Technotools/Technotools (Chestnut CD-ROM)(1993).ISO/lang_c/cpptut22/chap09.txt

Text File | 1992-01-19 | 23KB | 531 lines

                                                        Chapter 9
                       MULTIPLE INHERITANCE AND FUTURE DIRECTIONS

C++ version 2.0 was released by AT&T during the summer of 1989, and
the major addition to the language is multiple inheritance, the
ability to inherit data and methods from more than one class into
a subclass.  Multiple inheritance and a few of the other additions
to the language will be discussed in this chapter along with some
of the expected future directions of the language.

Several companies have C++ compilers available in the marketplace,
and many others are sure to follow.  Because the example programs
in this tutorial are designed to be as generic as possible, most
should be compilable with any good quality C++ compiler provided
it follows the AT&T definition of version 2.1 or newer.  Many of
these examples will not work with earlier definitions because the
language was significantly changed with the version 2.1 update.

After completing this tutorial, you should have enough experience
with the language to study additional new constructs on your own
as they are implemented by the various compiler writers.  We will
update the entire tutorial as soon as practical following
procurement of any new compiler, but hopefully the language will
not change rapidly enough now to warrant an update oftener than
annually.  Please feel free to contact us for information on
updates to the Coronado Enterprises C++ tutorial.


MULTIPLE INHERITANCE
_________________________________________________________________

A major recent addition to the C++ language is the ability to
inherit methods and variables from two or more parent classes when
building a new class.  This is called multiple inheritance, and is
purported by many people to be a major requirement for an object
oriented programming language.  Some writers, however, have
expressed doubts as to the utility of multiple inheritance.  To
illustrate the validity of this, it was not easy to think up a good
example of the use of multiple inheritance as an illustration for
this chapter.  In fact, the resulting example is sort of a forced
example that really does nothing useful.  It does however,
illustrate the mechanics of the use of multiple inheritance with
C++, and that is our primary concern at this time.

The biggest problem with multiple inheritance involves the
inheritance of variables or methods from two or more parent classes
with the same name.  Which variable or method should be chosen as
the inherited variable or method if two or more have the same name?
This will be illustrated in the next few example programs.


                                                         Page 9-1

           Chapter 9 - Multiple Inheritance and Future Directions


SIMPLE MULTIPLE INHERITANCE
_________________________________________________________________

An examination of the file named MULTINH1.CPP    ================
will reveal the definition of two very simple      MULTINH1.CPP
classes in lines 4 through 27 named moving_van   ================
and driver.

In order to keep the program as simple as possible, all of the
member methods are defined as inline functions.  This puts the code
for the methods where it is easy to find and study.  You will also
notice that all variables in both classes are declared to be
protected so they will be readily available for use in any class
which inherits them.  The code for each class is kept very simple
so that we can concentrate on studying the interface to the methods
rather than spending time trying to understand complex methods.
As mentioned previously, chapter 12 will illustrate the use of non-
trivial methods.

In line 30, we define another class named driven_truck which
inherits all of the data and all of the methods from both of the
previously defined classes.  In the last two chapters, we studied
how to inherit a single class into another class, and to inherit
two or more classes, the same technique is used except that we use
a list of inherited classes separated by commas as illustrated in
line 30.  The observant student will notice that we use the keyword
public prior to the name of each inherited class in order to be
able to freely use the methods within the subclass.  In this case,
we didn't define any new variables, but we did introduce two new
methods into the subclass in lines 32 through 39.

We declared an object named chuck_ford which presumably refers to
someone named Chuck who is driving a Ford moving van.  The object
named chuck_ford is composed of four variables, three from the
moving_van class, and one from the driver class.  Any of these four
variables can be manipulated in any of the methods defined within
the driven_truck class in the same way as in a singly inherited
situation.  A few examples are given in lines 47 through 56 of the
main program and the diligent student should be able to add
additional output messages to this program if he understands the
principles involved.

All of the rules for private or protected variables and public or
private method inheritance as used with single inheritance extends
to multiple inheritance.


