6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 14.4
Slide 14.4.1
So we have seen a first pass at building an object-oriented
system,using Scheme as the base,There are a few details that
we still need to clean up however,These include what to do if a
class does not have a method to handle some request,the need to
be able to refer to the object within methods belong to an object
(i.e,an ability to recursively use methods of an object within an
object),and the need to identify types of objects,
Slide 14.4.2
What happens if the object doesn't know how to handle a
message? We need a way to detect if we actually have a method
available,
To do this,we first need a kind of tagged system that will tell us
when we have no method,But remember our convention,
anytime we ask an object for something,it should return a
procedure,Thus our way of saying no method exists also needs
to be a procedure,as shown with no-method,a procedure
of no arguments with access to a local frame enclosing a special
symbol,
Slide 14.4.3
Then,to check if we have a method,we take an object,such as
might be returned by askand we do the following,First,is the
argument a procedure? If it is,we assume it is a method,and we
method,We do this by applying no-method to get out
the tag,then checking to see if the argument is eq? to it,If it is,
then we return false to indicate that no method exists for this
object and message,Otherwise,we signal an error,
So why go through all of this? We need to distinguish between
an error in our implementation and an error in trying to get an
instance to handle a message it doesn't know about,This lets us
separate the details of the implementation from use of the implementation,
proceed,If it is not,we check to see if the value passed in is a no-
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.4
We have now seen an example of creating a class definition,
(one of these maker-procedures),and we have seen an
example of using that class definition to create a particular
instance (in this case a person),We have also seen how to
separate out getting methods from actually invoking those
methods to execute an action,And we have seen how to build a
generic interface to the objects,so that we uniformly ask any
object to do anything,
Slide 14.4.5
So now we can use this,We can ask g to say "The sky is blue",
with the behavior shown,
We can also ask the object its name,and we can ask the object to
change its name,as shown,But in this latter case,it would be
nice if we could get the object to "say" its new name whenever it
changed its name,So doing the mutation to change the variable
binding is easy,But how do we get a person to use it's own
method,that is,how do we get a person to recursively,within
one of its methods,ask itself to do something (e.g,SAY)?
Slide 14.4.6
Well,we have a problem here,In particular,we don't have inside
the code of this object any access to the object itself,We have
nothing that points to the procedure representing the object,
which would allow us to "ask" it to do something,
Slide 14.4.7
What we need is an explicit reference to the object itself,Our
way of doing this is to add an explicit argument (called self)
to all our methods,The goal in doing this is to allow us to have
an object be able to refer to itself,so that not only can it do
something within a method,it can ask itself to get other methods
to do execute other actions,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.8
So here is the first change needed to make this happen,In our
class definition,we ensure that each method has a self
argument as its first argument,by convention,
Slide 14.4.9
The second modification is within the CHANGE-NAME
method,We want the object to ask itself to say its new name,
Thus,we askthe self object to handle a SAYmessage,
Notice that this should then ask the same procedure,the
procedure representing the object,to handle this new message,
which should in turn return a new method for doing exactly that,
Slide 14.4.10
Of course we will also need to modify ask,This is easy since
we just need to explicitly include the object as an argument when
applying the method,since each method expects an object in that
place in its argument list,Notice an important design issue here,
By separating out askfrom other parts of the system,and
especially separating out the idea of getting a method from the
idea of applying it,we have made it easier to incorporate
changes such as this one,
Slide 14.4.11
So how does this change give us more capabilities in our system,
especially our ability to ask an object to do something within the
context of another method? Suppose we evaluate (ask g
'CHANGE-NAME ‘ishmael),This will first get the
method for changing names from g,That returned procedure is
then applied to the object that corresponds to g(i.e,the value
bound to gin this environment) and the argument ishmael,
This will first mutate the binding for fnamein this
environment,from george to ishmael,And then it will
ask this object; the value associated with gor more particularly
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
the value associated with objectin this frame,to "say" something,
Slide 14.4.12
As before,let's step back from the details to consider what is
being accomplished here,We have been designing an interface
for objects,We have seen that we can use a generic ask
procedure to get methods for instances,but we have also seen
that in some cases we want the methods to interact with one
another,This requires letting an object be able to refer to itself
within a method,This led to the extension to our design that we
just went through,While this was a small change in terms of
code,it was a big change in terms of impact on behavior of the
system,
Slide 14.4.13
One other detail that we need to consider is how to identify the
type of an object,Remember in our earlier example,we wanted
to add a method to an arrogant-prof so that he would respond in
one manner if the person asking a question was a student,and in
a different manner if the person asking a question was a
professor,How do we add the equivalent of,type tags” to our
objects in an object-oriented system?
