	Verify type and value of what (expr) throws	(See text)
TS_ASSERT_THROWS_ANYTHING(expr)	Verify that (expr) throws an exception	TS_ASSERT_THROWS_ANYTHING(buggy());
TS_ASSERT_THROWS_NOTHING(expr)	Verify that (expr) doesn't throw anything	TS_ASSERT_THROWS_NOTHING(robust());
TS_WARN(message)	Print message as a warning	TS_WARN("TODO: Check invalid parameters");
TS_TRACE(message)	Print message as an informational message	TS_TRACE(errno);

3.1.1 TS_FAIL

TS_FAIL just fails the test. It is like an assert(false) with an error message. For example:

     
     void testSomething( void )
     {
        TS_FAIL( "I don't know how to test this!" );
     }
     
     

3.1.2 TS_ASSERT

TS_ASSERT is the basic all-around tester. It works just like the well-respected assert() macro (which I sincerely hope you know and use!) An example:

     
     void testSquare( void )
     {
        MyFileLibrary::createEmptyFile( "test.bin" );
        TS_ASSERT( access( "test.bin", 0 ) == 0 );
     }
     
     

3.1.3 TS_ASSERT_EQUALS

This is the second most useful tester. As the name hints, it is used to test if two values are equal.

     
     void testSquare( void )
     {
        TS_ASSERT_EQUALS( square(-5), 25 );
     }
     
     

3.1.4 TS_ASSERT_SAME_DATA

(v3.5.1) This assertion is similar to TS_ASSERT_EQUALS(), except that it compares the contents of two buffers in memory. If the comparison fails, the standard runner dumps the contents of both buffers as hex values.

     
     void testCopyMemory( void )
     {
        char input[77], output[77];
        myCopyMemory( output, input, 77 );
        TS_ASSERT_SAME_DATA( input, output, 77 );
     }
     
     

3.1.5 TS_ASSERT_DELTA

Similar to TS_ASSERT_EQUALS(), this macro verifies two values are equal up to a delta. This is basically used for floating-point values.

     
     void testSquareRoot( void )
     {
        TS_ASSERT_DELTA( squareRoot(4.0), 2.0, 0.00001 );
     }
     
     

3.1.6 TS_ASSERT_DIFFERS

The opposite of TS_ASSERT_EQUALS(), this macro is used to assert that two values are not equal.

     
     void testNumberGenerator( void )
     {
        int first = generateNumber();
        int second = generateNumber();
        TS_ASSERT_DIFFERS( first, second );
     }
     
     

3.1.7 TS_ASSERT_LESS_THAN

This macro asserts that the first operand is less than the second.

     
     void testFindLargerNumber( void )
     {
        TS_ASSERT_LESS_THAN( 23, findLargerNumber(23) );
     }
     
     

3.1.8 TS_ASSERT_LESS_THAN_EQUALS

(v3.7.0) Not surprisingly, this macro asserts that the first operand is less than or equals the second.

     
     void testBufferSize( void )
     {
        TS_ASSERT_LESS_THAN_EQUALS( bufferSize(), MAX_BUFFER_SIZE );
     }
     
     

3.1.9 TS_ASSERT_PREDICATE

(v3.8.2) This macro can be seen as a generalization of TS_ASSERT(). It takes as an argument the name of a class, similar to an STL unary_function, and evaluates operator(). The advantage this has over TS_ASSERT() is that you can see the failed value.

     
     class IsPrime
     {
     public:
        bool operator()( unsigned ) const;
     };
     
     // ...
     
     void testPrimeGenerator( void )
     {
        TS_ASSERT_PREDICATE( IsPrime, generatePrime() );
     }
     
     

3.1.10 TS_ASSERT_RELATION

(v3.8.0) Closely related to TS_ASSERT_PREDICATE(), this macro can be seen as a generalization of TS_ASSERT_EQUALS(), TS_ASSERT_DIFFERS(), TS_ASSERT_LESS_THAN() and TS_ASSERT_LESS_THAN_EQUALS(). It takes as an argument the name of a class, similar to an STL binary_function, and evaluates operator(). This can be used to very simply assert comparisons which are not covered by the builtin macros.

     
     void testGreater( void )
     {
        TS_ASSERT_RELATION( std::greater<int>, ticketsSold(), 1000 );
     }
     
     

3.1.11 TS_ASSERT_THROWS and friends

These assertions are used to test whether an expression throws an exception. TS_ASSERT_THROWS is used when you want to verify the type of exception thrown, and TS_ASSERT_THROWS_ANYTHING is used to just make sure something is thrown. As you might have guessed, TS_ASSERT_THROWS_NOTHING asserts that nothing is thrown. (v3.10.0) TS_ASSERT_THROWS_EQUALS checks the type of the exception as in TS_ASSERT_THROWS then allows you to compare two value (one of which will presumably be the caught object). TS_ASSERT_THROWS_ASSERT is the general case, and allows you to make any assertion about the thrown value. These macros may seem a little complicated, but they can be very useful; see below for an example.

