System for generating and using programs in an object-oriented environment with a message dispatch architecture5991538Abstract An object-oriented development system of the present invention includes a development system, which may include, among other features, a compiler, a linker, standard libraries, class libraries, and a debugger. Methods of the present invention include constructing C++ classes having response functions--C++ class methods which process specific system messages of interest. More particularly, a C++ class includes a registry object--an object which associates the message of interest with a particular response function. The registry object includes C++ template definition, whereby the object includes a "generic" function, that is, one which is not tied to any specific parameter type. In this manner, the message-response functions of the development system do not require compiler-specific extensions or unsafe casting operations. Claims What is claimed is:
______________________________________
template <class T>
void register( int msg, void (T::*func)(int, int) )
T::r.add(msg, func);
};
______________________________________
where msg represents an identifier of the message, *func represents an address of the method (function), T represents an identifier of the class, and (int, int) represents parameters passed to the method (function). 5. The process of claim 1, wherein said instructions include a register function receiving an identifier for the message and an address for the method. 6. The process of claim 1, wherein said message information includes an identifier for a single message, and wherein said method information includes a pointer (address) of the method. 7. In a system for creating programs executable in a computer, the computer including a message-driven operating system whereby events are communicated to the operating system and the programs by dispatching messages, a process for generating a computer program which processes a plurality of said messages, the process comprising: (a) providing source listings comprising data and methods operative on the data, said source listings specifying a method to be invoked in response to a selected message, said method to be invoked receiving a parameter related to the message; (b) storing within an executable program information representing said data and methods; and (c) storing within the executable program information indicating the method to be invoked in response to said selected message, said information including type-safe parameter information of the method, whereby said selected message is dispatched only to the method during execution of the computer program. 8. The process of claim 7, wherein said data and methods comprise at least one class of objects. 9. The process of claim 8, wherein said at least one class of objects includes a C++ class, wherein said data include data members of the C++ class, and wherein said methods include member functions of the C++ class. 10. The process of claim 9, wherein the C++ class includes a registry object for registering said selected message with the method. 11. The process of claim 10, wherein said information indicating the method to be invoked in response to said selected message includes an identifier for the message and a pointer (address) to the method. 12. The process of claim 8, wherein said information representing said data and methods includes instructions executable by a microprocessor. 13. The process of claim 7, wherein the message-driven operating system includes a graphical user interface (GUI). 14. The process of claim 13, wherein the graphical user interface includes a Microsoft.RTM. Windows-compatible interface and wherein said messages include Windows messages. 15. In a system for generating code executable in a computer from object-oriented source listings, the computer including a memory for storing data and a processor for performing methods operative on the data members, each method belonging to at least one class of methods, an improved process for message handling comprising: (a) creating a table for dispatching messages to selected ones of the methods; and (b) for each said method which is to be performed in response to a message, storing an entry in the table information identifying the message, its method, and the class for the method. 16. The process of claim 15 further comprising: dispatching the message to its method by reading the entry in the table. 17. The process of claim 15, wherein said method receives parameters identifying a state of the system upon occurrence of the message, the process further comprising: storing type-checking information for the parameters in the table entry, whereby parameters passed to the method are checked for type safety. 18. In a development system for creating message-driven programs from source listings, the source listings including at least one class having data members and function members, an improved process for handling messages, the improvement comprising; (a) selecting a message to be acted upon; (b) providing a function member of a class to be invoked upon occurrence of the message, the function member to receive at least one parameter for indicating a system state; and (c) storing in the program an entry for processing the message, the information including an identifier for the message, a reference to said function member, information identifying the class of the function, and information specifying a type for each said at least one parameter. 19. The process of claim 18, further comprising: (d) dispatching said message in response to a system event; (e) locating an entry in the program for the dispatched message by matching the message with an identifier stored in the program; (f) identifying the function member and its class from the located entry; and (g) executing the identified function member of the identified class with said at least one parameter, said at least one parameter having the specified type. 20. In a computer system having an event-driven architecture, the system operating under control of a program having at least one class of objects, each class specifying data and functions (methods) operative on data, and where events within the system are identified by dispatching messages, a process for making a function responsive to a message, the process comprising: (a) declaring the function to be a response function for the message by: (i) storing an identifier for the message as an entry in a response table, (ii) storing an address of the function in the entry, and (iii) storing information identifying the class of the function; (b) upon dispatch of the message, invoking the function of the class by referencing the entry in the response table. 21. The process of claim 20, wherein the function receives at least one parameter upon invocation, and wherein step (a) further includes: (iv) storing parameter type-checking information for the function in the entry, whereby parameters passed to the function are checked for type safety. 22. The process of claim 20, wherein step (a) includes: providing a function template for registering a selected message to the function of the class, the template having the general form of:
______________________________________
template <class T>
void register( int msg, void (T::*func)(int, int) )
T::r.add(msg, func);