Slide 14.4.14
One easy way to do this is to use the basic component of an
object,namely a method,Here is an example for our person
class definition,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.15
Now,if we check to see if someone is a person,we get the
response we expect,Clearly we could write methods for classes
in which an argument is checked for its type,and different
behaviors are used based on that type,
Slide 14.4.16
But we need to be careful! If we ask an object about a different
type,we get an error (due to a lack of method) rather than the
behavior we wanted,namely returning false to say that the object
is not of this type,
Slide 14.4.17
To fix this,we need to add one more detail,namely a way of
asking if an object is of a particular type,This is shown here,
This procedure can be used to check in an object is of a
particular type,but will return true or false,rather than failing
due to a lack of method,
Slide 14.4.18
The point of this last example is to show that we can add the
same abilities we saw earlier with tagged data types,We simply
need to ensure that our mechanism for checking type tags is
consistent with the object-oriented framework,But with this
ability,we can now inherit all the power we saw earlier of
dispatching on type,in this case to different objects,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.19
Thus,we have seen how to create (in Scheme) a system for
describing simple object-oriented frameworks,The system
includes a means of defining classes,a means of creating
instances of those classes,and ways of referring to instances of
classes,including oneself,
We have seen how we can use the idea of an environment to
capture local state,and to use procedures created on demand to
provide methods that can access and change the state of
instances of classes,
In the next lecture,we will turn to more complex ways of
creating and using classes,especially the issues of inheritance
and delegation,
6.001 Notes,Section 14.5
Slide 14.5.1
In the last part of this set of lectures,we looked at the basic
elements of object-oriented programming,We examined the
role of objects as a paradigm for structuring systems,and we saw
how we could use our knowledge of Scheme to construct a
framework for building classes and instances in an object-
oriented system,
So where are we? We have established ways of creating classes
and instances in our object oriented system,as well as
conventions for dealing with messages and methods,and
conventions for allowing objects to refer to themselves to
support methods calling other methods,
Now we want to look at using these ideas to explore the idea of
inheritance,Recall that inheritance meant having the ability to create subclasses of objects,or specialized classes of
objects,which could inherit behaviors from the superclass of objects,The goal was to have different specializations
of a general class so that the common methods between specializations could be captured in the superclass,and the
variations of unique methods could be isolated within subclasses,
Slide 14.5.2
Now why would we want the ability to inherit,within a
hierarchy of classes? First,by isolating a shared value in a
single variable,we make it easier to maintain consistency of
values,In other words,we avoid having multiple versions of the
same variable,and thus isolate changes in that variable to a
single location,
Second,under this new view of programming,classes become
our basic units of code construction,As a consequence,we
would like to enforce modularity as much as possible on classes,
and by enabling the construction of hierarchies of classes,in
which a subclass can inherit methods (as well as variables) from
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
superclass instances,we ensure consistency of behavior and isolate changes to a single method,
Slide 14.5.3
Let's extend our current example to look at this issue,We already
have a class,called a person,We can create a subclass of
person,called a professor,A professor,because it is a kind of
person,has the same capabilities as a person,It has an internal
variable for its names; it has methods for returning its name,for
changing its name,and for saying things,However,a professor
has a unique capability,different from normal people,When a
professor is "lecturing",it prefaces all said material with the
word "Therefore",Thus it has a new method,LECTURE,
which does that,
Slide 14.5.4
"Therefore" the behavior we expect is the following,If we define
ein the global environment to be an instance of a professor,
using a make-procedure that we will discuss shortly,then we
can ask eto SAYthings,In this case,it behaves just like a
person,as it inherits the SAYmethod from its superclass,
If we ask eits name,we get a different behavior from a normal
person,which suggests that we will need a new method that
shadows the person’s method,
Finally,we can ask e to LECTURE about the chromaticity of
the atmosphere,In that case,eas a professor should use the
LECTURE method,This will,as a consequence,say therefore the sky is blue,
So we see that this object eshould both have the ability to use methods specific to a professor and the ability to
inherit methods from an underlying person,
Slide 14.5.5
Here is the approach we will take to make this happen,In
particular,we are going to allow the inheritance of methods from
superclasses to subclasses,by within each subclass creating an
internal instance of a superclass,In other words,a professor will
have within it an internal instance of a person,Then,if a
message is passed to the professor,the professor will first see if
it has an explicit method for that message,If so,it does its thing,
If not,it "passes the buck" to the internal person instance,asking
it to handle this message,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.6
Using that idea,we can then build a make-procedure for
professors,that is,we can implement the class definition of
professors,Recall that it is going to be a subclass,which is going
to inherit from persons,Note how the constructor works,It
creates an internal person,i.e.,it literally calls the make-
procedure for persons and creates a binding for int
personto that internal instance,It then returns an object that
represents an instance of this class,As before,it is a message-
passing procedure,Here,it has one thing it explicitly knows how
to do,to LECTURE,Otherwise,it gets the method for the
message from its internal instance of a person,This means it
literally passes "the buck" back to the internal person,saying "You figure out how to handle this message and return
a method for me",
Notice the form used,This object satisfies our convention,as it will return a method (or procedure) for all messages,
But the default,rather than saying there is no method,is to ask the superclass instance to handle things,Thus,if we
ask eto SAYsomething,since eis a professor it will first look for an explicit SAY method,and then deducing it
doesn't have one,it will ask its internal person to SAY,This will,as we saw in the earlier slides,then return a
method for saying things,and apply it,
Slide 14.5.7
So let’s trace this through,We will suppress some of the details
of the environment model,so that we can see the general flow of
computation,
Defining eto be a professor using the appropriate make-
procedure should have the following behavior,
Slide 14.5.8
First,applying make-professor will drop a frame in
which the variable fname is bound to the symbol eric
and the variable lname bound to the symbol Grimson,
Thus,frame E1 is created by the application of make-
professor,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.9
Recall that the body of make-professorhas within it a
letexpression,You have seen that evaluating a letwill
cause a new frame to be created that is scoped by the current
frame,Within that frame,we bind int-personto some
value,Thus,frame E2 is created by the evaluation of the let
within make-professor,and is scoped by E1,
Slide 14.5.10
..,and what is int-personbound to? … to the result of
evaluating (make-person,..),And we know what
that does as we saw it earlier,It creates a new frame through the
application of make-personand notice that this frame is
scoped by the global environment,because that is where the
make-person procedure's environment pointer points to,
Within that frame we bind the variable fnameto the argument
passed in,and relative to that frame we evaluate the body of
make-person which returns a message-passing object
corresponding to an instance of a person,Int-person is
then bound to this object,as that is the value returned by the application of make-person,
Slide 14.5.11
Finally,we evaluate the body of the let expression inside
make-professor,with respect to E2 (remember that
was the frame created by evaluating the first part of the let),That
creates the message-passing object corresponding to an instance
of a person,It's environment pointer points to this frame,and the
procedure is the value returned by the application of make-
professor,Therefore,eis bound to this value in the
global environment,
This has a structure that is very useful in our system!