     
     void testFunctionsWhichThrowExceptions( void )
     {
        TS_ASSERT_THROWS_NOTHING( checkInput(1) );
        TS_ASSERT_THROWS( checkInput(-11), std::runtime_error );
        TS_ASSERT_THROWS_ANYTHING( thirdPartyFunction() );
     
        TS_ASSERT_THROWS_EQUALS( validate(), const std::exception &e, 
                                 e.what(), "Invalid value" );
        TS_ASSERT_THROWS_ASSERT( validate(), const Error &e, 
                                 TS_ASSERT_DIFFERS( e.code(), SUCCESS ) );
     }
     
     

3.1.12 TS_TRACE and TS_WARN

(v3.0.1) TS_WARN just prints out a message, like the #warning preprocessor directive. I find it very useful for "to do" items. For example:

     
     void testToDoList( void )
     {
        TS_WARN( "TODO: Write some tests!" );
        TS_WARN( "TODO: Make $$$ fast!" );
     }
     
     

In the GUI, TS_WARN sets the bar color to yellow (unless it was already red). (v3.9.0) TS_TRACE is the same, except that it doesn't change the color of the progress bar.

3.1.13 The ETS_ macros

The TS_ macros mentioned above will catch exceptions thrown from tested code and fail the test, as if you called TS_FAIL(). Sometimes, however, you may want to catch the exception yourself; when you do, you can use the ETS_ versions of the macros.

     
     void testInterestingThrower()
     {
        // Normal way: if an exception is caught we can't examine it
        TS_ASSERT_EQUALS( foo(2), 4 );
     
        // More elaborate way:
        try { ETS_ASSERT_EQUALS( foo(2), 4 ); } 
        catch( const BadFoo &e ) { TS_FAIL( e.bar() ); }
     }
     
     

3.1.14 The TSM_ macros

Sometimes the default output generated by the ErrorPrinter doesn't give you enough information. This often happens when you move common test functionality to helper functions inside the test suite; when an assertion fails, you do not know its origin.

In the example below (which is the file sample/MessageTest.h from the CxxTest distribution), we need the message feature to know which invocation of checkValue() failed:

     
     class MessageTest : public CxxTest::TestSuite
     {
     public:
        void testValues()
        {
           checkValue( 0, "My hovercraft" );
           checkValue( 1, "is full" );
           checkValue( 2, "of eels" );
        }
     
        void checkValue( unsigned value, const char *message )
        {
           TSM_ASSERT( message, value );
           TSM_ASSERT_EQUALS( message, value, value * value );
        }
     };
     
     

3.1.14.1 The ETSM_ macros

Note: As with normal asserts, all TSM_ macros have their non-exception-safe counterparts, the ETSM_ macros.

3.2 Running the samples

CxxTest comes with some samples in the sample/ subdirectory of the distribution. If you look in that directory, you will see three Makefiles: Makefile.unix, Makefile.msvc and Makefile.bcc32 which are for Linux/Unix, MS Visual C++ and Borland C++, repectively. These files are provided as a starting point, and some options may need to be tweaked in them for your system.

If you are running under Windows, a good guess would be to run nmake -fMakefile.msvc run_win32 (you may need to run VCVARS32.BAT first). Under Linux, make -fMakefile.unix run_x11 should probably work.

3.3 Test fixtures

When you have several test cases for the same module, you often find that all of them start with more or less the same code--creating objects, files, inputs, etc. They may all have a common ending, too--cleaning up the mess you left.

You can (and should) put all this code in a common place by overriding the virtual functions TestSuite::setUp() and TestSuite::tearDown(). setUp() will then be called before each test, and tearDown() after each test.

     
     class TestFileOps : public CxxTest::TestSuite 
     {
     public:
        void setUp() { mkdir( "playground" ); }
        void tearDown() { system( "rm -Rf playground"); }
     
        void testCreateFile()
        {
           FileCreator fc( "playground" );
           fc.createFile( "test.bin" );
           TS_ASSERT_EQUALS( access( "playground/test.bin", 0 ), 0 );
        }
     };
     
     

Note new users: This is probably the single most important feature to use when your tests become non-trivial.

3.3.1 Test suite level fixtures

setUp()/tearDown() are executed around each test case. If you need a fixture on the test suite level, i.e. something that gets constructed once before all the tests in the test suite are run, see Dynamically creating test suites below.

3.4 Integrating with your build environment

It's very hard to maintain your tests if you have to generate, compile and run the test runner manually all the time. Fortunately, that's why we have build tools!