};
______________________________________
where msg represents an identifier of the message, *func represents an address of the function, T represents an identifier of the class, and (int, int) represents parameters passed to the function. 23. A computer comprising: user input, a display and memory coupled to a central processor, said memory storing at least one program, said computer dispatching event messages from windows displayed on said display to said program, said program comprising a function template for providing said messages to a selected method of a class, said template comprising instructions for storing message, method, and class information together as an entry in a response table; said computer declaring a class to comprise user defined data and methods, said class comprising a registry object for dispatching messages to methods of said class, said registry object including said instructions from said function template, whereby message, method, and class information are stored together as an entry in said response table for each method of a class that is responsive to a message. Description COPYRIGHT NOTICE
______________________________________
while (GetMessage (&msg, NULL, 0, 0))
{
TranslateMessage (&msg) ;
DispatchMessage (&msg) ;
}
return msg.wParam ;
______________________________________
where a message (&msg) is retrieved by a call to GetMessage (step 263); if needed, the retrieved message may be translated by a call to TranslateMessage and then dispatched by a call to DispatchMessage (step 269). This "while" loop continues until the GetMessage function returns a value of zero--indicating that the loop has read a WM.sub.13 QUIT message from the queue, telling the application to end (yes at step 265). The general mechanism for retrieving and dispatching messages in an event-based system, such as Microsoft Windows, is known in the art; see, e.g., Petzold, C., Programming Windows, Second Edition, Microsoft Press, 1990. Additional information can be found in Microsoft's Window Software Development Kit, including: 1) Guide to Programming, 2) Reference, Vols. 1 and 2, and 3) Tools, all available from Microsoft Corp. of Redmond, Wash. The disclosures of each of the foregoing are hereby incorporated by reference. C. Development System Referring now to FIG. 3A, the development system 300 of the present invention will now be described in further detail. System 300 includes an compiler 320, a linker 350, and an interface 310; with additional reference to FIG. 3B, the general operation of the system is as follows. Through interface 310, the user or developer supplies source listings 301 to the compiler 320. Interface 310 includes both command-line driven 313 and Integrated Development Environment (IDE) 311 interfaces, the former accepting user instructions through command-line parameters, the latter providing menuing equivalents thereof. From the source code 301 and header/include files 330, the compiler 320 "compiles" or generates object modules or files 303. Upon successful creation of object (.obj) files, linker 350 next "links" or combines the object files 303 with standard libraries 360 (e.g., graphics, I/O routines, startup code, and the like) to generate program(s) 305, which may be executed by a target processor (e.g., processor 101 of FIG. 1). In addition to standard libraries, development system 300 provides class libraries 365, C++ libraries which simplify windows development (as described in further detail hereinbelow). A debugging module 370 may be added, as desired, for eliminating errors in the program listings 301. In a preferred embodiment, system 300 includes Borland C++ & Application Frameworks.TM. 3.1, available from Borland International of Scotts Valley, Calif. For further discussion of the general construction and operation of compilers; see e.g., Fischer et al., Crafting a Compiler with C, Benjamin/Cummings Publishing Company, Inc., 1991, the disclosure of which is hereby incorporated by reference. Simplified Windows Development Object-oriented Programming A. Introduction When one examines what makes Windows programming difficult, two things stand out: the event-driven model and the voluminous code required to make anything happen. Object-oriented programming (OOP) provides a simplified approach to both. With an object-oriented language, the programmer does not have to write a lot of tedious Windows message-dispatching code with special case arms for different types of messages. Instead, he or she creates reusable objects which know how to respond directly to individual messages. Of particular interest to the present invention is an object-oriented environment supporting the C++ programming language, which includes features of data encapsulation, inheritance, and polymorphism. For a general introduction to C++ , see e.g., Ellis, M. and Stroustrup, B., The Annotated C++ Reference Manual, Addison-Wesley, 1990. Additional information about object-oriented programming and C++ in particular can be found in Borland's C++ 3.1: 1) User's Guide, 2) Programmer's Guide, and 3) Library Reference, all available from Borland International of Scotts Valley, Calif. The disclosures of each of the foregoing are hereby incorporated by reference. B. Object-oriented approach to messages MS-Windows itself is not an object-oriented system, but it possesses characteristics which facilitate application development with OOP tools. Most notably, Windows is an event-driven or message-based system; its messages are analogous to the OOP concept of sending messages to objects. Combining the event-driven nature of Windows with an OOP language creates a single, more consistent development model. Just as a high-level language frees a programmer from the need to understand machine-level details, an OOP language, by hiding Windows details, eases the task of learning an event-driven architecture. To write Windows applications in an OOP language, one preferably defines "response methods"--special class methods (member functions) that handle messages for the class. More particularly, a window object is instantiated with message-response member functions, that is, methods which know how to process messages for that window. Message response functions are usually not called directly but instead are invoked automatically in response to events of the user or system. This basic idea holds true whether one is trapping user input or responding to system messages, DDE messages, or other Windows messages. In operation, a message is automatically routed to a virtual function that has been assigned an identifier corresponding to the message (using a special dynamic dispatching syntax). As seen in the following C++ class definition, for instance, response methods may be defined within a class (e.g., TMyWindow) as:
______________________________________
.sub.-- CLASSDEF (TMyWindow)
class TMyWindow : public TWindow
public:
TMywindow(PTWindowsObject AParent, LPSTR ATitle)
: TWindow(AParent, ATitle) {};
virtual BOOL CanClose();
virtual void WMLButtonDown(RTMessage Msg)
= [WM.sub.-- FIRST + WM.sub.-- LBUTTONDOWN];
virtual void WMRButtonDown(RTMessage Msg)
= [WM.sub.-- FIRST + WM.sub.-- RBUTTONDOWN];
};
/*
The addition of the constant WM.sub.-- FIRST to the command
ID constant guarantees that the IDs do not conflict with
standard Windows message constants.
