Programming with Objects and Classes
Overview
FPC includes several language extensions to "standard" Pascal implementations for object oriented programming.
- Objects (chapter 5)
- Classes (chapter 6)
- Interfaces (chapter 7)
- Generics (chapter 8)
These provisions are described in the indicated chapters of the FPC Language Reference Guide: http://www.freepascal.org/docs.var which includes syntax diagrams and further information not contained in this introductory guide. Of these four language features, Objects and Classes form the basis of object oriented programming (OOP) in FPC and for Lazarus. Users migrating from Delphi may want to skip the section on Objects and go to Classes since the Classes implementation/version of OOP is based on Delphi syntax. Conversely, users familiar with older Turbo Pascal or Freevision may want to skip the section on Classes since the Objects implementations is based on the syntax of the latter Pascal dialects. For users familiar with the Apple or THINK Pascal dialects on the Mac OS, neither the Objects or Classes implementation provides a direct migration path. In general, the Classes implementation seems to be more widely in use and is used for Lazarus and also in a future bridge being worked on interfacing Apple's Objective C / Cocoa framework to FPC.
General Concepts
OOP provides different ways to manage and encapsulate data and to manage program flow compared with other available programming language features. This method of programming often lends itself to modeling certain applications such as Graphic User Interfaces and physical systems in a more facile manner. However it is not appropriate for all applications. Program control is not as explicit as using the more standard Pascal procedural constructs and to get the most benefit of OOP, understanding of large class libraries is often required. Also, maintaining large OOP application code has its advantages and disadvantages compared to maintaining procedural code. There are many sources for learning OOP and OO design which are beyond the scope of this guide.
There are a lot of programming languages which incorporate OOP features as extensions or the basis of their language. As such, there are many different terms for describing OO concepts. Even within FPC, some of the terminology overlaps. In general, OOP usually consists of the concept of a programming object (or information unit) which explicitly combines and encapsulates a set of data and procedural code which are related. Usually, this data is persistent during program execution without having to be declared globally. In addition, there are usually features which enable additional objects to be incrementally modified and/or extended based on previously defined objects which is usually referred to by the term "inheritance." Many languages refer to the procedures which belong to an object as an object "method." Much of the power of OOP is realized by late dynamic binding of methods at run time rather than at compile time. This dynamic binding of methods is similar to using procedural variables and procedural parameters but with greater syntactic cohesion with the data it is related to.
Objects - Basics
Of the two OOP implementations FPC provides, the one used less often seems to be what is referred to as "Objects" which probably gets its name from the type definition syntax. The syntax diagram for an object type declaration can be found in the FPC language reference - Chapter 5. Basically, an object type looks like a record type with additional fields including procedure fields and also optional keywords which indicate the scope of the fields. In fact, an oversimplified (but valid) simple object is hard to distinguish from a record structure as can be seen below.
<delphi> Type
MyObject = Object
f_Integer : integer;
f_String : ansiString;
f_Array : array [1.3] of char;
end;
</delphi>
The only difference in the above example is that the Pascal keyword record has been replaced with the keyword object. Things start to be different when methods are added to the object. Object methods are declared in FPC using the keywords procedure or function. These object methods (procedures or functions) are declared the same way as normal Pascal procedures and functions; just that they are declared within the object declaration itself. Lets create a different object; this time possibly as part of an oversimplified graphics drawing program.
<delphi> Type
DrawingObject = Object
x, y : single;
height, width : single;
procedure Draw;
end;
Var
Rectangle : DrawingObject;
</delphi>
Besides the simple datatype fields providing some application specific location and size attributes, the object shown above declares an additional parameter; a procedure called Draw. The type declaration is followed by a variable identifier called Rectangle of the type DrawingObject. Next, the Draw procedural code itself needs to be written as well as well as code for accessing and manipulating the data fields, as well as how to invoke the Draw procedure. The following simple program shows how all this works. It should compile and run on any system with FPC 2.2.2 and above. Note: For Mac OS X, the -macpas compiler directive must be turned off.