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.12
In particular,ein the global environment points to a structure,a
professor object (the thing enclosed in blue),which is a
procedure with access to local frames,Within that object,there is
an object,an internal person (the thing enclosed in red),Notice
that it is referred to by a local variable that points to one of these
structures,a message-passing object that has access to some
local frames,Thus,we have an internal instance within another
instance,and this will support the idea of inheritance,
All we have to do is specify how the top-level object will pass
along requests for methods to the internal object,
Slide 14.5.13
First,suppose we ask this object to identify itself,In this case,
the process of,ask”ing will eventually cause the system to apply
the object e to the message whoareyou? Because the
professor class has an explicit method for handling this
message,it will create a procedure relative to this frame,And
since this frame is scoped by the frames created during the
construction of the object,the procedure will have access to the
internal variables,and can thus identify the object for us,
Slide 14.5.14
But suppose we ask this object to say something,In this case,
the process of,ask”ing will again cause the system to apply the
object e to the message say,Because the professor class
does not have an explicit method for handling this message,it
will try to inherit a method from its internal person object,
Slide 14.5.15
This means that the same message will be sent to the object that
corresponds to the internal instance of the superclass,As a
consequence,we will create a method for handling this request
with respect to this frame,and then proceed,
Thus we see that an object can inherit methods from internal
instances of superclasses,As a consequence,if we decide to
change how a class handles a method,we need only do it in the
definition of that class,and the changed behavior will then be
automatically inherited by subclasses,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 14.6
Slide 14.6.1
So we have seen how we can create subclasses,procedures that
accept messages as before,have access to internal state,but
include within them a pointer to another object that belongs only
to that instance,but that has information or capabilities of the
superclass,Now let’s look at how that provides power in
controlling behaviors of objects,
First,suppose we want our professor to lecture,We said that
“lecturing” simply meant adding the word,therefore” to the
beginning of each utterance,so here is a simple way of
accomplishing this,
Slide 14.6.2
But,a little thought suggests that at the level of code and
implementation,there is a significant overlap in the code to
implement lecture in a professor and the code for say in
person,As well,it seems reasonable as a statement about our
“world” that lectureing is actually a form of saying,That is,the
overlap is not just an accident of code,but rather these two
methods concern actions that are variants of one another at the
conceptual level,
So we want to indicate that a professor lectureing is a variant on
a person saying something,both on the implementation level
and on the conceptual level,Object-oriented programming
offers a way to acknowledge both of these cases,it’s called
delegation,and the idea is that a professor lectureing is done by having it delegate (or hand off) the job to its
superclass (person) and requesting it to say the right thing,
Here is the change we make to capture this behavior
Slide 14.6.3
When we want our professor to lecturewe really just want
him/her to saythe word "therefore" followed by whatever else
he/she was going to say,So our method for LECTUREis a
procedure (or method) with an argument,self,in order to be
able to refer to the object,plus the set of things to be said,This
procedure will then ask the internal person to SAYthe word
"therefore" followed by whatever else he/she was going to say,
Thus,we would like to delegate to the internal person a request
to say the appropriate stuff,And if we ask a professor to lecture
"the sky is blue" the internal person would be asked to say
"therefore the sky is blue",
We will have to implement delegate but the idea makes intuitive sense,Delegation would allow us to
designate from one object a request to a specific other object to do something,This should result in the other object
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
providing the method needed by the first object to accomplish the desired task,
Slide 14.6.4
To implement the idea of delegationwe want to pass a
message from one object to another,Notice that we can just get
the method of the to object associated with the message,then
apply that method with the fromobject (the object that asked
for this delegation) as the self object,
Slide 14.6.5
This looks a lot like ask,right? Ask took a single object and
a message,got the method for that message from that object,and
then applied that method to the same object.
Delegate extends this to differentiate the object providing
the method from the object to which that method is being
applied,
Slide 14.6.6
What we have done is extend the power of our object-oriented
system,We can create subclasses,we can inherit methods from
superclasses,and we can delegate specific requests to instances
of subclasses,This should greatly increase the range of behaviors
we can now simulate,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.7
Let's return to our example system and further extend it,We
have a person,we have a professor,now let’s add a new kind of
person,an arrogant-professor,This is someone
who ends every statement with the word "obviously",We would
like this to be a subclass that inherits from a professor which
itself inherits from a person,but has a different kind of behavior,
We want the arrogant professor,whenever he/she says anything
to end it with this word "obviously",
Slide 14.6.8
If we define e in this case to be an instance of an arrogant
professor,and ask it to SAY "the sky is blue",he then says "the
sky is blue obviously" as expected,And if we ask him to
LECTUREon the topic of the chromaticity of the atmosphere,
he says "therefore the sky is blue obviously",
Slide 14.6.9
With this class design in mind,it seems easy to implement this
idea using superclasses,Our make-arrogant-
professorprocedure should simply create an internal
instance of a professor,which will itself have within it an
internal instance of a person,Then the message-passing
procedure that represents instances of arrogant professors will
simply delegate to the professor,upon receiving a request to
SAY,the requirement to use the internal SAY method of the
professor to say the things,with "obviously" added at the end,Of
course,the professor should then add a "therefore" to the front of
this set of things to say,
Thus this object should be able to say things and to inherit the ability to lecture from the internal instance as well,
Let's check it out,Clearly it can say things as we expected,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.10
So I ask e to LECTURE " the sky is blue" and it says
"therefore the sky is blue",
OOPS! Where is the "obviously"? This didn't work! Why?