3.4.1 Overview

Let's assume you're developing an application. What I usually do is the following:

• Split the application into a library and a main module that just calls the library classes. This way, the test runner will be able to access all your classes through the library.
• Create another application (or target, or project, or whatever) for the test runner. Make the build tool generate it automatically.
• For extra points, make the build tool run the tests automatically.

3.4.2 Actually doing it

Unfortunately, there are way too many different build tools and IDE's for me to give ways to use CxxTest with all of them.

I will try to outline the usage for some cases.

3.4.2.1 Using Makefiles

Generating the tests with a makefile is pretty straightforward. Simply add rules to generate, compile and run the test runner.

     
     all: lib run_tests app
     
     # Rules to build your targets
     lib: ...
     
     app: ...
     
     # A rule that runs the unit tests
     run_tests: runner
             ./runner
     
     # How to build the test runner
     runner: runner.cpp lib
             g++ -o $@ $^
     
     # How to generate the test runner
     runner.cpp: SimpleTest.h ComplicatedTest.h
              cxxtestgen.pl -o $@ --error-printer $^
     
     

3.4.2.2 Using Cons

See sample/Construct in the CxxTest distribution for an example of building CxxTest test runners with Cons.

3.4.2.3 Using Microsoft Visual Studio

I have tried several ways to integrate CxxTest with visual studio, none of which is perfect. Take a look at sample/msvc in the distribution to see the best solution I'm aware of. Basically, the workspace has three projects:

• The project CxxTest_3_Generate runs cxxtestgen.
• The project CxxTest_2_Build compiles the generated file.
• The project CxxTest_1_Run runs the tests.

This method certainly works, and the test results are conveniently displayed as compilation errors and warnings (for TS_WARN()). However, there are still a few things missing; to integrate this approach with your own project, you usually need to work a little bit and tweak some makefiles and project options. I have provided a small script in sample/msvc/FixFiles.bat to automate some of the process.

3.4.2.4 Using Microsoft Windows DDK

Unit testing for device drivers?! Why not? And besides, the build utility can also be used to build user-mode application.

To use CxxTest with the build utility, you add the generated tests file as an extra dependency using the NTBUILDTARGET0 macro and the Makefile.inc file.

You can see an example of how to do this in the CxxTest distribution under sample/winddk.

3.5 Graphical user interface

There are currently three GUIs implemented: native Win32, native X11 and Qt. To use this feature, just specify --gui=X11Gui, --gui=Win32Gui or --gui=QtGui as a parameter for cxxtestgen (instead of e.g. --error-printer). A progress bar is displayed, but the results are still written to standard output, where they can be processed by your IDE (e.g. Emacs or Visual Studio). The default behavior of the GUI is to close the window after the last test.

Note that whatevr GUI you use, you can combine it with the --runner option to control the formatting of the text output, e.g. Visual Studio likes it better if you use --runner=ParenPrinter.

3.5.1 Starting the GUI minimized

If you run the generated Win32 or Qt GUIs with the command line -minimized, the test window will start minimized (iconified) and only pop up if there is an error (the bar turns red). This is useful if you find the progress bar distracting and only want to check it if something happens.

3.5.2 Leaving the GUI open

The Win32 GUI accepts the -keep which instructs it to leave the window open after the tests are done. This allows you to see how many tests failed and how much time it took.

3.5.3 Screenshots!

As with any self-respecting GUI application, here are some screenshots for you to enjoy:

• Using the Qt GUI on Linux (with the WindowMaker window manager):
  
  
  

• Using the Win32 GUI on Windows 98:
  
  
  

• Using the X11 GUI (with the venerable TWM):
  
  
  

• And of course, no GUI is complete without the ability to mess around with its appearance:
  
  
  

  Ahhh. Nothing like a beautiful user interface.

4 Advanced topics

Topics in this section are more technical, and you probably won't find them interesting unless you need them.

4.1 Aborting tests after failures

Usually, when a TS_ASSERT_* macro fails, CxxTest moves on to the next line. In many cases, however, this is not the desired behavior. Consider the following code:

     
     void test_memset()
     {
        char *buffer = new char[1024];
        TS_ASSERT( buffer );
        memset( buffer, 0, 1024 ); // But what if buffer == 0?
     }
     
     

If you have exception handling enabled, you can make CxxTest exit each test as soon as a failure occurs. To do this, you need to define CXXTEST_ABORT_TEST_ON_FAIL before including the CxxTest headers. This can be done using the --abort-on-fail command-line option or in a template file; see sample/aborter.tpl in the distribution. Note that if CxxTest doesn't find evidence of exception handling when scanning your files, this feature will not work. To overcome this, use the --have-eh command-line option.

4.1.1 Controlling this behavior at runtime

4.2 Commenting out tests

CxxTest does a very simple analysis of the input files, which is sufficient in most cases. This means, for example, that you can't indent you test code in "weird" ways.

