Key Concept:
Object-Oriented Programming
One of the most effective ways to really understand OOP is to see how the main OOP concepts are ultimately implemented at the low-level. This article uses this approach by describing how to implement OOP in the C programming language. Experience shows that software developers, who understand the concepts at the low-level apply them more efficiently and with greater confidence than developers, who only encountered the concepts at the higher-level.
What is Object-Oriented Programming?
Object-oriented programming (OOP) is a way of design based on the three fundamental concepts:
- Encapsulation – the ability to package data and functions together into classes
- Inheritance – the ability to define new classes based on existing classes in order to obtain reuse and code organization
- Polymorphism – the ability to substitute objects of matching interfaces for one another at run-time
Object-Oriented Programming in C
Although the fundamental OOP concepts have been traditionally associated with object-oriented languages, such as Smalltalk, C++, or Java, you can implement them in almost any programming language including portable, standard-compliant C (ISO-C90 Standard).
NOTES:
If you simply develop end-user programs in C, but you also want to do OOP, you probably should be using C++ instead of C. Compared to C++, OOP in C can be cumbersome and error-prone, and rarely offers any performance advantage.
However, if you build software libraries or frameworks the OOP concepts can be very useful as the primary mechanisms of organizing the code. In that case, most difficulties of doing OOP in C can be confined to the library and can be effectively hidden from the application developers. This document has this primary use case in mind.
This article describes how OOP is implemented in the QP/C real-time framework. As a user of these frameworks, you need to understand the techniques, because you will need to apply them also to your own application-level code. But these techniques are not limited only to developing QP/C applications and are applicable generally to any C program.
Encapsulation (Classes)
Encapsulation is the ability to package data with functions into classes. This concept should actually come as very familiar to any C programmer because it’s quite often used even in the traditional C. For example, in the Standard C runtime library, the family of functions that includes fopen(), fclose(), fread(), fwrite(), etc. operates on objects of type FILE. The FILE structure is thus encapsulated because client programmers have no need to access the internal attributes of the FILE struct and instead the whole interface to files consists only of the aforementioned functions. You can think of the FILE structure and the associated C-functions that operate on it as the FILE class. The following bullet items summarize how the C runtime library implements the FILE class:
- Attributes of the class are defined with a C struct (the
FILEstruct). - Operations of the class are defined as C functions. Each function takes a pointer to the attribute structure (
FILE *) as an argument. Class operations typically follow a common naming convention (e.g., allFILEclass methods start with prefixf). - Special functions initialize and clean up the attribute structure (
fopen()andfclose()). These functions play the roles of class constructor and destructor, respectively.
You can very easily apply these design principles to come up with your own “classes”. For example, suppose you have an application that employs two-dimensional geometric shapes (perhaps to be rendered on an embedded graphic LCD).
The basic Shape “class” in C can be declared as follows (NOTE: The code corresponding to this section is located in the sub-directory: oop_in_c/encapsulation/):
#ifndef SHAPE_H
#define SHAPE_H
/* Shape's attributes... */
typedef struct {
int16_t x; /* x-coordinate of Shape's position */
int16_t y; /* y-coordinate of Shape's position */
} Shape;
/* Shape's operations (Shape's interface)... */
void Shape_ctor(Shape * const me, int16_t x, int16_t y);
void Shape_moveBy(Shape * const me, int16_t dx, int16_t dy);
int16_t Shape_getX(Shape * const me);
int16_t Shape_getY(Shape * const me);
#endif /* SHAPE_H */ The Shape “class” declaration goes typically into a header file (e.g., shape.h), although sometimes you might choose to put the declaration into a file scope (.c file).
One nice aspect of classes is that they can be drawn in diagrams, which show the class name, attributes, operations, and relationships among classes. The following figure shows the UML class diagram of the Shape class:
And here is the definition of the Shape‘s operations (must be in a .c file):.
#include "shape.h" /* Shape class interface */ /* constructor implementation */ void Shape_ctor(Shape * const me, int16_t x, int16_t y) { me->x = x; me->y = y; } /* move-by operation implementation */ void Shape_moveBy(Shape * const me, int16_t dx, int16_t dy) { me->x += dx; me->y += dy; } /* "getter" operations implementation */ int16_t Shape_getX(Shape * const me) { return me->x; } int16_t Shape_getY(Shape * const me) { return me->y; }
You can create any number of Shape objects as instances of the Shape attributes struct. You need to initialize each instance with the “constructor” Shape_ctor(). You manipulate the Shapes only through the provided operations, which take the pointer "me" as the first argument.