*/
______________________________________
Here, two message-response functions are defined: WMLButtonDown() and WMRButtonDown(); the former responds to WM.sub.13 LBUTTONDOWN messages, while the latter responds to WM.sub.13 RBUTTONDOWN messages. At the compiler level, the message is dispatched to its corresponding response function through use of a dynamic dispatch virtual table (DDVT). Based on the message identifier appended to its definition, a corresponding virtual function is referenced and invoked. In other words, dynamic dispatching triggers an action (member function) in response to capturing a message of interest. The dynamic dispatching extension to C++ , based on the C++ "pure virtual" syntax, eliminates the unreadable, C-style case statement approach, thus facilitating the handling Windows messages. Moreover, within a response function itself, one only has to write the function body as he or she would normally do in a non-event driven system. Using message-response functions, while undoubtedly better than the standard, case-statement technique, is not a perfect solution. In particular, the technique requires use of an extension to the base language, C++ , which is arguably compiler specific. Hence, code written using the technique consequently may have limited portability to other environments. Moreover, all the message-response functions require the same list of argument types, notably a message (msg) which must be further dissected into components of interest (e.g., 1param or wParam); preferably, a function would instead take parameters which are actually utilized within the function. As an alternative to DDVT-invoked functions, one could define a "message map" (implemented through a macro) within a class definition, for example:
______________________________________
class C : public TWindow
public:
void move (int, int);
BEGIN.sub.-- MESSAGE.sub.-- MAP // macro begin
ON.sub.-- MOVE(move) // on WM.sub.-- MOVE msg call move()
END.sub.-- MESSAGE.sub.-- MAP // macro end
};
______________________________________
with the macro including facility for calling the move() function and splitting up the parameters passed to the function in a manner appropriate for the message (here, WM.sub.13 MOVE). The macro operates by storing the address of the member function (here, &C::move) together with the message number associated with it. While message maps are a rather straightforward implementation, they have distinct disadvantages. Declared within the class definition itself, the syntax of the macro is awkward at best and perhaps even illegal. Worse yet, the macro relies the rather outdated practice of casting--here, casting the address into a pointer-to-member function of the base class (where dispatching actually occurs). As is well known in the art, casting defeats the type checking which ordinarily would be provided by the compiler and is to be avoided. In the example at hand, casting would conceal errors in the argument list, creating difficult-to-track program bugs. Moreover, present day implementations of message mapping do not support multiple inheritance. A better solution is desired. Improved Message Response Functions C++ Templates A. Introduction to Templates According to the present invention, methods are provided for message dispatching using C++ templates. Before describing specific message dispatching methods of the present invention, however, it is helpful to first review templates--a recent extension to the proposed ANSI C++ syntax. The concept of C++ templates is perhaps best introduced by reviewing the age-old problem of providing a "generic" function, that is, one which is not tied to any specific parameter type. Consider, for instance, the problem of providing a function max(x,y) that returns the larger of its two arguments, x and y, which can be of any data type (that can be ordered). In a strongly typed language such as C++ (and the a lesser extent ANSI C), the problem, of course, is that the function expects the types of the parameters to be declared beforehand, that is, at compile time. The classic standard C approach to this problem has been to use a macro, for instance:
______________________________________
#define max(x,y) ((x) > (y)) ? (x) : (y))
______________________________________
This works, but it creates a new problem. Specifically, the approach circumvents the type-checking mechanism provided by ANSI C and C++ . And in the latter, use of macros is almost obsolete. Clearly, the intent of max(x,y) is to compare compatible types; unfortunately, using the macro allows a comparison between data types (e.g., an unsigned int and an int) with unintended results. Consider max(1u, -1), for instance. At the bit level, a comparison between 0x01 and 0xFF is made. Since 0xFF>0x01, such a comparison would return the erronous result that -1 is the larger number. As another problem, the substitution to be performed may not be what is desired. Consider the macro expansion in a C++ class definition, for instance:
______________________________________
class Foo
public
int max(int, int); // Results in syntax error; this gets expanded!!!
// . . .