Slide 14.6.11
The problem is not with the thing we just built! The new
arrogant professor subclass did the right thing,Arrogant
professor changed its SAY method with the expectation that
everything the arrogant professor says will be modified,That's
the behavior we want,
Slide 14.6.12
But think about what happens,When we ask an arrogant
professor to LECTURE something,it delegates to its internal
professor a request to LECTURE "...,obviously",But that will
then use the internal SAYmethod of the internal person and
we really should have asked the arrogant professor self to SAY
this,What is the SAYmethod for the internal person? It just
says the passed in argument,
In particular,the SAYmethod associated with the arrogant
professor did NOT get called when we asked it to lecture
because it delegated this job through to the internal person,Thus
it correctly SAYs but it incorrectly LECTUREs,The problem is that we were not careful in our design of the
professor class to say when we wanted to have something delegated,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.13
The way to fix this behavior is to use askbecause ask will
make it possible for a superclass to invoke a subclass' method as
we want in this case,Thus,we have two different kinds of
behavior mechanisms,delegation and asking,
Slide 14.6.14
So in this case,we can accomplish what we want with a simple
change,Inside of our make-professorprocedure we
change the behavior,We still have an internal person,but rather
than delegating when asked to LECTURE to the internal
person,we will ASK the self to SAY with "therefore" at the
front,
Thus when we make an arrogant professor,asking it to lecture
will use the SAY method of the arrogant lecturer,not the SAY
method of the internal person,
Slide 14.6.15
This is a rather subtle point,and the reason we are raising it is to
let you see the variations in behavior one can get,One of the
interesting challenges in designing object oriented systems is in
breaking up the system into the right sized modules and
associated behaviors and at the same time controlling the
interactions of behaviors between the classes,especially in the
presence of inheritance and hierarchies of classes,
As we can see,if there are conflicting or competing methods
within these hierarchies,we have to think carefully about asking
which object to execute which method,
6.001 Notes,Section 14.7
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.1
Now that we have seen inheritance,the ability for an internal
instance of a superclass to provide methods to specializations of
objects,what happens when we have multiple inheritance (i.e,
what happens when we have objects that inherit methods from
different kinds of superclasses)?
Slide 14.7.2
Let's add a new object,a new class,to our system,A singer is
distinct from a person,It has its own SAYmethod (which
always ends with "tra la la"),as well as having a SING method
(which starts with "the hills are alive"),
On top of this,we can then create a "singing arrogant professor",
God knows what it actually does although maybe you have seen
of few of these folks around MIT,The idea is that an s-a-p
should inherit methods from both an arrogant professor and from
a singer,This will lead to some interesting questions about how
one decides where to inherit a method from,when there are
multiple choices of methods,
Slide 14.7.3
First,we can build our base representation or base class,There is
no superclass here,because a singer is a basic class,The
definition for a singer is very simple,it’s a message-passing
object that handles methods for saying and singing,as shown
(noting that singing uses the objects SAY method),This is just
like our other class definitions in form,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.4
Now we can create the class of a singing arrogant professor,We
will have within the constructor for this class,something that
creates an internal singer,using make-singer,something
that creates an internal arrogant professor,using make-
arrogant-professorand we will have references to
both of those,Then the object that accepts messages for the
singing arrogant professor should simply take the message and
find a method,for example by first looking in the singer,then in
the arrogant professor if the singer does not have a method,The
behavior would then be what we expect as shown in the
examples of singing and lecturing,
Slide 14.7.5
To find a method,we will simply look through the objects in
order until one returns a method,Thus,find-method
takes a message and a list of objects,and scans through a loop
until it either runs out of objects or gets a returned method,
Slide 14.7.6
Clearly the order in which we list objects will determine the
behavior we see,Checking the singer first,then the professor
will give one kind of behavior,while checking the professor
first,and then the singer will give a different kind of behavior,
We could add more things to our ability to use multiple
inheritance,For example,suppose we want to pass a message on
to all the internal objects,and have them all do the appropriate
thing,For example,we could have a singing arrogant professor
with two internal objects,as before,but now when we ask it to
do some thing,we will pass the message on to all the internal
objects,Notice the different behavior we get for the examples in
this case,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.7
And this comes from a particular choice,as shown here in the
procedure that delegates a message to all objects,
Slide 14.7.8
So what we have shown you is how to build an object oriented
system,especially considering the kinds of behaviors we can get,
We saw the role of classes,instances and hierarchies of classes
that capture common behavior,
Once we have the ability to create variations on objects,we have
to worry about how to allocate requests for actions,We can
delegate to particular objects,We can inherit from super
classes,If we have multiple inheritance,we have lots of
variations in how objects inherit methods from different super
classes,
Thus we have begun to see the range of behaviors available in
such systems,The goal is to try to design classes that support the
desired behaviors,in a modular fashion,This includes deciding what each class should do and the interactions