A slight inconvenience arises, however, when you want to comment out tests. Commenting out the tests using C-style comments or the preprocessor will not work:

     
     class MyTest : public CxxTest::TestSuite
     {
     public:
     /*
        void testCommentedOutStillGetsCalled()
        {
        }
     */
     
     #if 0
        void testMarkedOutStillGetsCalled()
        {
        }
     #endif
     };
     
     

     
     class MyTest : public CxxTest::TestSuite
     {
     public:
     // void testFutureStuff()
     // {
     // }
     
        void xtestFutureStuff()
        {
        }
     };
     
     

4.3 Comparing equality for your own types

You may have noticed that TS_ASSERT_EQUALS() only works for built-in types. This is because CxxTest needs a way to compare object and to convert them to strings, in order to print them should the test fail.

If you do want to use TS_ASSERT_EQUALS() on your own data types, this is how you do it.

4.3.1 The equality operator

First of all, don't forget to implement the equality operator (operator==()) on your data types!

4.3.2 Value traits

Since CxxTest tries not to rely on any external library (including the standard library, which is not always available), conversion from arbitrary data types to strings is done using value traits.

For example, to convert an integer to a string, CxxTest does the following actions:

• int i = value to convert;
• CxxTest::ValueTraits<int> converter(i);
• string = converter.asString();

CxxTest comes with predefined ValueTraits for int, char, dobule etc. in cxxtest/ValueTraits.h in the cxxtest-selftest archive.

4.3.3 Unknown types

Obviously, CxxTest doesn't "know" about all possible types. The default ValueTraits class for unknown types dumps up to 8 bytes of the value in hex format.

For example, the following code

     
     #include <cxxtest/TestSuite.h>
     
     class TestMyData : public CxxTest::TestSuite 
     {
     public:
        struct Data
        {
           char data[3];
        };
     
        void testCompareData()
        {
           Data x, y;
           memset( x.data, 0x12, sizeof(x.data) );
           memset( y.data, 0xF6, sizeof(y.data) );
           TS_ASSERT_EQUALS( x, y );
        }
     };
     
     

     
     Running 1 test.
     TestMyData.h:16: Expected (x == y), found ({ 12 12 12 } != { F6 F6 F6 })
     Failed 1 of 1 test
     Success rate: 0%
     
     

4.3.4 Enumeration traits

     
     enum Status { STATUS_IDLE, STATUS_BUSY, STATUS_ERROR };
     
     CXXTEST_ENUM_TRAITS( Status,
                          CXXTEST_ENUM_MEMBER( STATUS_IDLE )
                          CXXTEST_ENUM_MEMBER( STATUS_BUSY )
                          CXXTEST_ENUM_MEMBER( STATUS_ERROR ) );
     
     

See sample/EnumTraits.h for a working sample.

4.3.5 Defining new value traits

Defining value traits for new (non-enumeration) types is easy. All you need is to define a way to convert an object of your class to a string. You can use this example as a possible skeleton:

     
     class MyClass 
     {
        int _value;
     
     public:
        MyClass( int value ) : _value( value ) {}
        int value() const { return _value; }
     
        // CxxTest requires a copy constructor
        MyClass( const MyClass &other ) : _value( other._value ) {}
     
        // If you want to use TS_ASSERT_EQUALS
        bool operator== ( const MyClass &other ) const { return _value == other._value; }
     
        // If you want to use TS_ASSERT_LESS_THAN
        bool operator== ( const MyClass &other ) const { return _value < other._value; }
     };
     
     #ifdef CXXTEST_RUNNING
     #include <cxxtest/ValueTraits.h>
     #include <stdio.h>
     
     namespace CxxTest 
     {
        CXXTEST_TEMPLATE_INSTANTIATION
        class ValueTraits<MyClass> 
        {
           char _s[256];
     
        public:
           ValueTraits( const MyClass &m ) { sprintf( _s, "MyClass( %i )", m.value() ); }
           const char *asString() const { return _s; }
        };
     };
     #endif // CXXTEST_RUNNING
     
     

4.3.5.1 Defining value traits for template classes

A simple modification to the above scheme allows you to define value traits for your template classes. Unfortunately, this syntax (partial template specialization) is not supported by some popular C++ compilers. Here is an example:

     
     template<class T>
     class TMyClass
     {
        T _value;
     
     public:
        TMyClass( const T &value ) : _value( value );
        const T &value() const { return _value; }
     
        // CxxTest requires a copy constructor
        TMyClass( const TMyClass<T> &other ) : _value( other._value ) {}
        
        // If you want to use TS_ASSERT_EQUALS
        bool operator== ( const TMyClass<T> &other ) const { return _value == other._value; }
     };
     
     #ifdef CXXTEST_RUNNING
     #include <cxxtest/ValueTraits.h>
     #include <typeinfo>
     #include <sstream>
     
     namespace CxxTest 
     {
        template<class T>
        class ValueTraits< TMyClass<T> > 
        {
           std::ostringstream _s;
     
        public:
           ValueTraits( const TMyClass<T> &t ) 
              { _s << typeid(t).name() << "( " << t.value() << " )"; }
           const char *asString() const { return _s.str().c_str(); }
        };
     };
     #endif // CXXTEST_RUNNING
     
     

4.3.6 Overriding the default value traits

You can see a sample of this technique in test/UserTraits.tpl in the cxxtest-selftest archive.