};
______________________________________
Here, int max (mint, int) is a method declaration of class Foo; its expansion here is unintended, leading to a hard-to-diagnose error. Of course, one could create separate functions for each type, for instance declaring:
______________________________________
int max.sub.-- int(int x, int y)
{
return (x > y) ? x : y;
}
long max.sub.-- long(long x, long y)
{
return (x > y) ? x : y;
}
float max.sub.-- float(float x, float y)
{
return (x > y) ? x : y;
}
______________________________________
But this is highly awkward at best. The approach is prone to error as the programmer would have to undertake extra effort to match each data pair with one of three different function names (i.e., max.sub.13 int(), max.sub.-- long(), or max.sub.13 float()). The approach is unworkable for all but the simplest of programs. With C++ , programmers have another solution--overloading. Overloading is a technique which allows the system to correctly determine, based on parameter type passed at runtime, which version to invoke. For example, many overloaded versions of maxo could be provided, one for each data type to be supported. For example,
______________________________________
int max(int x, int y)
{
return (x > y) ? x : y;
}
long max(long x, long y)
{
return (x > y) ? x : y;
}
float max(float x, float y)
{
return (x > y) ? x : y;
}
______________________________________
In each instance, regardless of parameter type, only a single function, maxo, is called. As the code for each version is essentially identical, however, the code is highly duplicative. A better solution is desired. C++ templates, also called generic or parameterized types, provide the best solution to the problem. By using a template, one can define a pattern for a family of related overloaded functions by letting the data type itself be a parameter. Thus instead of the programmer explicitly coding several overloaded functions, the compiler creates the needed function based on the parameter types which are provided at runtime. Consider, for example, the following function template definition:
______________________________________
template <class T>
T max(T x, T y)
{
return (x > y) ? x : y;
};
______________________________________
where the data type is represented by the template argument, <class T>. When used in an application, the compiler generates a call to the appropriate function according to the data type actually passed:
______________________________________
int i;
Myclass a, b;
int j = max(i,0); // arguments are integers
Myclass m = max(a,b); // arguments are type Myclass
______________________________________
Any data type, not just a class, can be used for <class T>. The compiler calls the appropriate operator>(), so the programmer can use maxo with arguments of any type for which operator>() is defined. Templates can be further divided into function templates and class templates. The foregoing example is generally called a function template; the argument type(s) of a template function use all of the template formal arguments. If not, there is no way of deducing the actual values for the unused template arguments when the function is called. A specific instantiation of a function template is called a template function. A class template (also called a generic class or class generator), on the other hand, allows one to define a pattern for class definitions, such as generic container classes.
______________________________________
// Vector is a class template.
//
template <class T> class Vector
T *data;
public:
Vector( int i );
T& operator [ ] ( int );
};
______________________________________
From this class template, the following classes are created (upon instantiation with an object):
______________________________________
// Vector<int> is a class, as are Vectar<long> and Vector<float>.
//
int main( )
Vector<int> vi;
Vector<long> vl;
Vector<float> vf;
return 0;
}
______________________________________
For further discussion of the operation of C++ templates, including class templates, the reader should refer to Chapter 14 (Templates) of The Annotated C++ Reference Manual, supra. B. Template-driven Response Functions 1. Introduction According to the present invention, message-response functions are provided which eliminate reliance on DDVTs. In particular, the message-response functions that are provided by the development system 300 do not dependent on compiler-specific extensions nor unsafe casting operations. An exemplary class with response capability may be constructed from the following class definition:
______________________________________
class C : public TWindow
public:
void move(int, int);
static registry r; // this object understands the dispatching
};
______________________________________
As shown, the class includes its own registry object, r, which understands the dispatching of messages. The function template which registers a member function, may be constructed as:
______________________________________
template <class T>
void register( int msg, void (T::*func)(int, int) )
T::r.add(msg, func);
};
______________________________________
Like the macro operation above, the template provides the address of a member function. Unlike the foregoing macro, however, no particular class need be specified. Instead, a desired function can be registered in the right class, without having to cast it into a pointer to the base class and without discarding argument type information. In essence, the compiler can determine which class is referenced and, hence, which registry object shall be invoked. Thus by preserving class information and argument type information through the use of templates, the compiler can generate the right function call without relying on unsafe casting operations. Referring now to FIG. 4A, a general overview of message dispatching in accordance with the present invention is illustrated. In step 410, the user defines his or her own C++ class; the class includes at least one method (function) for responding to a message, and including an object--the registry object--which knows how to respond to the message (by calling an appropriate response method). At step 420, a "register" function template is provided to simplify the registering of the message/method pair for the class of interest. As shown by its prototype, the register function takes two parameters: an identifier (msg) for the message and a pointer to the function which knows about that message (i.e., how the message or its parameters are to be interpreted). More particularly, the function calls an add method (i.e., generic container function) for registering message/response function pairs; as shown, the class information is retained with the message. At step 430, a particular message of interest (e.g., WM.sub.