Home | All Classes | Main Classes | Annotated | Grouped Classes | Functions

Using the Meta Object Compiler

The Meta Object Compiler, moc among friends, is the program which handles TQt's C++ extensions.

The moc reads a C++ source file. If it finds one or more class declarations that contain the Q_OBJECT macro, it produces another C++ source file which contains the meta object code for the classes that use the Q_OBJECT macro. Among other things, meta object code is retquired for the signal/slot mechanism, runtime type information and the dynamic property system.

The C++ source file generated by the moc must be compiled and linked with the implementation of the class (or it can be #included into the class's source file).

If you use qmake to create your Makefiles, build rules will be included that call the moc when retquired, so you will not need to use the moc directly. For more background information on moc, see Why doesn't TQt use templates for signals and slots?.

Usage

The moc is typically used with an input file containing class declarations like this:

    class MyClass : public TQObject
    {
        Q_OBJECT
    public:
        MyClass( TQObject * parent=0, const char * name=0 );
        ~MyClass();

    signals:
        void mySignal();

    public slots:
        void mySlot();

    };

In addition to the signals and slots shown above, the moc also implements object properties as in the next example. The Q_PROPERTY macro declares an object property, while Q_ENUMS declares a list of enumeration types within the class to be usable inside the property system. In this particular case we declare a property of the enumeration type Priority that is also called "priority" and has a get function priority() and a set function setPriority().

    class MyClass : public TQObject
    {
        Q_OBJECT
        Q_PROPERTY( Priority priority READ priority WRITE setPriority )
        Q_ENUMS( Priority )
    public:
        MyClass( TQObject * parent=0, const char * name=0 );
        ~MyClass();

        enum Priority { High, Low, VeryHigh, VeryLow };
        void setPriority( Priority );
        Priority priority() const;
    };

Properties can be modified in subclasses with the Q_OVERRIDE macro. The Q_SETS macro declares enums that are to be used as sets, i.e. OR'ed together. Another macro, Q_CLASSINFO, can be used to attach additional name/value-pairs to the class' meta object:

    class MyClass : public TQObject
    {
        Q_OBJECT
        Q_CLASSINFO( "Author", "Oscar Peterson")
        Q_CLASSINFO( "Status", "Active")
    public:
        MyClass( TQObject * parent=0, const char * name=0 );
        ~MyClass();
    };

The three concepts, signals and slots, properties and class meta-data, can be combined.

The output produced by the moc must be compiled and linked, just like the other C++ code in your program; otherwise the build will fail in the final link phase. By convention, this is done in one of the following two ways:

Method A: The class declaration is found in a header (.h) file

If the class declaration above is found in the file myclass.h, the moc output should be put in a file called moc_myclass.cpp. This file should then be compiled as usual, resulting in an object file moc_myclass.o (on Unix) or moc_myclass.obj (on Windows). This object should then be included in the list of object files that are linked together in the final building phase of the program.

Method B: The class declaration is found in an implementation (.cpp) file

If the class declaration above is found in the file myclass.cpp, the moc output should be put in a file called myclass.moc. This file should be #included in the implementation file, i.e. myclass.cpp should contain the line

    #include "myclass.moc"

at the end. This will cause the moc-generated code to be compiled and linked together with the normal class definition in myclass.cpp, so it is not necessary to compile and link it separately, as in Method A.

Method A is the normal method. Method B can be used in cases where you want the implementation file to be self-contained, or in cases where the Q_OBJECT class is implementation-internal and thus should not be visible in the header file.

Automating moc Usage with Makefiles

For anything but the simplest test programs, it is recommended that you automate running the moc. By adding some rules to your program's Makefile, make can take care of running moc when necessary and handling the moc output.

We recommend using Trolltech's free makefile generation tool, qmake, for building your Makefiles. This tool recognizes both Method A and B style source files, and generates a Makefile that does all the necessary moc handling.

If you want to create your Makefiles yourself, here are some tips on how to include moc handling.

For Q_OBJECT class declarations in header files, here is a useful makefile rule if you only use GNU make:

    moc_%.cpp: %.h
            moc $< -o $@

If you want to write portably, you can use individual rules with the following form:

    moc_NAME.cpp: NAME.h
            moc $< -o $@

You must also remember to add moc_NAME.cpp to your SOURCES (substitute your favorite name) variable and moc_NAME.o or moc_NAME.obj to your OBJECTS variable.

(While we prefer to name our C++ source files .cpp, the moc doesn't care, so you can use .C, .cc, .CC, .cxx or even .c++ if you prefer.)

For Q_OBJECT class declarations in implementation (.cpp) files, we suggest a makefile rule like this:

    NAME.o: NAME.moc

    NAME.moc: NAME.cpp
            moc -i $< -o $@

This guarantees that make will run the moc before it compiles NAME.cpp. You can then put

    #include "NAME.moc"

at the end of NAME.cpp, where all the classes declared in that file are fully known.

Invoking moc

Here are the command-line options supported by the moc:

You can explicitly tell the moc not to parse parts of a header file. It recognizes any C++ comment (//) that contains the substrings MOC_SKIP_BEGIN or MOC_SKIP_END. They work as you would expect and you can have several levels of them. The net result as seen by the moc is as if you had removed all lines between a MOC_SKIP_BEGIN and a MOC_SKIP_END.