DUPLICATED METHOD NAMES
_________________________________________________________________

You will notice that both of the parent classes have a method named
initialize(), and both of these are inherited into the subclass
with no difficulty.  However, if we attempt to send a message to

                                                         Page 9-2

           Chapter 9 - Multiple Inheritance and Future Directions

one of these methods, we will have a problem, because the system
does not know which we are referring to.  This problem will be
solved and illustrated in the next example program.

Before going on to the next example program, it should be noted
that we have not declared any objects of the two parent classes in
the main program.  Since the two parent classes are simply normal
classes themselves, it should be apparent that there is nothing
magic about them and they can be used to define and manipulate
objects in the usual fashion.  You may wish to do this to review
your knowledge of simple classes and objects of those classes.

Be sure to compile and execute this program after you understand
its operation completely.



MORE DUPLICATE METHOD NAMES
_________________________________________________________________

The second example program in this chapter named ================
MULTINH2.CPP, illustrates the use of classes       MULTINH2.CPP
with duplicate method names being inherited into ================
a derived class.

If you study the code, you will find that a new method has been
added to all three of the classes named cost_per_full_day().  This
was done intentionally to illustrate how the same method name can
be used in all three classes.  The class definitions are no problem
at all, the methods are simply named and defined as shown.  The
problem comes when we wish to use one of the methods since they are
all the same name and they have the same numbers and types of
parameters and identical return types.  This prevents some sort of
an overloading rule to disambiguate the message sent to one or more
of the methods.

The method used to disambiguate the method calls are illustrated
in lines 60, 64, and 68 of the main program.  The solution is to
prepend the class name to the method name with the double colon as
used in the method implementation definition.  This is referred to
as qualifying the method name.  Qualification is not necessary in
line 68 since it is the method in the derived class and it will
take precedence over the other method names.  Actually, you could
qualify all method calls, but if the names are unique, the compiler
can do it for you and make your code easier to write and read.

Be sure to compile and execute this program and study the results.
The observant student will notice that there is a slight
discrepancy in the results given in lines 79 through 81, since the
first two values do not add up to the third value exactly.  This
is due to the limited precision of the float variable but should
cause no real problem.



                                                         Page 9-3

           Chapter 9 - Multiple Inheritance and Future Directions


DUPLICATED VARIABLE NAMES
_________________________________________________________________

If you will examine the example program named    ================
MULTINH3.CPP, you will notice that each base       MULTINH3.CPP
class has a variable with the same name.         ================

According to the rules of inheritance, an object
of the driven_truck class will have two variables with the same
name, weight.  This would be a problem if it weren't for the fact
that C++ has defined a method of accessing each one in a well
defined way.  You have probably guessed that we will use
qualification to access each variable.  Lines 38 and 45 illustrate
the use of the variables.  It may be obvious, but it should be
explicitly stated, that there is no reason that the subclass itself
cannot have a variable of the same name as those inherited from the
parent classes.  In order to access it, no qualification would be
required.

It should be apparent to you that once you understand single
inheritance, multiple inheritance is nothing more than an extension
of the same rules.  Of course, if you inherit two methods or
variables of the same name, you must use qualification to allow the
compiler to select the correct one.


PRACTICAL MULTIPLE INHERITANCE
_________________________________________________________________

Examine the example program named DATETIME.H     ================
for a practical example using multiple              DATETIME.H
inheritance.  You will notice that we are        ================
returning to our familiar date and time classes
from earlier chapters.

There is a good deal to be learned from this very short header file
since it is our first example of member initialization.  There are
two constructors for this class, the first being a very simple
constructor that does nothing in itself as is evident from an
examination of line 12.  This constuctor allows the constructors
to be executed for the classes new_date and time_of_day.  In both
cases a constructor will be executed that requires no parameters,
and such a constructor is available for each of these two classes.

The second constuctor is more interesting since it does not simply
use the default constructor, but instead passes some of the input
parameters to the inherited class constructors.  Following the
colon in line 13 are two member initializers which are used to
initialize members of this class.  Since the two parent classes are
inherited, they are also members of this class and can be
initialized as shown.  Each of the member initializers is actually
a call to a constructor of the parent classes and it should be
evident that there must be a constructor with the proper number of

                                                         Page 9-4

           Chapter 9 - Multiple Inheritance and Future Directions

input parameters to respond to the messages given.  You will note
that in line 14, we are actually calling the constructor with no
parameters given explicitly.  If we chose, we could simply let the
system call that constructor automatically, but this gives us an
explicit comment on what is happening.