NOTES:
The "me" pointer in C corresponds directly to the implicit "this" pointer in C++. The "this" identifier is not used, however, because it is a keyword in C++ and such a program wouldn't compile with a C++ compiler.
#include "shape.h" /* Shape class interface */ #include/* for printf() */ int main() { Shape s1, s2; /* multiple instances of Shape */ Shape_ctor(&s1, 0, 1); Shape_ctor(&s2, -1, 2); printf("Shape s1(x=%d,y=%d)\n", Shape_getX(&s1), Shape_getY(&s1)); printf("Shape s2(x=%d,y=%d)\n", Shape_getX(&s2), Shape_getY(&s2)); Shape_moveBy(&s1, 2, -4); Shape_moveBy(&s2, 1, -2); printf("Shape s1(x=%d,y=%d)\n", Shape_getX(&s1), Shape_getY(&s1)); printf("Shape s2(x=%d,y=%d)\n", Shape_getX(&s2), Shape_getY(&s2)); return 0; }
Inheritance
Inheritance is the ability to define new classes based on existing classes in order to reuse and organize code. You can easily implement single inheritance in C by literally embedding the inherited class attribute structure as the first member of the derived class attribute structure.
For example, instead of creating a Rectangle class from scratch, you can inherit most what’s common from the already existing Shape class and add only what’s different for rectangles. Here’s how you declare the Rectangle “class” (NOTE: The code corresponding to this section is located in the sub-directory: oop_in_c/inheritance/)
#ifndef RECT_H
#define RECT_H
#include "shape.h" /* the base class interface */
/* Rectangle's attributes... */
typedef struct {
Shape super; /* <== inherits Shape */
/* attributes added by this subclass... */
uint16_t width;
uint16_t height;
} Rectangle;
/* constructor prototype */
void Rectangle_ctor(Rectangle * const me, int16_t x, int16_t y,
uint16_t width, uint16_t height);
#endif /* RECT_H */ With this arrangement, you can always safely pass a pointer to Rectangle to any C function that expects a pointer to Shape. Specifically, all functions from the Shape class (called the superclass or the base class) are automatically available to the code>Rectangle/code>class (called the subclass or the derived class). So, not only all attributes, but also all functions from the superclass are inherited by all subclasses.
NOTES:
The alignment of the Rectangle structure and the inherited attributes from the Shape structure is guaranteed by the C Standard WG14/N1124. Section 6.7.2.1.13 of this Standard, says: “… A pointer to a structure object, suitably converted, points to its initial member. There may be unnamed padding within a structure object, but not at its beginning”.
#include "rect.h"
/* constructor implementation */
void Rectangle_ctor(Rectangle * const me, int16_t x, int16_t y,
uint16_t width, uint16_t height)
{
/* first call superclass’ ctor */
Shape_ctor(&me->super, x, y);
/* next, you initialize the attributes added by this subclass... */
me->width = width;
me->height = height;
} main.c)
#include "rect.h" /* Rectangle class interface */ #include/* for printf() */ int main() { Rectangle r1, r2; /* multiple instances of Rect */ /* instantiate rectangles... */ Rectangle_ctor(&r1, 0, 2, 10, 15); Rectangle_ctor(&r2, -1, 3, 5, 8); printf("Rect r1(x=%d,y=%d,width=%d,height=%d)\n", r1.super.x, r1.super.y, r1.width, r1.height); printf("Rect r2(x=%d,y=%d,width=%d,height=%d)\n", r2.super.x, r2.super.y, r2.width, r2.height); /* re-use inherited function from the superclass Shape... */ Shape_moveBy((Shape *)&r1, -2, 3); Shape_moveBy(&r2.super, 2, -1); printf("Rect r1(x=%d,y=%d,width=%d,height=%d)\n", r1.super.x, r1.super.y, r1.width, r1.height); printf("Rect r2(x=%d,y=%d,width=%d,height=%d)\n", r2.super.x, r2.super.y, r2.width, r2.height); return 0; }
As you can see, to call the inherited functions you need to either explicitly up-cast the first “me” parameter to the superclass (Shape *), or alternatively, you can avoid casting and take the address of the member “super” (&r2->super).
NOTES:
There are no additional costs to using the "inherited" functions for instances of the subclasses. In other words, the cost of calling a function for an object of a subclass is exactly as expensive as calling the same function for an object of the superclass. This overhead is also very similar (identical really) as in C++.