-- MOVE) is dispatched in accordance with the template definition of step 420, whereupon the message invokes the proper response function. 2. Response Tables In a preferred embodiment, message dispatching is effected through "response tables" of the present invention. As shown in FIGS. 4B-C, a user-defined class 451 has associated with it a single response table, TResponseTable 453. Thus, a 1:1 relationship exists between a user class and its response table. The detailed construction and operation of TResponseTable will now be described. a) Class TResponseTable As shown by the block diagram 450 of FIG. 4B, TResponseTable, which is preferably implemented as a C++ template class, primarily functions to carry class hierarchy information about its user class. When instantiated, it stores a pointer to an array of response table entries; the entries themselves are stored in TResponseTableEntry 455 (described hereinbelow). It also stores a list of pointers to one or more base classes, thereby supporting multiple inheritance. As described below, single inheritance is treated as a special case. This approach minimizes overhead for users who do not want to use multiple inheritance, yet makes the dispatching of messages extensible. In an exemplary embodiment, a TResponseTable class may be constructed from the following C++ class template:
______________________________________
template <class T> class TResponseTable : public TAnyResponseTable {
public:
const BOOL isSI; // is single-inheritance (tag for union)
TResponseTableEntry<T> *entries;
union {
TAnyResponseTable *base;
BI.sub.-- ICVectorImp<TMIBaseForT<T> > *bases;
};
// construct a single-inheritance response table by supplying the base
// class (NULL terminated list of response table entries is retrieved
// automatically)
//
TResponseTable (TAnyResponseTable *b) : isSI (TRUE),
entries (T::.sub.-- entries) (base = b;)
// construct a multiply-inherited response table (NULL terminated list
// of response table entries is retrieved automatically)
TResponseTable ( ) : isSI (FALSE),
entries (T::.sub.-- entries) {bases = new
BI.sub.-- ICVectorImp<TMIBaSeForT<T> >(2, 1);}
};
______________________________________
In this fashion, the TResponseTable 453 may be employed to dispatch messages to specific functions. Specifically, TResponseTable 453 calls into TResponseTableEntry 455, which stores information referencing the response method. Since the response table 453 "knows" (stores a reference to) which class it is associated with (e.g., user class 451), it references the correct response table entries stored in TResponseTableEntry 455. In the event that a match is not found, the response table, storing the class hierarchy information for its user class, may locate a base class having the response method of interest. b) Class TResponseTableEntry According to the present invention, the dispatcher should be extensible, yet compile-time type checking should be provided. In an exemplary embodiment, a TResponseTableEntry class may be constructed as follows:
______________________________________
class TResponseTableEntry {
public:
UINT msg, id;
virtual LRESULT Dispatch (GENERIC &, WPARAM, LPARAM) = 0;
TResponseTableEntry (UINT m, UINT i) : msg (m), id (i) { }
};
______________________________________
This approach supports multiple inheritance and is extensible. In addition, type checking may be added in derived classes to ensure that the pointer to member function the user supplies has a correct signature. More importantly, the array of response table entries can be built statically by the compiler, thereby eliminating any overhead in code size and runtime initialization. As shown, TResponseTableEntry 455 includes a Dispatcho method, which takes a generic reference (GENERIC &) and windows parameters (LPARAM and WPARAM) as its parameters. Since TResponseTableEntry class 455 is an abstract class (i.e., it includes at least one pure virtual function), at least one derived class must be created. Specifically, a class is derived which overrides the virtual dispatch function, adding specific knowledge for processing the message parameters associated with the message of interest (i.e., for "cracking out" or extracting the information required for its specific message type). For instance, a TCommonExtractPoint class 461 for extracting "Point" information from Windows may be created from the following C++ template:
______________________________________
template <class T> class TCommonExtractPoint : public
TResponseTableEntry {
protected:
// extract "POINT &" from lParam
typedef void (T::*PMF) (POINT &);
PMF pmf;
public:
LRESULT Dispatch (T &object, WPARAM, LPARAM lParam)
{(object.*pmf) ((POINT &) lParam); return TRUE;}
TExtractCommonPoint (UINT msg, UINT id; PMF p);
// store pointer-to-member in "p"
};
______________________________________
For TCommonExtractPoint 461, the Windows message parameter lParam includes a point (screen point) of interest; thus, the cracker need only extract this point from the lParam data structure (e.g., by casting lParam to a Windows POINT type). In a preferred embodiment, the system provides default message crackers for standard Windows messages. Next, a family of template classes are derived, one for each type of message cracking. The template classes do not know which specific classes they are invoked with, but they do provide the correct parameter type and parameter type safety. The desired type checking is provided by making the derived classes template classes and their constructors accept a pointer to method function with an appropriate signature. For instance, template class 463 may be created from TCommonExtractPoint as follows:
______________________________________
template <class T> class TExtractpoint : public TCommonExtractpoint {
public:
typedef void (T::*TPMF) (POINT &);
TExtractPoint
(UINT msg,
UINT id,
TPMF pmf) : TCommonExtractPoint
(msg, id,
(PMF &)
pmf) { }
};
______________________________________
From the template classes, various dispatchers (i.e., those which use this particular message cracking) are derived which are type safe as to class. A simple example demonstrates this use:
______________________________________
class A {
public:
void Func (POINT &);