MORE ABOUT MEMBER INITIALIZERS
_________________________________________________________________

Actually, we can use the member initializer to initialize class
members also.  If we had a class member of type int named
member_var, we could initialize it also by mentioning the name of
the member followed by the value we desired to initialize it to in
parentheses.  If we wished to initialize it to the value 13, we
could use the following line of code in the member initializer
list;

   member_var(13);

Following all member initialization, the normal constructor code
is executed which in this case is given in line 16.


ORDER OF MEMBER INITIALIZATION
_________________________________________________________________

The order of member initialization may seem a bit strange, but it
does follow a few simple rules.  The order of member initialization
does not follow the order given by the initialization list, but
another very strict order over which you have complete control.
All inherited classes are initialized first in the order they are
listed in the class header.  If lines 14 and 15 were reversed,
class new_date would still be initialized first because it is
mentioned first in line 8.  It has been mentioned that C++ respects
its elders and initializes its parents prior to itself.  That
should be a useful memory aid in the use of member initializers.

Next, all local class members are initialized in the order in which
they are declared in the class, not the order in which they are
declared in the initialization list.  Actually, it would probably
be good practice to not use the member initializer to initialize
class members but instead to initialize them in the normal
constructor code.

Finally, after the member initializers are all executed in the
proper order, the main body of the constructor is executed in the
normal manner.

USING THE NEW CLASS
_________________________________________________________________

The example program named USEDTTM.CPP uses the datetime class we
just built, and like our previous examples, the main program is

                                                         Page 9-5

           Chapter 9 - Multiple Inheritance and Future Directions

kept very simple and straight forward.  You will  ===============
note that the default constructor is used for       USEDTTM.CPP
the object named now, and the constructor with    ===============
the member initializers is used with the objects
named birthday and special.

The diligent student should have no trouble understanding the
remaining code in this example.


FUTURE DIRECTIONS OF C++
_________________________________________________________________

An ANSI committee has been formed to write an ANSI standard for
C++.  They first met in the Spring of 1990 and are expected to
complete the standard in about three years.  Until the new standard
is released, the C++ language is expected to stay fairly stable.
However, due to the nature of compiler writers and their desire to
slightly improve their offerings over their competitors, you can
bet that the language will not remain static during this three year
period.

Many small changes have been added during the past year that barely
affect the casual programmer, or even the heavy user of the
language.  You can be sure that the language will evolve slowly and
surely into a very usable and reliable language.  There are two
areas, however, that should be discussed in a little detail because
they will add so much to the language in future years.  Those two
topics are parameterized types and exception handling.



FUTURE DIRECTIONS - PARAMETERIZED TYPES
_________________________________________________________________

Many times, when developing a program, you wish to perform some
operation on more than one data type.  For example you may wish to
sort a list of integers, another list of floating point numbers,
and a list of alphabetic strings.  It seems silly to have to write
a separate sort function for each of the three types when all three
are sorted in the same logical way.  With parameterized types, you
will be able to write a single sort routine that is capable of
sorting all three of the lists.

This is already available in the Ada language as the generic
package or procedure.  Because it is available in Ada, there is a
software components industry that provides programmers with
prewritten and thoroughly debugged software routines that work with
many different types.  When this is generally available in C++,
there will be a components industry for C++ and precoded, debugged
and efficient source code will be available off the shelf to
perform many of the standard operations.  These operations will
include such things as sorts, queues, stacks, lists, etc.


                                                         Page 9-6

           Chapter 9 - Multiple Inheritance and Future Directions


Bjarne Stroustrup has announced that parameterized types, otherwise
known as templates or generics, will be available in a future
version of C++.  He has presented a paper with details of one way
to implement them, but this is only a suggestion, not a
specification.

Borland International has included templates in version 3.0 of
Borland C++, and hopefully their implementation will be very close
to the final definition of templates.
The next three example programs will illustrate the use of
templates with Borland's compiler, but may not work with other
compilers.


THE FIRST TEMPLATE
_________________________________________________________________

The example program named TEMPLAT1.CPP is the    ================
first example of the use of a template.  This      TEMPLAT1.CPP
program is so simple it seems silly to even      ================
bother with it but it will illustrate the use of
the parameterized type.