<delphi> Program TestObjects;
Type
DrawingObject = Object
x, y : single;
height, width : single;
procedure Draw; // procedure declared in here
end;
procedure DrawingObject .Draw;
begin
writeln('Drawing an Object');
writeln(' x = ', x, ' y = ', y); // object fields
writeln(' width = ', width);
writeln(' height = ', height);
writeln;
// moveto (x, y); // probably would need to include a platform dependent drawing unit to do actual drawing // ... more code to actually draw a shape on the screen using the other parameters
end;
Var
Rectangle : DrawingObject;
begin
Rectangle.x:= 50; // the fields specific to the variable "Rectangle" Rectangle.y:= 100; Rectangle.width:= 60; Rectangle.height:= 40;
writeln('x = ', Rectangle.x);
Rectangle.Draw; // Calling the method (procedure)
with Rectangle do // With works the same way even with the method (procedure) field
begin
x:= 75;
Draw;
end;
end. </delphi>
As can be seen in the above program, the body of the Draw method (procedure) is declared after the object type declaration by concatenating the type identifier with the procedure name. In a more realistic situation, the object would likely be declared in the interface section of a separate unit while the procedure body would be written in the implementation section of the same external unit. In this example, only standard vanilla Pascal is used, but if actual graphic primitives are available, they can be used if desired. The second thing to notice is that inside the Draw method (procedure), the object's data fields are referenced as if they were regular local variables. The only difference to regular local variables is that the values of these fields will persist between calls to the Draw procedure.
In the main program, the fields are assigned values and accessed just like record fields are. Similarly, the Draw procedure is invoked using the same dot notation as the fields. And like records, the with keyword works the same for accessing object fields and invoking methods. Finally, notice that the second time the Draw method is called, all the fields persisted between calls; the only field different is the x attribute which was explicitly changed.
The above example is meant to show the basic mechanics of simple Objects. There are a couple of problems, however. The first problem is that the Draw method only draws one thing and that is a rectangle (although even that is questionable.) Additional methods could be declared in the DrawingObject object such as DrawRectangle, DrawCircle, DrawTriangle, etc..., but this would not be too much different than declaring separate regular procedures and having to use case statements to select the appropriate procedure. So doing it this way with objects would require more work (and is not how it is accomplished.) The other problem is, the object's fields are able to be accessed and modified anywhere in the program very similar to using global variables which makes it harder to write well encapsulated and robust programs. These problems will be addressed in the next section.
Objects - Static Inheritance
The next step will expand upon the previous code handle rectangles, squares, and triangles. An attempt will be made to leverage the existing code, if possible, by creating two new objects to handle the two new shapes. In order to make the code and output easier to follow, the main object type, DrawingObject has been renamed to TShape. Other object types will be named TRectangle, TSquare, and TTriangle with corresponding variables named Shape, Rectangle, Square and Triangle. The letter "T" is used as a prefix to the object type names since it is a commonly used convention in many object libraries and frameworks.
What follows are type declarations for: the main TShape object type (previously DrawingObject) along with the new types TRectangle and TSquare. TShape now includes a new method (procedure GetParams) to obtain values for its fields. Next, the types TRectangle and TSquare are declared. TShape is usually referred to as an ancestor or parent object, while TRectangle is said to be a child or subobject or subclass of TShape or that it descends from TShape. This parent child, grandchild heiracry is identified by the qualifier type name following the Object keyword. Similarly, TSquare is a child of TRectangle.
<delphi> Type
TShape = Object
x, y : single;
height, width : single;
procedure GetParams;
procedure Draw;
end;
TRectangle = Object(TShape)
procedure Draw;
end;
TSquare = Object(TRectangle)
procedure GetParams;
procedure Draw;
end;
Var
Shape : TShape; Rectangle : TRectangle; Square : TSquare;
</delphi>
Notice that TRectangle lists only the Draw procedure while TSquare lists both the GetParams and Draw procedures. Neither of the subobject types show any fields. The "missing" fields and procedure names are said to be inherited from the fields and procedures declared in the ancestor objects. In this case, any object variables declared and instantiated of type TRectangle will inherit all four fields, (x, y, height, and width) and the GetParams method from the TShape type. At runtine, a variable of the type TRectangle will look and behave like a TShape variable except that the Draw procedure will use a different code block than the procedure by the same name in an instantiated variable of the TShape type. Similarly, the TSquare object type will inherit all the "grandparent" fields from TShape and have its own different procedure blocks. If desired, TRectangle and TSquare could have declared additional fields in their type definitions which would show up as additional fields (memory locations) available at runtime which instantiated parent object variables would not have.