Polymorphism
Polymorphism is the ability to substitute objects of matching interfaces for one another at run-time. C++ implements polymorphism with virtual functions. In C, you can also implement virtual functions in a number of ways [1,4,10]. The implementation presented here (and used in the QP/C real-time framework) has very similar performance and memory overhead as virtual functions in C++ [4,7,8].
As an example of how virtual functions could be useful, consider again the Shape class introduced before. This class could provide many more useful operations, such as area() (to let the shape compute its own area) or draw() (to let the shape draw itself on the screen), etc. But the trouble is that the Shape class cannot provide the actual implementation of such operations, because Shape is too abstract and doesn’t “know” how to calculate, say its own area. The computation will be very different for a Rectangle subclass (width * height) than for the Circle subclass (pi * radius2). However, this does not mean that Shape cannot provide at least the interface for the operations, like Shape_area() or Shape_draw(), as follows (NOTE: The code corresponding to this section is located in the sub-directory: oop_in_c/polymorphism/):
#ifndef SHAPE_H
#define SHAPE_H
#include
/* Shape's attributes... */
struct ShapeVtbl; /* forward declaration */
typedef struct {
(1) struct ShapeVtbl const *vptr; /* <== Shape's Virtual Pointer */
int16_t x; /* x-coordinate of Shape's position */
int16_t y; /* y-coordinate of Shape's position */
} Shape;
/* Shape's virtual table */
(2) struct ShapeVtbl {
(3) uint32_t (*area)(Shape const * const me);
(4) void (*draw)(Shape const * const me);
};
/* Shape's operations (Shape's interface)... */
void Shape_ctor(Shape * const me, int16_t x, int16_t y);
void Shape_moveBy(Shape * const me, int16_t dx, int16_t dy);
static inline uint32_t Shape_area(Shape const * const me) {
(5) return (*me->vptr->area)(me);
}
static inline void Shape_draw(Shape const * const me) {
(6) (*me->vptr->draw)(me);
}
/* generic operations on collections of Shapes */
(7) Shape const *largestShape(Shape const *shapes[], uint32_t nShapes);
(8) void drawAllShapes(Shape const *shapes[], uint32_t nShapes);
#endif /* SHAPE_H */ In fact, such an interface could be very useful, because it would allow you to write generic code to manipulate shapes uniformly. For example, given such an interface, you will be able to write a generic function to draw all shapes on the screen or to find the largest shape (with the largest area). This might sound a bit theoretical at this point, but it will become more clear when you see the actual code later in this Section.
Virtual Table (vtbl) and Virtual Pointer (vptr)
By now it should be clear that a single virtual function, such as Shape_area(), can have many different implementations in the subclasses of Shape. For example, the Rectangle subclass of Shape will have a different way of calculating its area than the Circle subclass.
This means that a virtual function call cannot be resolved at link-time, as it is done for ordinary function calls in C, because the actual version of the function to call depends on the type of the object (Rectangle, Circle, etc.) So, instead the binding between the invocation of a virtual function and the actual implementation must happen at run-time, which is called late binding (as opposed to the link-time binding, which is also called early binding).
Practically all C++ compilers implement late binding by means of one Virtual Table (vtbl) per class and a Virtual Pointer (vptr) per each object [4,7]. This method can be applied to C as well.
Virtual Table is a table of function pointers corresponding to the virtual functions introduced by the class. In C, a Virtual Table can be emulated by a structure of pointers-to-functions, as shown in Listing 7(2-4).
Virtual Pointer (vptr) is a pointer to the Virtual Table of the class. This pointer must be present in every instance (object) of the class, and so it must go into the attribute structure of the class. For example, the attribute structure of the Shape class augmented with the vptr member added at the top, as sown in Listing 7(1).
The vptr is declared as pointer to an immutable object (see the const keyword in front of the *), because the Virtual Table should not be changed and is, in fact, allocated in ROM.
The Virtual Pointer (vptr) is inherited by all subclasses, so the vptr of the Shape class will be automatically available in all its subclasses, such as Rectangle, Circle, etc.
Setting the vptr in the Constructor
The Virtual Pointer (vptr) must be initialized to point to the corresponding Virtual Table (vtbl) in every instance (object) of a class. The ideal place to perform such initialization is the class’ constructor. In fact, this is exactly where the C++ compilers generate an implicit initialization of the vptr.