4.4 Global Fixtures

Note: Unlike setUp() and tearDown() in TestSuite, global fixtures should return a bool value to indicate success/failure.

4.4.1 World fixtures

4.5 Mock Objects

So far, so good. But the problem when developing in C/C++ is that your code probably needs to call global functions which you cannot override. Just consider any code which uses fopen(), fwrite() and fclose(). It is not very elegant to have this code actually create files while being tested. Even more importantly, you (should) want to test how the code behaves when "bad" things happen, say when fopen() fails. Although for some cases you can cause the effects to happen in the test code, this quickly becomes "hairy" and unmaintainable.

CxxTest solves this problem by allowing you to override any global function while testing. Here is an outline of how it works, before we see an actual example:

• For each function you want to override, you use the macro CXXTEST_MOCK_GLOBAL to "prepare" the function (all is explained below in excruciating detail).
• In the tested code you do not call the global functions directly; rather, you access them in the T (for Test) namespace. For instance, your code needs to call T::fopen() instead of fopen(). This is the equivalent of using abstract interfaces instead of concrete classes.
• You link the "real" binary with a source file that implements T::fopen() by simply calling the original fopen().
• You link the test binary with a source file that implements T::fopen() by calling a mock object.
• To test, you should create a class that inherits T::Base_fopen and implement its fopen() function. Simply by creating an object of this class, calls made to T::fopen() will be redirected to it.

This may seem daunting at first, so let us work our way through a simple example. Say we want to override the well known standard library function time().

• Prepare a header file to be used by both the real and test code.

            
            // T/time.h
            #include <time.h>
            #include <cxxtest/Mock.h>
            
            CXXTEST_MOCK_GLOBAL( time_t,        /* Return type          */
                                 time,          /* Name of the function */
                                 ( time_t *t ), /* Prototype            */
                                 ( t )          /* Argument list        */ );
            
            

• In our tested code, we now include the special header instead of the system-supplied one, and call T::time() instead of time().

            
            // code.cpp
            #include <T/time.h>
            
            int generateRandomNumber()
            {
                return T::time( NULL ) * 3;
            }
            
            

• We also need to create a source file that implements T::time() by calling the real function. This is extremely easy: just define CXXTEST_MOCK_REAL_SOURCE_FILE before you include the header file:

            
            // real_time.cpp
            #define CXXTEST_MOCK_REAL_SOURCE_FILE
            #include <T/time.h>
            
            

• Before we can start testing, we need a different implementation of T::time() for our tests. This is just as easy as the previous one:

            
            // mock_time.cpp
            #define CXXTEST_MOCK_TEST_SOURCE_FILE
            #include <T/time.h>
            
            

• Now comes the fun part. In our test code, all we need to do is create a mock, and the tested code will magically call it:

            
            // TestRandom.h
            #include <cxxtest/TestSuite.h>
            #include <T/time.h>
            
            class TheTimeIsOne : public T::Base_time
            {
            public:
                time_t time( time_t * ) { return 1; }
            };
            
            class TestRandom : public CxxTest::TestSuite
            {
            public:
                void test_Random()
                {
                    TheTimeIsOne t;
                    TS_ASSERT_EQUALS( generateRandomNumber(), 3 );
                }
            };
            
            

4.5.1 Actually doing it

I know that this might seem a bit heavy at first glance, but once you start using mock objects you will never go back. The hardest part may be getting this to work with your build system, which is why I have written a simple example much like this one in sample/mock, which uses GNU Make and G++.

4.5.2 Advanced topic with mock functions

4.5.2.1 Void functions

Void function are a little different, and you use CXXTEST_MOCK_VOID_GLOBAL to override them. This is identical to CXXTEST_MOCK_GLOBAL except that it doesn't specify the return type. Take a look in sample/mock/T/stdlib.h for a demonstation.

4.5.2.2 Calling the real functions while testing

From time to time, you might want to let the tested code call the real functions (while being tested). To do this, you create a special mock object called e.g. T::Real_time. While an object of this class is present, calls to T::time() will be redirected to the real function.