between the classes,If we make a poor design decision,we can get very unexpected behavior,and our goal is to
guard against this,
6.001 Notes,Section 14.4
Slide 14.4.1
So we have seen a first pass at building an object-oriented
system,using Scheme as the base,There are a few details that
we still need to clean up however,These include what to do if a
class does not have a method to handle some request,the need to
be able to refer to the object within methods belong to an object
(i.e,an ability to recursively use methods of an object within an
object),and the need to identify types of objects,
Slide 14.4.2
What happens if the object doesn't know how to handle a
message? We need a way to detect if we actually have a method
available,
To do this,we first need a kind of tagged system that will tell us
when we have no method,But remember our convention,
anytime we ask an object for something,it should return a
procedure,Thus our way of saying no method exists also needs
to be a procedure,as shown with no-method,a procedure
of no arguments with access to a local frame enclosing a special
symbol,
Slide 14.4.3
Then,to check if we have a method,we take an object,such as
might be returned by askand we do the following,First,is the
argument a procedure? If it is,we assume it is a method,and we
method,We do this by applying no-method to get out
the tag,then checking to see if the argument is eq? to it,If it is,
then we return false to indicate that no method exists for this
object and message,Otherwise,we signal an error,
So why go through all of this? We need to distinguish between
an error in our implementation and an error in trying to get an
instance to handle a message it doesn't know about,This lets us
separate the details of the implementation from use of the implementation,
proceed,If it is not,we check to see if the value passed in is a no-
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.4
We have now seen an example of creating a class definition,
(one of these maker-procedures),and we have seen an
example of using that class definition to create a particular
instance (in this case a person),We have also seen how to
separate out getting methods from actually invoking those
methods to execute an action,And we have seen how to build a
generic interface to the objects,so that we uniformly ask any
object to do anything,
Slide 14.4.5
So now we can use this,We can ask g to say "The sky is blue",
with the behavior shown,
We can also ask the object its name,and we can ask the object to
change its name,as shown,But in this latter case,it would be
nice if we could get the object to "say" its new name whenever it
changed its name,So doing the mutation to change the variable
binding is easy,But how do we get a person to use it's own
method,that is,how do we get a person to recursively,within
one of its methods,ask itself to do something (e.g,SAY)?
Slide 14.4.6
Well,we have a problem here,In particular,we don't have inside
the code of this object any access to the object itself,We have
nothing that points to the procedure representing the object,
which would allow us to "ask" it to do something,
Slide 14.4.7
What we need is an explicit reference to the object itself,Our
way of doing this is to add an explicit argument (called self)
to all our methods,The goal in doing this is to allow us to have
an object be able to refer to itself,so that not only can it do
something within a method,it can ask itself to get other methods
to do execute other actions,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.8
So here is the first change needed to make this happen,In our
class definition,we ensure that each method has a self
argument as its first argument,by convention,
Slide 14.4.9
The second modification is within the CHANGE-NAME
method,We want the object to ask itself to say its new name,
Thus,we askthe self object to handle a SAYmessage,
Notice that this should then ask the same procedure,the
procedure representing the object,to handle this new message,
which should in turn return a new method for doing exactly that,
Slide 14.4.10
Of course we will also need to modify ask,This is easy since
we just need to explicitly include the object as an argument when
applying the method,since each method expects an object in that
place in its argument list,Notice an important design issue here,
By separating out askfrom other parts of the system,and
especially separating out the idea of getting a method from the
idea of applying it,we have made it easier to incorporate
changes such as this one,
Slide 14.4.11
So how does this change give us more capabilities in our system,
especially our ability to ask an object to do something within the
context of another method? Suppose we evaluate (ask g
'CHANGE-NAME ‘ishmael),This will first get the
method for changing names from g,That returned procedure is
then applied to the object that corresponds to g(i.e,the value
bound to gin this environment) and the argument ishmael,
This will first mutate the binding for fnamein this
environment,from george to ishmael,And then it will
ask this object; the value associated with gor more particularly
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
the value associated with objectin this frame,to "say" something,
Slide 14.4.12
As before,let's step back from the details to consider what is
being accomplished here,We have been designing an interface
for objects,We have seen that we can use a generic ask
procedure to get methods for instances,but we have also seen
that in some cases we want the methods to interact with one
another,This requires letting an object be able to refer to itself
within a method,This led to the extension to our design that we
just went through,While this was a small change in terms of
code,it was a big change in terms of impact on behavior of the
system,
Slide 14.4.13
One other detail that we need to consider is how to identify the
type of an object,Remember in our earlier example,we wanted
to add a method to an arrogant-prof so that he would respond in
one manner if the person asking a question was a student,and in
a different manner if the person asking a question was a
professor,How do we add the equivalent of,type tags” to our
objects in an object-oriented system?