Next are the specific procedure implementations for these object types.
<delphi> procedure TShape.GetParams;
begin
write('TShape.GetParams : '); readln(x, y, width, height); writeln;
end;
procedure TShape.Draw;
begin
writeln('TShape.Draw');
writeln('Position: x = ', x:4:0, ' y = ', y:4:0); writeln(' Size: w = ', width:4:0, ' h = ', height:4:0); writeln;
end;
procedure TRectangle.Draw;
begin
writeln('TRectangle.Draw');
end;
procedure TSquare.GetParams; begin
write('TSquare.GetParams : '); readln(x, y, width, height); height := width; writeln('making sure all sides are equal for Square'); writeln;
end;
procedure TSquare.Draw;
begin
writeln('TSquare.Draw');
end;
</delphi>
To provide clarity, no actual (platform specific) drawing routines are used. Rather, generic Pascal I/O routines are included to illustrate the runtime behavior of objects. This particular hierarchy of objects, fields and methods would likely not be the best implementation for an actual shape drawing application and this will become apparent after new concepts are introduced.
The GetParams method is used to obtain and to perform any needed processing of the field values. Although the object fields could accessed directly as was done in the previous section, it is considered better programming style to encapsulate the "getting" a "setting" of object fields using object methods. In fact, FPC provides additional language features to help support and enforce narrowing the accessibility and visibility of internal object data.
Since the different types of objects in this program all use the same fields, using a common GetParams method would seem to make sense to include at the top level object which all other descendent objects can inherit. Thus, this method is implemented in the TShape object type.
The TShape object type includes a Draw method which is probably not needed in an actual shape drawing program. However, a variable declared of type TShape likely will be assigned to a variable of a descendent object type such as TRectangle and as such needs to have a draw method which will draw using the code of the descendent object type. Normally, this method would just be declared as a stub or not implemented at all. The syntax for this type of situation and others will be described in a later section. In this case, we want to provide some feedback for field values and program flow so it includes some writeln statements.
The TRectangle object type does not include a GetParams method but does have its own Draw method. The TSquare object type has it's own GetParams method which repeats much of the the same code from the TShape.GetParams method but also includes code to ensure that the height and width fields are the same. It then defines its own Draw method. In an actual program, the Draw method for squares and rectangles would probably be the same and the Draw method could be omitted for TSquare.
Here is a sample program which demonstrates the behavior of the various objects.
<delphi> Var
Shape : TShape; Rectangle : TRectangle; Square : TSquare;
begin writeln; writeln ('Getting parameters for Shape'); Shape.GetParams;
write ('Calling Shape.Draw : '); Shape.Draw;
writeln; writeln ('Getting parameters for Rectangle'); Rectangle.GetParams;
write ('Calling Rectangle.Draw : '); Rectangle.Draw;
writeln; writeln ('Getting parameters for Square'); Square.GetParams;
write ('Calling Square.Draw : '); Square.Draw; </delphi>
The above code produces the following output.
Getting parameters for Shape
TShape.GetParams : 1 2 3 4
Calling Shape.Draw : TShape.Draw
Position: x = 1 y = 2
Size: w = 3 h = 4
Getting parameters for Rectangle
TShape.GetParams : 11 22 33 44
Calling Rectangle.Draw : TRectangle.Draw
Getting parameters for Square
TSquare.GetParams : 111 222 333 444
making sure all sides are equal for Square
Calling Square.Draw : TSquare.Draw
Tthe Shape object is initialized by calling the GetParams method and then the Draw method is called which prints out the field values. Next, the same is done for the Rectangle object. Notice that since there is no GetParams method for Rectangles, the compiler used the parent object's method, TShape.GetPArams to carry out this action. Finally, the Square object is initialized by calling the GetPArams method which is explicitly defined for Squares and the output shows this method was called and did extra processing. The Draw method is called and the Draw method specifically defined for Squares was executed. Note that if the draw procedure was left out for the T-square type (as would be reasonable for an actual program) the last output line would look like this.
Calling Square.Draw : TRectangle.Draw
Now what happens if we assign a sub object to a parent object as follows?