static TResponseTableEntry *.sub.-- entries[ ];
};
TResponseTableEntry *A::.sub.-- entries[ ] = {new TExtractPoint<A>
(0, 0, &A::Func));
______________________________________
Thus, any function for which the lParam is a point is registered as a TExtractPoint, which is derived from TCommonExtractPoint 461. As shown, the derived dispatchers include type-safe constructors. In this fashion, a family of classes may be derived from TCommonExtractPoint 461, yet differ in that each has its own type safe constructor. As shown by the TExtractPoint class template, TPMF provides type safety. In particular, TPMF is a pointer to member of class T which takes a point reference (POINT &). The parameter of type TPMF is passed down to TCommonExtractPoint 461, but only after it has first been cast to PMF--a pointer to member function which is generic. Hence, the class information is no longer needed when this is passed to TCommonExtractPoint. In this manner, the constructor is initialized to point to a function which is passed in with the constructor. The above approach benefits from extensibility and type safety. Extensibility is achieved using delegation, not inheritance (as before). The pointer to function that is used to do the cracking/dispatching--TAnyDispatcher--may be constructed as:
______________________________________
class GENERIC;
typedef void (GENERIC::*TAnyPMF) ( );
//
// all message cracker functions take four parameters:
//
reference to a "generic" object
//
pointer to "generic" member function (signature varies
// according to the cracking that the function performs)
//
WPARAM
//
LPARAM
//
typedef LRESULT (*TAnyDispatcher) (GENERIC &, TAnyPMF,
WPARAM, LPAMAM);
______________________________________
A family of functions to do message cracking/dispatching, one for each type of signature, is provided. According to the present invention, type checking is also provided. Perhaps the most important type checking to provide is that which ensures that the pointer-to-member function that is used has a signature that is appropriate for the message cracking performed. This is the where the the greatest opportunity for error exists. Since the cracking/dispatching is done using delegation, class TResponseTableEntry does not know in advance what signatures to expect and declares the pointer to member function as returning no argument and having no formal parameters. Macros are provided for building a response table entry which will perform a cast of the user supplied pointer to member function to the type expected by class TResponseTableEntry (i.e,. void return and no formal parameters). Preferably, the macros check that the signature matches that expected by the dispatcher function. This can be accomplished using template functions: for every dispatching/cracking function there will be a corresponding template function that accepts a pointer-to-member function of the correct signature and simply returns it. Using the example of a dispatcher that "cracks" wParam and lParam to produce a POINT which it passes to the pointer-to-member function, a template function v.sub.13 POINT.sub.13 Sig()may be constructed from:
______________________________________
template <class T>
inline void (T::*v.sub.-- POINT.sub.-- Sig (void (T::*pmf) (POINT &)))
(POINT &)
return pmf;
}
______________________________________
v.sub.13 POINT.sub.13 Sig()is a template function that has two formal parameters: a pointer-to-member function of class T (where T is the template argument) and a reference to a POINT; the function includes a void return. Preferably declared in-line, the function simply returns its actual parameter. A definition of an array of response table entries for a class "B", for example, may be constructed from:
__________________________________________________________________________
class B {
public:
void MouseMove (POINT &);
};
TResponseTableEntry<B> entries[ ] =
{{WM.sub.-- MOUSEMOVE, 0, (TAnyDispatcher) v.sub.-- POINT.sub.--
Dispatch,
(PMF) v.sub.-- POINT.sub.-- Sig (&B::MouseMove)}};
__________________________________________________________________________
This approach correctly ensures that the pointer-to-member function has the correct signature. At this point, assuring type safety is a simple matter of using the macros that are used to create a response table entry aggregate. c) Class TEventHandler According to the present invention, multiple inheritance is provided by introducing class TEventHandler into the class hierarchy:
______________________________________
// class from which to derive classes that can handle events
//
class TEventHandler {
};
Class TWindowsObject will be changed to derive from TEventHandler,
class TwindowsObject : virtual public TEventHandler
Class TEventHandler will have one method:
Dispatch( )
Function Dispatch is declared as follows:
LRESULT
Dispatch (GENERIC &object,
TResponseTableEntry<GENERIC>
&entry,
WPARAM wParam,
LPARAM lParam);
______________________________________
Given a response table entry and a reference to an object of class GENERIC, Dispatcho invokes the dispatcher/cracker pointer-to-function passing parameters wParam and lParam as arguments, as follows:
______________________________________
return (*entry.cracker) (object, (TAnyPMF &) entry.pmf, wParam,
lParam);
______________________________________
d) TEventHandler and Class GENERIC TEventHandler is defined to be the base class for all classes that want to handle events. The real purpose of TEventHandler is to serve as a base class for multiply inherited branches of the hierarchy. Accordingly, class TResponseTableEntry accepts a pointer-to-member function of classTEventHandler; the cracker/dispatcher functions, in turn, operate on TEventHandler(s), for instance:
______________________________________
class TResponseTableEntry {
public:
typedef void
(TEvent.Handler::*PMF) ( );
UINT msg;
UINT id;
TAnyCracker
cracker;
// pointer to func to do message cracking
PMF pmf; // and dispatch
};
______________________________________
Since one cannot convert a pointer-to-member function to be of type pointer-to-member function of a virtual base class (because the inverse operation from virtual base class back to derived class is not valid), the approach does not work. Consider the following example:
______________________________________
class A : virtual public TEventHandler { };
class E : virtual public TEventHandler { };
class C : public A, public B {
public:
void Func ( );
};
void (A::*pmt) ( ) = (void (A::*) ( )) &C::Func; // invalid!!!