Slide 14.4.14
One easy way to do this is to use the basic component of an
object,namely a method,Here is an example for our person
class definition,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.15
Now,if we check to see if someone is a person,we get the
response we expect,Clearly we could write methods for classes
in which an argument is checked for its type,and different
behaviors are used based on that type,
Slide 14.4.16
But we need to be careful! If we ask an object about a different
type,we get an error (due to a lack of method) rather than the
behavior we wanted,namely returning false to say that the object
is not of this type,
Slide 14.4.17
To fix this,we need to add one more detail,namely a way of
asking if an object is of a particular type,This is shown here,
This procedure can be used to check in an object is of a
particular type,but will return true or false,rather than failing
due to a lack of method,
Slide 14.4.18
The point of this last example is to show that we can add the
same abilities we saw earlier with tagged data types,We simply
need to ensure that our mechanism for checking type tags is
consistent with the object-oriented framework,But with this
ability,we can now inherit all the power we saw earlier of
dispatching on type,in this case to different objects,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.4.19
Thus,we have seen how to create (in Scheme) a system for
describing simple object-oriented frameworks,The system
includes a means of defining classes,a means of creating
instances of those classes,and ways of referring to instances of
classes,including oneself,
We have seen how we can use the idea of an environment to
capture local state,and to use procedures created on demand to
provide methods that can access and change the state of
instances of classes,
In the next lecture,we will turn to more complex ways of
creating and using classes,especially the issues of inheritance
and delegation,
6.001 Notes,Section 14.5
Slide 14.5.1
In the last part of this set of lectures,we looked at the basic
elements of object-oriented programming,We examined the
role of objects as a paradigm for structuring systems,and we saw
how we could use our knowledge of Scheme to construct a
framework for building classes and instances in an object-
oriented system,
So where are we? We have established ways of creating classes
and instances in our object oriented system,as well as
conventions for dealing with messages and methods,and
conventions for allowing objects to refer to themselves to
support methods calling other methods,
Now we want to look at using these ideas to explore the idea of
inheritance,Recall that inheritance meant having the ability to create subclasses of objects,or specialized classes of
objects,which could inherit behaviors from the superclass of objects,The goal was to have different specializations
of a general class so that the common methods between specializations could be captured in the superclass,and the
variations of unique methods could be isolated within subclasses,
Slide 14.5.2
Now why would we want the ability to inherit,within a
hierarchy of classes? First,by isolating a shared value in a
single variable,we make it easier to maintain consistency of
values,In other words,we avoid having multiple versions of the
same variable,and thus isolate changes in that variable to a
single location,
Second,under this new view of programming,classes become
our basic units of code construction,As a consequence,we
would like to enforce modularity as much as possible on classes,
and by enabling the construction of hierarchies of classes,in
which a subclass can inherit methods (as well as variables) from
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
superclass instances,we ensure consistency of behavior and isolate changes to a single method,
Slide 14.5.3
Let's extend our current example to look at this issue,We already
have a class,called a person,We can create a subclass of
person,called a professor,A professor,because it is a kind of
person,has the same capabilities as a person,It has an internal
variable for its names; it has methods for returning its name,for
changing its name,and for saying things,However,a professor
has a unique capability,different from normal people,When a
professor is "lecturing",it prefaces all said material with the
word "Therefore",Thus it has a new method,LECTURE,
which does that,
Slide 14.5.4
"Therefore" the behavior we expect is the following,If we define
ein the global environment to be an instance of a professor,
using a make-procedure that we will discuss shortly,then we
can ask eto SAYthings,In this case,it behaves just like a
person,as it inherits the SAYmethod from its superclass,
If we ask eits name,we get a different behavior from a normal
person,which suggests that we will need a new method that
shadows the person’s method,
Finally,we can ask e to LECTURE about the chromaticity of
the atmosphere,In that case,eas a professor should use the
LECTURE method,This will,as a consequence,say therefore the sky is blue,
So we see that this object eshould both have the ability to use methods specific to a professor and the ability to
inherit methods from an underlying person,
Slide 14.5.5
Here is the approach we will take to make this happen,In
particular,we are going to allow the inheritance of methods from
superclasses to subclasses,by within each subclass creating an
internal instance of a superclass,In other words,a professor will
have within it an internal instance of a person,Then,if a
message is passed to the professor,the professor will first see if
it has an explicit method for that message,If so,it does its thing,
If not,it "passes the buck" to the internal person instance,asking
it to handle this message,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.6
Using that idea,we can then build a make-procedure for
professors,that is,we can implement the class definition of
professors,Recall that it is going to be a subclass,which is going
to inherit from persons,Note how the constructor works,It
creates an internal person,i.e.,it literally calls the make-
procedure for persons and creates a binding for int
personto that internal instance,It then returns an object that
represents an instance of this class,As before,it is a message-
passing procedure,Here,it has one thing it explicitly knows how
to do,to LECTURE,Otherwise,it gets the method for the
message from its internal instance of a person,This means it
literally passes "the buck" back to the internal person,saying "You figure out how to handle this message and return
a method for me",
Notice the form used,This object satisfies our convention,as it will return a method (or procedure) for all messages,
But the default,rather than saying there is no method,is to ask the superclass instance to handle things,Thus,if we
ask eto SAYsomething,since eis a professor it will first look for an explicit SAY method,and then deducing it
doesn't have one,it will ask its internal person to SAY,This will,as we saw in the earlier slides,then return a
method for saying things,and apply it,
Slide 14.5.7
So let’s trace this through,We will suppress some of the details
of the environment model,so that we can see the general flow of
computation,
Defining eto be a professor using the appropriate make-
procedure should have the following behavior,
Slide 14.5.8
First,applying make-professor will drop a frame in
which the variable fname is bound to the symbol eric
and the variable lname bound to the symbol Grimson,
Thus,frame E1 is created by the application of make-
professor,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.9
Recall that the body of make-professorhas within it a
letexpression,You have seen that evaluating a letwill
cause a new frame to be created that is scoped by the current
frame,Within that frame,we bind int-personto some
value,Thus,frame E2 is created by the evaluation of the let
within make-professor,and is scoped by E1,
Slide 14.5.10
..,and what is int-personbound to? … to the result of
evaluating (make-person,..),And we know what
that does as we saw it earlier,It creates a new frame through the
application of make-personand notice that this frame is
scoped by the global environment,because that is where the
make-person procedure's environment pointer points to,
Within that frame we bind the variable fnameto the argument
passed in,and relative to that frame we evaluate the body of
make-person which returns a message-passing object
corresponding to an instance of a person,Int-person is
then bound to this object,as that is the value returned by the application of make-person,
Slide 14.5.11
Finally,we evaluate the body of the let expression inside
make-professor,with respect to E2 (remember that
was the frame created by evaluating the first part of the let),That
creates the message-passing object corresponding to an instance
of a person,It's environment pointer points to this frame,and the
procedure is the value returned by the application of make-
professor,Therefore,eis bound to this value in the
global environment,
This has a structure that is very useful in our system!