<delphi> writeln; writeln ('Assigning Rectangle to Shape'); Shape := Rectangle; writeln; write ('Calling Shape.Draw : '); Shape.Draw;
writeln; writeln ('Assigning Square to Shape'); Shape := Square; writeln; write ('Calling Shape.Draw : '); Shape.Draw;
writeln; writeln ('Assigning Square to Rectangle'); Rectangle := Square; writeln; write ('Calling Rectangle.Draw : '); Rectangle.Draw; </delphi>
Assigning Rectangle to Shape
Calling Shape.Draw : TShape.Draw
Position: x = 11 y = 22
Size: w = 33 h = 44
Assigning Square to Shape
Calling Shape.Draw : TShape.Draw
Position: x = 111 y = 222
Size: w = 333 h = 333
Assigning Square to Rectangle
Calling Rectangle.Draw : TRectangle.Draw
The first block of code assigns the Rectangle variable to the Shape variable. When the Shape.Draw method is invoked, it executes the TShape.Draw code and not the TRectangle.Draw code but as can be seen, the Rectangle fields are what is printed out. This behavior is called Static method inheritance in FPC. If it is desired to instead invoke the TRectangle.Draw method in this situation, FPC provides way to do this which will be covered in the next section. Continuing on, Shape is assigned to Square and the TShape.Draw method is invoked which prints out the field values of the Square object. Finally, the Rectangle variable is assigned the Square and the TRectangle.Draw method is invoked similarly to the behavior of the TShape.Draw methods.
Assigning an ancestor object to a descendent object is not allowed. The compiler will flag the following line as an error.
<delphi> Rectangle := Shape; <--- can't assign a parent to a child, does not compile </delphi>
Objects - Dynamic Inheritance
For a drawing application, a common task would be to refresh the display and stepping through an array or linked list of shapes and calling the draw method. Instead of needing a case statement inside the loop which selects one of many possible specific drawing procedures
<delphi> for k := 1 to NumShapes do
with ShapeRec[k] do case ShapeRec[k].ShapeKind of cRectangle : DrawRectangle (x, y, width, .height); cSquare: DrawSquare : (x, y, width, .height); cTriangle: DrawTriangle:(x, y, angle1, angle2, base); end; // case
</delphi> the code would look something like this <delphi> for k:= 1 to Numshapes do
Shape[k].draw;
</delphi>
Where each Shape object could be one of any sub objects descended from TShape. Code maintenance is made easier since there is one fewer locations in code which needs to be modified. All changes to the behavior of a particular sub object of shape is kept together in one place.
However, as seen in the last section, calling the Draw method for the Shape variable this way will not invoke the appropriate draw method of the sub object which is the behavior that is desired in this (and most) cases. To obtain the desired behavior, the virtual keyword must be inserted after the method declaration in the type definition as follows:
<delphi> Type
TShape = Object
x, y : single;
height, width : single;
procedure GetParams;
procedure Draw; virtual;
end;
TRectangle = Object(TShape)
procedure Draw; virtual;
end;
</delphi>
Now if a Rectangle object is assigned to the Shape variable, the Draw method of TRectangle will be used. The term often used to describe this situation is called overriding a parent method. Although the body of main program will the same as in the previous section, the execution behavior will be different. The virtual keyword tells the compiler to hold off fixing the specific procedure used and instead lets the binding of the method be determined at runtime dynamically.
By adding the virtual keyword to the type declarations of TShape, TRectangle and TSquare in the last section and running the program again, the latter portion of the output would look like this:
Assigning Rectangle to Shape Calling Shape.Draw : TRectangle.Draw Assigning Square to Shape Calling Shape.Draw : TSquare.Draw Assigning Square to Rectangle Calling Rectangle.Draw : TSquare.Draw
Mixing virtual methods and non virtual (static) methods in the inheritance hierarchy may result in confusing behavior.
Objects - Constructors and Destructors
If you try to run the previous program after simply adding the virtual keyword to the Draw methods of each object type, you will find the program will halt with a run time error. Due to the peculiarities of this particular OOP implementation, when virtual methods are declared, special initialization code must be included for that object. Specifically, a special method must be included in the object type definition called a constructor and this constructor must be called at runtime when creating the object.
Constructors don't need to be declared virtual to act virtual.
Classes
Waiting for the construction contract