In C, you need to initialize the vptr explicitly. Here is an example of setting up the vtbl and the initialization of the vptr in the Shape’s constructor:
#include "shape.h"
#include
/* Shape's prototypes of its virtual functions */
(1) static uint32_t Shape_area_(Shape const * const me);
(2) static void Shape_draw_(Shape const * const me);
/* constructor */
void Shape_ctor(Shape * const me, int16_t x, int16_t y) {
(3) static struct ShapeVtbl const vtbl = { /* vtbl of the Shape class */
&Shape_area_,
&Shape_draw_
};
(4) me->vptr = &vtbl; /* "hook" the vptr to the vtbl */
me->x = x; me->y = y;
}
/* move-by operation */
void Shape_moveBy(Shape * const me, int16_t dx, int16_t dy) {
me->x += dx;
me->y += dy;
}
/* Shape class implementations of its virtual functions... */
static uint32_t Shape_area_(Shape const * const me) {
(5) assert(0); /* purely-virtual function should never be called */
return 0U; /* to avoid compiler warnings */
}
static void Shape_draw_(Shape const * const me) {
(6) assert(0); /* purely-virtual function should never be called */
}
/* the following code finds the largest-area shape in the collection */
Shape const *largestShape(Shape const *shapes[], uint32_t nShapes) {
Shape const *s = (Shape *)0;
uint32_t max = 0U;
uint32_t i;
for (i = 0U; i < nShapes; ++i) {
(7) uint32_t area = Shape_area(shapes[i]); /* virtual call */
if (area > max) {
max = area;
s = shapes[i];
}
}
return s; /* the largest shape in the array shapes[] */
}
/* The following code will draw all Shapes on the screen */
void drawAllShapes(Shape const *shapes[], uint32_t nShapes) {
uint32_t i;
for (i = 0U; i < nShapes; ++i) {
(8) Shape_draw(shapes[i]); /* virtual call */
}
}
The vtbl is initialized with pointer to functions that implement the corresponding operations (see Listing 8(4)). In this case, the implementations are Shape_area_() and Shape_draw_() (see Listing 8(1-2)).
If a class cannot provide a reasonable implementation of some of its virtual functions (because this is an abstract class, as Shape is), the implementations should assert internally. This way, you would know at least at run-time, that an unimplemented (purely virtual) function has been called (see Listing 8(5,6)).
Inheriting the vtbl and Overriding the vptr in the Subclasses
As mentioned before, if a superclass contains the vptr, it is inherited automatically by all the derived subclasses at all levels of inheritance, so the technique of inheriting attributes (via the “super“ member) works automatically for polymorphic classes.
However, the vptr typically needs to be re-assigned to the vtbl of the specific subclass. Again, this re-assignment must happen in the subclass’ constructor. For example, here is the constructor of the Rectangle subclass of Shape:
#include "rect.h" /* Rectangle class interface */
#include /* for printf() */
/* Rectangle's prototypes of its virtual functions */
/* NOTE: the "me" pointer has the type of the superclass to fit the vtable */
(1) static uint32_t Rectangle_area_(Shape const * const me);
(2) static void Rectangle_draw_(Shape const * const me);
/* constructor */
void Rectangle_ctor(Rectangle * const me, int16_t x, int16_t y,
uint16_t width, uint16_t height)
{
(3) static struct ShapeVtbl const vtbl = { /* vtbl of the Rectangle class */
&Rectangle_area_,
&Rectangle_draw_
};
(4) Shape_ctor(&me->super, x, y); /* call the superclass' ctor */
(5) me->super.vptr = &vtbl; /* override the vptr */
me->width = width;
me->height = height;
}
/* Rectangle's class implementations of its virtual functions... */
static uint32_t Rectangle_area_(Shape const * const me) {
(6) Rectangle const * const me_ = (Rectangle const *)me; /* explicit downcast */
return (uint32_t)me_->width * (uint32_t)me_->height;
}
static void Rectangle_draw_(Shape const * const me) {
Rectangle const * const me_ = (Rectangle const *)me; /* explicit downcast */
printf("Rectangle_draw_(x=%d,y=%d,width=%d,height=%d)\n",
me_->super.x, me_->super.y, me_->width, me_->height);
}
Please note that the superclass’ constructor (Shape_ctor()) is called first to initialize the me->super member inherited from Shape (Listing 9(4)). This constructor sets the vptr to point to the Shape’s vtbl. However, the vptr is overridden in the next statement, where it is assigned to the Rectangle’s vtbl (Listing 9(5))
Please also note that the subclass’ implementation of the virtual functions must precisely match the signatures defined in the superclass in order to fit into the vtbl. For example, the implementation Rectangle_area_() takes the pointer “me” of class Shape*, instead of its own class Rectangle*. The actual implementation from the subclass must then perform an explicit downcast of the “me” pointer, as illustrated in (Listing 9(1-2)).