Diagnostics

The moc will warn you about a number of dangerous or illegal constructs in the Q_OBJECT class declarations.

If you get linkage errors in the final building phase of your program, saying that YourClass::className() is undefined or that YourClass lacks a vtbl, something has been done wrong. Most often, you have forgotten to compile or #include the moc-generated C++ code, or (in the former case) include that object file in the link command.

Limitations

The moc does not expand #include or #define, it simply skips any preprocessor directives it encounters. This is regrettable, but is not usually a problem in practice.

The moc does not handle all of C++. The main problem is that class templates cannot have signals or slots. Here is an example:

    class SomeTemplate<int> : public TQFrame {
        Q_OBJECT
        ...
    signals:
        void bugInMocDetected( int );
    };

Less importantly, the following constructs are illegal. All of them have alternatives which we think are usually better, so removing these limitations is not a high priority for us.

Multiple inheritance retquires TQObject to be first

If you are using multiple inheritance, moc assumes that the first inherited class is a subclass of TQObject. Also, be sure that only the first inherited class is a TQObject.

    class SomeClass : public TQObject, public OtherClass {
        ...
    };

(This limitation is almost impossible to remove; since the moc does not expand #include or #define, it cannot find out which one of the base classes is a TQObject.)

Function pointers cannot be arguments to signals or slots

In most cases where you would consider using function pointers as signal/slot arguments, we think inheritance is a better alternative. Here is an example of illegal syntax:

    class SomeClass : public TQObject {
        Q_OBJECT
        ...
    public slots:
        // illegal
        void apply( void (*apply)(List *, void *), char * );
    };

You can work around this restriction like this:

    typedef void (*ApplyFunctionType)( List *, void * );

    class SomeClass : public TQObject {
        Q_OBJECT
        ...
    public slots:
        void apply( ApplyFunctionType, char * );
    };

It may sometimes be even better to replace the function pointer with inheritance and virtual functions, signals or slots.

Friend declarations cannot be placed in signals or slots sections

Sometimes it will work, but in general, friend declarations cannot be placed in signals or slots sections. Put them in the private, protected or public sections instead. Here is an example of the illegal syntax:

    class SomeClass : public TQObject {
        Q_OBJECT
        ...
    signals:
        friend class ClassTemplate<char>; // WRONG
    };

Signals and slots cannot be upgraded

The C++ feature of upgrading an inherited member function to public status is not extended to cover signals and slots. Here is an illegal example:

    class Whatever : public TQButtonGroup {
        ...
    public slots:
        TQButtonGroup::buttonPressed; // WRONG
        ...
    };

The TQButtonGroup::buttonPressed() slot is protected.

C++ tquiz: What happens if you try to upgrade a protected member function which is overloaded?

1. All the functions are overloaded.
2. That is not legal C++.

Type macros cannot be used for signal and slot parameters

Since the moc does not expand #define, type macros that take an argument will not work in signals and slots. Here is an illegal example:

    #ifdef ultrix
    #define SIGNEDNESS(a) unsigned a
    #else
    #define SIGNEDNESS(a) a
    #endif

    class Whatever : public TQObject {
        ...
    signals:
        void someSignal( SIGNEDNESS(int) );
        ...
    };

A #define without parameters will work as expected.

Nested classes cannot be in the signals or slots sections nor have signals or slots

Here's an example:

    class A {
        Q_OBJECT
    public:
        class B {
        public slots:   // WRONG
            void b();
            ...
        };
    signals:
        class B {       // WRONG
            void b();
            ...
        }:
    };

Constructors cannot be used in signals or slots sections

It is a mystery to us why anyone would put a constructor in either the signals or slots sections. You can't anyway (except that it happens to work in some cases). Put them in private, protected or public sections, where they belong. Here is an example of the illegal syntax:

    class SomeClass : public TQObject {
        Q_OBJECT
    public slots:
        SomeClass( TQObject *parent, const char *name )
            : TQObject( parent, name ) { } // WRONG
        ...
    };

Properties need to be declared before the public section that contains the respective get and set functions

Declaring the first property within or after the public section that contains the type definition and the respective get and set functions does not work as expected. The moc will complain that it can neither find the functions nor resolve the type. Here is an example of the illegal syntax:

    class SomeClass : public TQObject {
        Q_OBJECT
    public:
        ...
        Q_PROPERTY( Priority priority READ priority WRITE setPriority ) // WRONG
        Q_ENUMS( Priority ) // WRONG
        enum Priority { High, Low, VeryHigh, VeryLow };
        void setPriority( Priority );
        Priority priority() const;
        ...
    };

Work around this limitation by declaring all properties at the beginning of the class declaration, right after Q_OBJECT:

    class SomeClass : public TQObject {
        Q_OBJECT
        Q_PROPERTY( Priority priority READ priority WRITE setPriority )
        Q_ENUMS( Priority )
    public:
        ...
        enum Priority { High, Low, VeryHigh, VeryLow };
        void setPriority( Priority );
        Priority priority() const;
        ...
    };