4.5.2.3 When there is no real function

Sometimes your code needs to call functions which are not available when testing. This happens for example when you test driver code using a user-mode test runner, and you need to call kernel functions. You can use CxxTest's mock framework to provide testable implementations for the test code, while maintaing the original functions for the real code. This you do with CXXTEST_SUPPLY_GLOBAL (and CXXTEST_SUPPLY_VOID_GLOBAL). For example, say you want to supply your code with the Win32 kernel function IoCallDriver:

     
     CXXTEST_SUPPLY_GLOBAL( NTSTATUS,                /* Return type */
                            IoCallDriver,            /* Name        */
                            ( PDEVICE_OBJECT Device, /* Prototype   */
                              PIRP Irp ),
                            ( Device, Irp )          /* How to call */ );
     
     

Note: Since these macros can also be used to actually declare the function prototypes (e.g. in the above example you might not be able to include the real <ntddk.h> from test code), they also have an extern "C" version which declares the functions with C linkage. These are CXXTEST_SUPPLY_GLOBAL_C and CXXTEST_SUPPLY_GLOBAL_VOID_C.

4.5.2.4 Functions in namespaces

Sometimes the functions you want to override are not in the global namespace like time(): they may be global functions in other namespaces or even static class member functions. The default mock implementation isn't suitable for these. For them, you can use the generic CXXTEST_MOCK, which is best explained by example. Say you have a namespace Files, and you want to override the function bool Files::FileExists( const String &name ), so that the mock class will be called T::Base_Files_FileExists and the function to implement would be fileExists. You would define it thus (of course, you would normally want the mock class name and member function to be the same as the real function):

     
     CXXTEST_MOCK( Files_FileExists,       /* Suffix of mock class  */
                   bool,                   /* Return type           */
                   fileExists,             /* Name of mock member   */
                   ( const String &name ), /* Prototype             */
                   Files::FileExists,      /* Name of real function */
                   ( name )                /* Parameter list        */ );
     
     

There is also an equivalent version for CXXTEST_SUPPLY_GLOBAL, as demonstrated by another function from the Win32 DDK:

     
     CXXTEST_SUPPLY( AllocateIrp,         /* => T::Base_AllocateIrp */
                     PIRP,                /* Return type            */
                     allocateIrp,         /* Name of mock member    */
                     ( CCHAR StackSize ), /* Prototype              */
                     IoAllocateIrp,       /* Name of real function  */
                     ( StackSize )        /* Parameter list         */ );
     
     

4.5.2.5 Overloaded functions

If you have two or more global functions which have the same name, you cannot create two mock classes with the same name. The solution is to use the general CXXTEST_MOCK/CXXTEST_MOCK_VOID as above: just give the two mock classes different names.

4.5.2.6 Changing the mock namespace

Finally, if you don't like or for some reason can't use the T:: namespace for mock functions, you can change it by defining CXXTEST_MOCK_NAMESPACE. Have fun.

4.6 Test Listeners and Test Runners

A TestListener is a class that receives notifications about the testing process, notably which assertions failed. CxxTest defines a standard test listener class, ErrorPrinter, which is responsible for printing the dots and messages seen above. When the test runners generated in the examples run, they create an ErrorPrinter and pass it to TestRunner::runAllTests(). As you might have guessed, this functions runs all the test you've defined and reports to the TestListener it was passed.

4.6.1 Other test listeners

If you don't like or can't use the ErrorPrinter, you can use any other test listener. To do this you have to omit the --error-printer, --runner= or --gui= switch when generating the tests file. It is then up to you to write the main() function, using the test listener of your fancy.

4.6.1.1 The stdio printer

If the ErrorPrinter's usage of std::cout clashes with your environment or is unsupported by your compiler, don't dispair! You may still be able to use the StdioPrinter, which does the exact same thing but uses good old printf().

To use it, invoke cxxtestgen.pl with the --runner=StdioPrinter option. (v3.8.5) Note: cxxtest/StdioPrinter makes reference to stdout as the default output stream. In some environments you may have <stdio.h> but not stdout, which will cause compiler errors. To overcome this problem, use --runner=StdioFilePrinter, which is exactly the same as --runner=StdioPrinter, but with no default output stream.

4.6.1.2 The Yes/No runner

As an example, CxxTest also provides the simplest possible test listener, one that just reports if there were any failures. You can see an example of using this listener in sample/yes_no_runner.cpp.

4.6.1.3 Template files

To use you own test runner, or to use the supplied ones in different ways, you can use CxxTest template files. These are ordinary source files with the embedded "command" <CxxTest world> which tells cxxtestgen.pl to insert the world definition at that point. You then specify the template file using the --template option.

See samples/file_printer.tpl for an example.

Note: CxxTest needs to insert certain definitions and #include directives in the runner file. It normally does that before the first #include <cxxtest/*.h> found in the template file. If this behvaior is not what you need, use the "command" <CxxTest preamble>. See test/preamble.tpl in the cxxtest-selftest archive for an example of this.