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.5.12
In particular,ein the global environment points to a structure,a
professor object (the thing enclosed in blue),which is a
procedure with access to local frames,Within that object,there is
an object,an internal person (the thing enclosed in red),Notice
that it is referred to by a local variable that points to one of these
structures,a message-passing object that has access to some
local frames,Thus,we have an internal instance within another
instance,and this will support the idea of inheritance,
All we have to do is specify how the top-level object will pass
along requests for methods to the internal object,
Slide 14.5.13
First,suppose we ask this object to identify itself,In this case,
the process of,ask”ing will eventually cause the system to apply
the object e to the message whoareyou? Because the
professor class has an explicit method for handling this
message,it will create a procedure relative to this frame,And
since this frame is scoped by the frames created during the
construction of the object,the procedure will have access to the
internal variables,and can thus identify the object for us,
Slide 14.5.14
But suppose we ask this object to say something,In this case,
the process of,ask”ing will again cause the system to apply the
object e to the message say,Because the professor class
does not have an explicit method for handling this message,it
will try to inherit a method from its internal person object,
Slide 14.5.15
This means that the same message will be sent to the object that
corresponds to the internal instance of the superclass,As a
consequence,we will create a method for handling this request
with respect to this frame,and then proceed,
Thus we see that an object can inherit methods from internal
instances of superclasses,As a consequence,if we decide to
change how a class handles a method,we need only do it in the
definition of that class,and the changed behavior will then be
automatically inherited by subclasses,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
6.001 Notes,Section 14.6
Slide 14.6.1
So we have seen how we can create subclasses,procedures that
accept messages as before,have access to internal state,but
include within them a pointer to another object that belongs only
to that instance,but that has information or capabilities of the
superclass,Now let’s look at how that provides power in
controlling behaviors of objects,
First,suppose we want our professor to lecture,We said that
“lecturing” simply meant adding the word,therefore” to the
beginning of each utterance,so here is a simple way of
accomplishing this,
Slide 14.6.2
But,a little thought suggests that at the level of code and
implementation,there is a significant overlap in the code to
implement lecture in a professor and the code for say in
person,As well,it seems reasonable as a statement about our
“world” that lectureing is actually a form of saying,That is,the
overlap is not just an accident of code,but rather these two
methods concern actions that are variants of one another at the
conceptual level,
So we want to indicate that a professor lectureing is a variant on
a person saying something,both on the implementation level
and on the conceptual level,Object-oriented programming
offers a way to acknowledge both of these cases,it’s called
delegation,and the idea is that a professor lectureing is done by having it delegate (or hand off) the job to its
superclass (person) and requesting it to say the right thing,
Here is the change we make to capture this behavior
Slide 14.6.3
When we want our professor to lecturewe really just want
him/her to saythe word "therefore" followed by whatever else
he/she was going to say,So our method for LECTUREis a
procedure (or method) with an argument,self,in order to be
able to refer to the object,plus the set of things to be said,This
procedure will then ask the internal person to SAYthe word
"therefore" followed by whatever else he/she was going to say,
Thus,we would like to delegate to the internal person a request
to say the appropriate stuff,And if we ask a professor to lecture
"the sky is blue" the internal person would be asked to say
"therefore the sky is blue",
We will have to implement delegate but the idea makes intuitive sense,Delegation would allow us to
designate from one object a request to a specific other object to do something,This should result in the other object
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
providing the method needed by the first object to accomplish the desired task,
Slide 14.6.4
To implement the idea of delegationwe want to pass a
message from one object to another,Notice that we can just get
the method of the to object associated with the message,then
apply that method with the fromobject (the object that asked
for this delegation) as the self object,
Slide 14.6.5
This looks a lot like ask,right? Ask took a single object and
a message,got the method for that message from that object,and
then applied that method to the same object.
Delegate extends this to differentiate the object providing
the method from the object to which that method is being
applied,
Slide 14.6.6
What we have done is extend the power of our object-oriented
system,We can create subclasses,we can inherit methods from
superclasses,and we can delegate specific requests to instances
of subclasses,This should greatly increase the range of behaviors
we can now simulate,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.7
Let's return to our example system and further extend it,We
have a person,we have a professor,now let’s add a new kind of
person,an arrogant-professor,This is someone
who ends every statement with the word "obviously",We would
like this to be a subclass that inherits from a professor which
itself inherits from a person,but has a different kind of behavior,
We want the arrogant professor,whenever he/she says anything
to end it with this word "obviously",
Slide 14.6.8
If we define e in this case to be an instance of an arrogant
professor,and ask it to SAY "the sky is blue",he then says "the
sky is blue obviously" as expected,And if we ask him to
LECTUREon the topic of the chromaticity of the atmosphere,
he says "therefore the sky is blue obviously",
Slide 14.6.9
With this class design in mind,it seems easy to implement this
idea using superclasses,Our make-arrogant-
professorprocedure should simply create an internal
instance of a professor,which will itself have within it an
internal instance of a person,Then the message-passing
procedure that represents instances of arrogant professors will
simply delegate to the professor,upon receiving a request to
SAY,the requirement to use the internal SAY method of the
professor to say the things,with "obviously" added at the end,Of
course,the professor should then add a "therefore" to the front of
this set of things to say,
Thus this object should be able to say things and to inherit the ability to lecture from the internal instance as well,
Let's check it out,Clearly it can say things as we expected,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.10
So I ask e to LECTURE " the sky is blue" and it says
"therefore the sky is blue",
OOPS! Where is the "obviously"? This didn't work! Why?