NOTES:
To simplify the discussion, Listing 9 shows the case where Rectangle does not introduce any new virtual functions of its own. In this case, Rectangle can just re-use the ShapeVtbl “as is”. However, it is also fairly straightforward to extend the implementation to the generic case where Rectangle would introduce its own RectangleVtbl that would inherit ShapeVtbl.
Virtual Call (Late Binding)
With the infrastructure of Virtual Tables and Virtual Pointers in place, the virtual call (late binding) can be realized as follows (see also Listing 7(5,6)):
uint32_t Shape_area(Shape const * const me) {
return (*me->vptr->area)(me);
}
This function definition can be either placed in the .c file scope, but the downside is that you incur additional function call overhead for every virtual call. To avoid this overhead, if your compiler supports in-lining of functions (C99 standard), you can put the definition in the header file like this:
static inline uint32_t Shape_area(Shape const * const me) {
return (*me->vptr->area)(me);
}
Alternatively, for older compilers (C89) you can use function-like macro, like this:
#define Shape_area(me_) ((*(me_)->vptr->area)((me_)))
AEither way, the virtual call works by first de-referencing the vtbl of the object to find the corresponding vtbl, and only then calling the appropriate implementation from this vtbl via a pointer-to-function. The figure below illustrates this process:
Examples of Using Virtual Functions
As mentioned in the beginning of this section on polymorphism, virtual functions allow you to write generic code that is very clean and independent on the specific implementation details for subclasses. Moreover, the code automatically supports an open-ended number of sub-classes, which can be added long after the generic code has been developed (and compiled!).
For example, Listing 8(7) shows the generic implementation of a function that finds the largest-area shape in a given collection, while Listing 8(8) shows the generic implementation of a function that draws all shapes in a given collection.
The following listing shows how to exercise all these features.
#include "rect.h" /* Rectangle class interface */ #include "circle.h" /* Circle class interface */ #include/* for printf() */ int main() { Rectangle r1, r2; /* multiple instances of Rectangle */ Circle c1, c2; /* multiple instances of Circle */ Shape const *shapes[] = { /* collection of shapes */ &c1.super, &r2.super, &c2.super, &r1.super }; Shape const *s; /* instantiate rectangles... */ Rectangle_ctor(&r1, 0, 2, 10, 15); Rectangle_ctor(&r2, -1, 3, 5, 8); /* instantiate circles... */ Circle_ctor(&c1, 1, -2, 12); Circle_ctor(&c2, 1, -3, 6); s = largestShape(shapes, sizeof(shapes)/sizeof(shapes[0])); printf("largetsShape s(x=%d,y=%d)\n", Shape_getX(&s), Shape_getY(&s)); drawAllShapes(shapes, sizeof(shapes)/sizeof(shapes[0])); return 0; }
Summary
OOP is a design method rather than the use of a particular language or a tool. This article described how to implement the concepts of encapsulation, (single) inheritance, and polymorphism in portable ISO-C. The first two of these concepts (classes and inheritance) turned out to be quite simple to implement without adding any extra costs or overheads.
Polymorphism turned out to be quite involved, and if you intend to use it extensively, you would be probably better off by switching to C++. However, if you build or use libraries (such as the QP/C real-time embedded framework), the complexities of the OOP in C can be confined to the library and can be effectively hidden from the application developers.
References
[1] Miro Samek, “Portable Inheritance and Polymorphism in C”, Embedded Systems Programming December, 1997
[2] Miro Samek, “Practical Statecharts in C/C++”, CMP Books 2002, ISBN 978-1578201105
[3] Miro Samek, “Practical UML Statecharts in C/C++, 2nd Edition”, Newnes 2008, ISBN 978-0750687065
[4] Dan Saks, “Virtual Functions in C”, “Programming Pointers” column August, 2012, Embedded.com.
[5] Dan Saks, “Impure Thoughts”, “Programming Pointers” column September, 2012, Embedded.com.
[6] Dan Saks, “Implementing a derived class vtbl in C”, “Programming Pointers” column February, 2013, Embedded.com.
[7] Stanley Lippman, “Inside the C++ Object Model”, Addison Wesley 1996, ISBN 0-201-83454-5
[8] Bruce Eckel, “Thinking in C++”, http://mindview.net/Books/TICPP/ThinkingInCPP2e.html
[9] StackOverflow: Object-Orientation in C, August 2011
[10] Axel-Tobias Schreiner, “Object-Oriented Programming in ANSI-C”, Hanser 1994, ISBN 3-446-17426-5