4.7 Dynamically creating test suites

Usually, your test suites are instantiated statically in the tests file, i.e. say you defined class MyTest : public CxxTest::TestSuite, the generated file will contain something like static MyTest g_MyTest;.

If, however, your test suite must be created dynamically (it may need a constructor, for instance), CxxTest doesn't know how to create it unless you tell it how. You do this by writing two static functions, createSuite() and destroySuite().

See sample/CreatedTest.h for a demonstration.

4.8 Static initialization

Appendix A Command line options

Here are the different command line options for cxxtestgen:

A.1 --version

A.2 --output

Specify --output=FILE or -o FILE to determine the output file name.

A.3 --error-printer

This option creates a test runner which uses the standard error printer class.

A.4 --runner

Specify --runner=CLASS to generate a test runner that #includes <cxxtest/CLASS.h> and uses CxxTest::CLASS as the test runner.

The currently available runners are:

--runner=ErrorPrinter

This is the standard error printer, which formats its output to std::cout.

--runner=ParenPrinter

Identical to ErrorPrinter except that it prints line numbers in parantheses. This is the way Visual Studio expects it.

--runner=StdioPrinter

The same as ErrorPrinter except that it uses printf instead of cout.

--runner=YesNoRunner

This runner doesn't produce any output, merely returns a true/false result.

A.5 --gui

Specify --gui=CLASS to generate a test runner that #includes <cxxtest/CLASS.h> and uses CxxTest::CLASS to display a graphical user interface. This option can be combined with the --runner option to determine the text-mode output format. The default is the standard error printer.

There are three different GUIs:

--gui=Win32Gui

A native Win32 GUI. It has been tested on Windows 98, 2000 and XP and should work unmodified on other 32-bit versions of Windows.

--gui=X11Gui

A native XLib GUI. This GUI is very spartan and should work on any X server.

--gui=QtGui

A GUI that uses the Qt library from Troll. It has been tested with Qt versiond 2.2.1 and 3.0.1.

A.6 --include

Note: If you want the runner to #inculde <FILE>, specify it on the command line, e.g. --include=<FILE>. You will most likely need to use shell escapes, e.g. "--include=<FILE>" or --include=\<FILE\>.

Examples: --include=TestDefs.h or --include=\<GlobalDefs.h\>.

A.7 --template

Specify --template=FILE to use FILE as a template file. This is for cases for which --runner and/or --include are not enough. One example is the Windows DDK; see sample/winddk in the distribution.

A.8 --have-eh

A.9 --no-eh

A.10 --have-std

Note: If you reference the standard library anywhere in your test files, CxxTest will (usually) recognize it and automatically define this.

A.11 --no-std

A.12 --longlong

A.13 --abort-on-fail

A.14 --part

A.15 --root

A.16 --no-static-init

Appendix B Controlling the behavior of CxxTest

Here are various #defines you can use to modify how CxxTest works. You will need to #define them before including any of the CxxTest headers, so use them in a template file or with the --include option.

B.1 CXXTEST_HAVE_STD

This is equivalent to the --have-std option.

B.2 CXXTEST_HAVE_EH

This is equivalent to the --have-eh option.

B.3 CXXTEST_ABORT_TEST_ON_FAIL

B.4 CXXTEST_USER_VALUE_TRAITS

This tells CxxTest you wish to define you own ValueTraits. It will only declare the default traits, which dump up to 8 bytes of the data as hex values.

B.5 CXXTEST_OLD_TEMPLATE_SYNTAX

Some compilers (e.g. Borland C++ 5) don't support the standard way of instantiating template classes. Use this define to overcome the problem.

B.6 CXXTEST_OLD_STD

Again, this is used to support pre-std:: standard libraries.

B.7 CXXTEST_MAX_DUMP_SIZE

This sets the standard maximum number of bytes to dump if TS_ASSERT_SAME_DATA() fails. The default is 0, meaning no limit.

B.8 CXXTEST_DEFAULT_ABORT

This sets the default value of the dynamic "abort on fail" flag. Of course, this flag is only used when "abort on fail" is enabled.

B.9 CXXTEST_LONGLONG

This is equivalent to --longlong.

Appendix C Runtime options

The following functions can be called during runtime (i.e. from your tests) to control the behavior of CxxTest. They are reset to their default values after each test is executed (more precisely, after tearDown() is called). Consequently, if you set them in the setUp() function, they will be valid for the entire test suite.

C.1 setAbortTestOnFail( bool )

This only works when you have exception handling. It can be used to tell CxxTest to temporarily change its behavior. The default value of the flag is false, true if you set --abort-on-fail, or CXXTEST_DEFAULT_ABORT if you #define it.

C.2 setMaxDumpSize( unsigned )

This temporarily sets the maximum number of bytes to dump if TS_ASSERT_SAME_DATA() fails. The default is 0, meaning no limit, or CXXTEST_MAX_DUMP_SIZE if you #define it.