The template is given in lines 4 through 8 with the first line
indicating that it is a template with a single type to be replaced,
the type ANY_TYPE.  This type can be replaced by any type which can
be used in the comparison operation in line 7.  If you have defined
a class, and you have overloaded the operator ">", then this
template can be used with objects of your class.  Thus, you do not
have to write a maximum function for each type or class in your
program.

This function is included automatically for each type it is called
with in the program, and the code itself should be very easy to
understand.

The diligent student should realize that nearly the same effect can
be achieved through use of a macro, except that when a macro is
used, the strict type checking is not done.  Because of this and
because of the availability of the inline method capability in C++,
the use of macros is essentially non-existent by experienced C++
programmers.



A CLASS TEMPLATE
_________________________________________________________________

The example program named TEMPLAT2.CPP is a      ================
little more involved since it provides a           TEMPLAT2.CPP
template for an entire class rather than a       ================
single function.  The template code is given in
lines 6 through 16 and a little study will show

                                                         Page 9-7

           Chapter 9 - Multiple Inheritance and Future Directions

that this is an entire class definition.  The diligent student will
recognize that this is a very weak stack class since there is
nothing to prevent popping data from an empty stack, and there is
no indication of a full stack.  Our intent, however, is to
illustrate the use of the parameterized type and to do so using the
simplest class possible.

In the main program we create an object named int_stack in line 25
which will be a stack designed to store integers, and another
object named float_stack in line 26 which is designed to store
float type values.  In both cases, we enclose the type we desire
this object to work with in "<>" brackets, and the system creates
the object by first replacing all instances of ANY_TYPE with the
desired type, then creating the object of that type.  You will note
that any type can be used that has an assignment capability since
lines 13 and 14 use the assignment operator on the parameterized
type.

Even though the strings are all of differing lengths, we can even
use the stack to store a stack of strings if we only store a
pointer to the strings and not the entire string.  This is
illustrated in the object named string_stack declared in line 27
and used later in the program.

This program should be fairly easy for you to follow if you spend
a bit of time studying it.  You should compile and run it if you
have a compiler that will handle this new construct.



REUSING THE STACK CLASS
_________________________________________________________________

The program named TEMPLAT3.CPP uses the same     ================
class with the template as defined in the last     TEMPLAT3.CPP
program but in this case, it uses the date class ================
developed earlier as the stack members.  More
specifically, it uses a pointer to the date
class as the stack member.

Because class assignment is legal, you could also store the actual
class in the stack rather than just the pointer to it.  To do so
however, would be very inefficient since the entire class would be
copied into the stack each time it is pushed and the entire class
would be copied out again when it was popped.  Use of the pointer
is a little more general, so it was illustrated here for your
benefit.

All three of the previous programs can be compiled and executed if
you have a copy of Borland C++ version 3.0.  Other compilers may
not work with these programs since parameterized types are not yet
a part of the C++ specification.



                                                         Page 9-8

           Chapter 9 - Multiple Inheritance and Future Directions


FUTURE DIRECTIONS - EXCEPTION HANDLING
_________________________________________________________________

A future version of C++ will have some form of exception handling
to allow the programmer to trap errors and prevent the system from
completely shutting down when a fatal error occurs.  The Ada
language allows the programmer to trap any error that occurs, even
system errors, execute some recovery code, and continue on with the
program execution in a very well defined way.  Bjarne Stroustrup,
working in conjunction with the ANSI-C++ committee, has announced
that some form of exception handling will be implemented but he has
not stated what form it would take as of this writing.


WHAT SHOULD BE YOUR NEXT STEP?
_________________________________________________________________

Once again, we have reached a major milestone in C++ programming.
With the ability to use inheritance, you have nearly all of the
tools you need to effectively use the object oriented programming
techniques of C++ and you would do well to stop studying again and
begin programming.  The only topic left with C++ is virtual methods
which are used for dynamic binding or polymorphism.  This will be
covered in the next two chapters.  The vast majority of all
programming can be done without dynamic binding, and in attempting
to force it into every program, you could wind up with an
unreadable mess, so you should approach it slowly.



























                                                         Page 9-9