______________________________________
Declaring TEventHandler to not be virtual does not help either. In that instance, class C would have two copies of class TEventHandler; as a result, class TEventHandler would be ambiguous (as to which one to use when a message is received). To overcome this problem, the present invention includes a class GENERIC. All pointer-to-objects and pointer-to-member functions are cast to be of type class GENERIC. This is a safe cast, which generates no code, converts the pointers to untyped pointers. This is safe because the type of function is also stored. The saved type information allows the call to be done correctly. As an alternative, TResponseTableEntry could be constructed as a template class that expects a pointer-to-member function of class "T", with the cracker/dispatcher specifications changed as follows:
______________________________________
// extracts a POINT & from lParam and pass it to "pmf"
//
void ExtractPointCracker
(TEventHandler *object,
void (TEventHandler::*pmf) (POINT &),
WPARAM,
LPARAM);
______________________________________
In this instance, the cracker/dispatcher functions are, from their perspective, being called with pointers to an object of class TEventHandler and a pointer-to-member function of class TEventHandler. Since, in reality, this is not the case, it is safer to treat it as a pointer to an unknown class; that way the compiler does not make any false assumptions (particularly, in trying to optimize the pointer-to-member function call). e) Macro Simplification In a preferred embodiment, the end user/programmer is shielded from unnecessary implementation details by the use of macros. Two macros, in particular, are provided for declaring a response table: DECLARE.sub.13 RESPONSE.sub.13 TABLE and DECLARE.sub.13 NV.sub.13 RESPONSE.sub.13 TABLE. The former is used declaring a response table for a class derived from TWindowsObject, the latter for a class which is not derived from TWindowsObject. The two macros may be defined as follows:
______________________________________
#define DECLARE.sub.-- NV.sub.-- RESPONSE.sub.-- TABLE(cls).backslash.
public:.backslash.
static TResponseTableEntry<cls>
.sub.-- entries[];.backslash.
typedef TResponseTableEntry<cls>::PMP
TMyPMF;.backslash.
static TResponseTable<cls>
responseTable
#define DECLARE.sub.-- RESPONSE.sub.-- TABLE(cls).backslash.
DECLARE.sub.-- NV.sub.-- RESPONSE.sub.-- TABLE (cls);.backslash.
virtual TResponseTable<TWindowsObject> &GetResponseTable ().backslash.
{return (TResponseTable<TWindowsObject> &) responseTable;}.backslash.
virtual GENERIC *ToDerived () {return (GENERIC *) this;}
______________________________________
Macro DECLARE.sub.13 RESPONSE.sub.13 TABLE, after invoking the macro DECLARE.sub.13 NV.sub.-- RESPONSE.sub.13 TABLE for defining an array of response entries and a response table, defines two the virtual functions: GetResponseTable() and ToDerived(). 3. Response Table Entries In a preferred embodiment, one macro is defined per Windows message. For a generic message that is passed a wParam and lParam without doing any message cracking, the macro may be defined as follows:
__________________________________________________________________________
#define EV.sub.-- MESSAGE(method, message).backslash.
{message, 0, (TAnyCracker) &::NoCrackingDispatcher, (TMyPMF)
__________________________________________________________________________
&method}
As shown, the macro hides the cast of the pointer-to-member function and specifies the cracker/dispatcher function that is to be used. In an exemplary embodiment, all macros begin with EV.sub.13 (for event) and use a predetermined method name signifying their actions. For a Windows WM.sub.13 PAINT message, for instance, the following macro could be declared:
__________________________________________________________________________
#define EV.sub.-- WM.sub.-- PAINT.backslash.
{WM.sub.-- PAINT, 0, (TAnyCracker) &::VoidVoidDispatcher, (TMyPMF)
&EvPaint}
__________________________________________________________________________
where VoidVoidDispatcher passes no arguments and expects no return value. An exemplary definition of the response table and array of response table entries without a macro for class "TMyWin" (derived from base class "TWin") with two message entries may be constructed as:
__________________________________________________________________________
TResponseTable<TMyWin> TMyWin::responseTable (&TWin::responseTable);.backs
lash.
TResponseTableEntry<TMyWin> TMyWin::.sub.-- entries[] = {EV PAINT,
EV.sub.-- LMOUSEDOWN,
EV.sub.-- MESSAGE (0, 0, 0);}
__________________________________________________________________________
The last entry is a NULL terminator. Thus, for:
______________________________________
#define DEFINE.sub.-- RESPONSE.sub.-- TABLE(cls,base).backslash.