Slide 14.6.11
The problem is not with the thing we just built! The new
arrogant professor subclass did the right thing,Arrogant
professor changed its SAY method with the expectation that
everything the arrogant professor says will be modified,That's
the behavior we want,
Slide 14.6.12
But think about what happens,When we ask an arrogant
professor to LECTURE something,it delegates to its internal
professor a request to LECTURE "...,obviously",But that will
then use the internal SAYmethod of the internal person and
we really should have asked the arrogant professor self to SAY
this,What is the SAYmethod for the internal person? It just
says the passed in argument,
In particular,the SAYmethod associated with the arrogant
professor did NOT get called when we asked it to lecture
because it delegated this job through to the internal person,Thus
it correctly SAYs but it incorrectly LECTUREs,The problem is that we were not careful in our design of the
professor class to say when we wanted to have something delegated,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.6.13
The way to fix this behavior is to use askbecause ask will
make it possible for a superclass to invoke a subclass' method as
we want in this case,Thus,we have two different kinds of
behavior mechanisms,delegation and asking,
Slide 14.6.14
So in this case,we can accomplish what we want with a simple
change,Inside of our make-professorprocedure we
change the behavior,We still have an internal person,but rather
than delegating when asked to LECTURE to the internal
person,we will ASK the self to SAY with "therefore" at the
front,
Thus when we make an arrogant professor,asking it to lecture
will use the SAY method of the arrogant lecturer,not the SAY
method of the internal person,
Slide 14.6.15
This is a rather subtle point,and the reason we are raising it is to
let you see the variations in behavior one can get,One of the
interesting challenges in designing object oriented systems is in
breaking up the system into the right sized modules and
associated behaviors and at the same time controlling the
interactions of behaviors between the classes,especially in the
presence of inheritance and hierarchies of classes,
As we can see,if there are conflicting or competing methods
within these hierarchies,we have to think carefully about asking
which object to execute which method,
6.001 Notes,Section 14.7
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.1
Now that we have seen inheritance,the ability for an internal
instance of a superclass to provide methods to specializations of
objects,what happens when we have multiple inheritance (i.e,
what happens when we have objects that inherit methods from
different kinds of superclasses)?
Slide 14.7.2
Let's add a new object,a new class,to our system,A singer is
distinct from a person,It has its own SAYmethod (which
always ends with "tra la la"),as well as having a SING method
(which starts with "the hills are alive"),
On top of this,we can then create a "singing arrogant professor",
God knows what it actually does although maybe you have seen
of few of these folks around MIT,The idea is that an s-a-p
should inherit methods from both an arrogant professor and from
a singer,This will lead to some interesting questions about how
one decides where to inherit a method from,when there are
multiple choices of methods,
Slide 14.7.3
First,we can build our base representation or base class,There is
no superclass here,because a singer is a basic class,The
definition for a singer is very simple,it’s a message-passing
object that handles methods for saying and singing,as shown
(noting that singing uses the objects SAY method),This is just
like our other class definitions in form,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.4
Now we can create the class of a singing arrogant professor,We
will have within the constructor for this class,something that
creates an internal singer,using make-singer,something
that creates an internal arrogant professor,using make-
arrogant-professorand we will have references to
both of those,Then the object that accepts messages for the
singing arrogant professor should simply take the message and
find a method,for example by first looking in the singer,then in
the arrogant professor if the singer does not have a method,The
behavior would then be what we expect as shown in the
examples of singing and lecturing,
Slide 14.7.5
To find a method,we will simply look through the objects in
order until one returns a method,Thus,find-method
takes a message and a list of objects,and scans through a loop
until it either runs out of objects or gets a returned method,
Slide 14.7.6
Clearly the order in which we list objects will determine the
behavior we see,Checking the singer first,then the professor
will give one kind of behavior,while checking the professor
first,and then the singer will give a different kind of behavior,
We could add more things to our ability to use multiple
inheritance,For example,suppose we want to pass a message on
to all the internal objects,and have them all do the appropriate
thing,For example,we could have a singing arrogant professor
with two internal objects,as before,but now when we ask it to
do some thing,we will pass the message on to all the internal
objects,Notice the different behavior we get for the examples in
this case,
6.001 Structure and Interpretation of Computer Programs,Copyright? 2004 by Massachusetts Institute of Technology,
Slide 14.7.7
And this comes from a particular choice,as shown here in the
procedure that delegates a message to all objects,
Slide 14.7.8
So what we have shown you is how to build an object oriented
system,especially considering the kinds of behaviors we can get,
We saw the role of classes,instances and hierarchies of classes
that capture common behavior,
Once we have the ability to create variations on objects,we have
to worry about how to allocate requests for actions,We can
delegate to particular objects,We can inherit from super
classes,If we have multiple inheritance,we have lots of
variations in how objects inherit methods from different super
classes,
Thus we have begun to see the range of behaviors available in
such systems,The goal is to try to design classes that support the
desired behaviors,in a modular fashion,This includes deciding what each class should do and the interactions
between the classes,If we make a poor design decision,we can get very unexpected behavior,and our goal is to
guard against this,