Appendix D Version history

• Version 3.10.0 (2004-11-20)
  • Added mock framework for global functions
  • Added TS_ASSERT_THROWS_ASSERT and TS_ASSERT_THROWS_EQUALS
  • Added CXXTEST_ENUM_TRAITS
  • Improved support for STL classes (vector, map etc.)
  • Added support for Digital Mars compiler
  • Reduced root/part compilation time and binary size
  • Support C++-style commenting of tests
• Version 3.9.1 (2004-01-19)
  • Fixed small bug with runner exit code
  • Embedded test suites are now deprecated
• Version 3.9.0 (2004-01-17)
  • Added TS_TRACE
  • Added --no-static-init
  • CxxTest::setAbortTestOnFail() works even without --abort-on-fail
• Version 3.8.5 (2004-01-08)
  • Added --no-eh
  • Added CxxTest::setAbortTestOnFail() and CXXTEST_DEFAULT_ABORT
  • Added CxxTest::setMaxDumpSize()
  • Added StdioFilePrinter
• Version 3.8.4 (2003-12-31)
  • Split distribution into cxxtest and cxxtest-selftest
  • Added sample/msvc/FixFiles.bat
• Version 3.8.3 (2003-12-24)
  • Added TS_ASSERT_PREDICATE
  • Template files can now specify where to insert the preamble
  • Added a sample Visual Studio workspace in sample/msvc
  • Can compile in MSVC with warning level 4
  • Changed output format slightly
• Version 3.8.1 (2003-12-21)
  • Fixed small bug when using multiple --part files.
  • Fixed X11 GUI crash when there's no X server.
  • Added GlobalFixture::setUpWorld()/tearDownWorld()
  • Added leaveOnly(), activateAllTests() and sample/only.tpl
  • Should now run without warnings on Sun compiler.
• Version 3.8.0 (2003-12-13)
  • Fixed bug where Root.cpp needed exception handling
  • Added TS_ASSERT_RELATION
  • TSM_ macros now also tell you what went wrong
  • Renamed Win32Gui::free() to avoid clashes
  • Now compatible with more versions of Borland compiler
  • Improved the documentation
• Version 3.7.1 (2003-09-29)
  • Added --version
  • Compiles with even more exotic g++ warnings
  • Win32 Gui compiles with UNICODE
  • Should compile on some more platforms (Sun Forte, HP aCC)
• Version 3.7.0 (2003-09-20)
  • Added TS_ASSERT_LESS_THAN_EQUALS
  • Minor cleanups
• Version 3.6.1 (2003-09-15)
  • Improved QT GUI
  • Improved portability some more
• Version 3.6.0 (2003-09-04)
  • Added --longlong
  • Some portability improvements
• Version 3.5.1 (2003-09-03)
  • Major internal rewrite of macros
  • Added TS_ASSERT_SAME_DATA
  • Added --include option
  • Added --part and --root to enable splitting the test runner
  • Added global fixtures
  • Enhanced Win32 GUI with timers, -keep and -title
  • Now compiles with strict warnings
• Version 3.1.1 (2003-08-27)
  • Fixed small bug in TS_ASSERT_THROWS_*()
• Version 3.1.0 (2003-08-23)
  • Default ValueTraits now dumps value as hex bytes
  • Fixed double invocation bug (e.g. TS_FAIL(functionWithSideEffects()))
  • TS_ASSERT_THROWS*() are now "abort on fail"-friendly
  • Win32 GUI now supports Windows 98 and doesn't need comctl32.lib
• Version 3.0.1 (2003-08-07)
  • Added simple GUI for X11, Win32 and Qt
  • Added TS_WARN() macro
  • Removed --exit-code
  • Improved samples
  • Improved support for older (pre-std::) compilers
  • Made a PDF version of the User's Guide
• Version 2.8.4 (2003-07-21)
  • Now supports g++-3.3
  • Added --have-eh
  • Fixed bug in numberToString()
• Version 2.8.3 (2003-06-30)
  • Fixed bugs in cxxtestgen.pl
  • Fixed warning for some compilers in ErrorPrinter/StdioPrinter
  • Thanks Martin Jost for pointing out these problems!
• Version 2.8.2 (2003-06-10)
  • Fixed bug when using CXXTEST_ABORT_TEST_ON_FAIL without standard library
  • Added CXXTEST_USER_TRAITS
  • Added --abort-on-fail
• Version 2.8.1 (2003-01-16)
  • Fixed charToString() for negative chars
• Version 2.8.0 (2003-01-13)
  • Added CXXTEST_ABORT_TEST_ON_FAIL for xUnit-like behaviour
  • Added sample/winddk
  • Improved ValueTraits
  • Improved output formatter
  • Started version history
• Version 2.7.0 (2002-09-29)
  • Added embedded test suites
  • Major internal improvements