TResponseTable<cls> cls::responseTable(&base::responseTable);.backslash.
TResponseTableEntry<cls> cls::.sub.-- entries[] = .backslash.
{
#define END.sub.-- RESPONSE.sub.-- TABLE.backslash.
{0, 0, 0, 0}}
______________________________________
using the macros, the definition is simplified as:
______________________________________
DEFINE.sub.-- RESPONSE.sub.-- TABLE (TMyWin, TWin)
EV.sub.-- PAINT,
EV.sub.-- LMOUSEDOWN,
END.sub.-- RESPONSE.sub.-- TABLE;
______________________________________
While the invention is described in some detail with specific reference to a single preferred embodiment and certain alternatives, there is no intent to limit the invention to that particular embodiment or those specific alternatives. Thus, the true scope of the present invention is not limited to any one of the foregoing exemplary embodiments but is instead defined by the following claims. Appendix Multiple Inheritance With multiple inheritance, the system should ensure that it has the correct "this" pointer when invoking a pointer-to-member function. This is accomplished by using virtual function ToDerivedo defined by macro DECLARE.sub.13 RESPONSE.sub.13 TABLE and virtual functions ToBase() defined by wrapper class TMIBase. The virtual functions convert a TWindowsObject pointer to the most derived class pointer using virtual function ToDerivedo. Thus at this point, the correct pointer needed is available. The need for this approach is illustrated by the following scenario:
______________________________________
class TMyWindow : public TMixIn, public TWindowsObject {
};
______________________________________
Since the user has specified class TMixIn before specifying class TWindowsobject (assuming the compiler layout of objects is in the order of TMixIn, TWindowsObject, and TMyWindow), a (TMyWindow *) is not the same as a (TWindowsObject *) pointer. Therefore, the virtual function ToDerived() is preferably used to obtain a pointer to the most derived object. Upon obtaining the correct pointer to start with, the response table is searched for a match. If one is not found, then the base class response tables is search. When the class has more than one base class, the "this" pointer is converted as appropriate to point to the correct sub-object for each base class that is searched; the ToBase() virtual function is employed for that purpose. The process continues up the class hierarchy in breadth-first order. Exemplary Multiple Inheritance Macros Multiple inheritance entails two additional macros. If one's class has no base class, one employs the macro END.sub.13 RESPONSE.sub.13 TABLE to signify the end:
______________________________________
#define DEFINE BASE RESPONSE.sub.-- TABLE(cls).backslash.
TResponseTable<cls>
cls::responseTable(0);.backslash.
TResponseTableEntry<cls>
cls::entries[ ]=.backslash.
{
______________________________________
For a class with multiple base classes, one uses the following macro to invoke a parameterless TResponseTable constructor (using the macro END.sub.13 RESPONSE.sub.13 TABLE to signify the end as like before):
______________________________________
#define DEFINE.sub.-- MI.sub.-- RESPONSE.sub.-- TABLE(cls).backslash.
TResponseTable<cls>
cls::responseTable;.backslash.
TResponseTableEntry<cls>
cls::.sub.-- entries[ ] =.backslash.
______________________________________
Thus, a multiple inheritance example for class TMyWin, which has two base classes TMyWin and TMixIn, may be constructed from:
______________________________________
DEFINE.sub.-- MI.sub.-- RESPONSE.sub.-- TABLE (TMyWin)
EV.sub.-- MESSAGE (Func, 15),
END.sub.-- RESPONSE.sub.-- TABLE;
ADD.sub.-- BASE.sub.-- TO (TMyWin, TMixIn);
ADD.sub.-- BASE.sub.-- TO (TMyWin, TWin);
______________________________________
Macro ADD.sub.13 BASE.sub.13 TO is provided for building a data structure for multiple inheritance and adding it to the collection of bases:
______________________________________
#define ADD.sub.-- BASE.sub.-- TO(derived, base).backslash.
static TMIBase<derived,base>.sub.-- ##derived##.sub.-- BASE.sub.--
______________________________________
##base
Finally, class TMiBase and its base class TMiBaseForT may be built as follows:
______________________________________
template <class T> class TMIBaseForT {
public:
BOOL operator == (const TMIBaseForT<T> &) const (return TRUE;}
};
//
// here's the class is used to construct a wrapper
// for a derived class "T" that has a base class "B"
//
// declaring an instance of this class will add the base
// to the list of base classes in "T's" response table
//
template <class T, class B> class TMIBase : public TMIBaseForT<T> {
public:
TAnyResponseTable *base;
//
// type-safe conversion from pointer to class "T" to pointer
// to base class "B"
//
virtual B *ToBase (T *object) {return object;}
TMIBase ( ) : base (&B::responseTable)
{T::responseTable.bases->add (this);}
};
______________________________________
Class TMIBaseForT exists for the purpose of type checking, ensuring that the base class wrappers are of the same type as the response table.
|
Same subclass Same class Consider this |
||||||||||