Revised Dec 1998
                                             Slightly edited 29 Dec 2015

This file was written originally for AI students learning Pop-11 (which
includes various object oriented extensions), but it should be useful
and intelligible to others. Since this was written, the world of
programming has changed in many ways, along with changes in computer
hardware and networking. Pop-11 is the core language of the Poplog
software development environment. For more information on both:
    https://en.wikipedia.org/wiki/Poplog
    https://en.wikipedia.org/wiki/POP-11

This is a short introduction to concepts of object oriented programming
(OOP), along with a potted history, and provides some pointers to online
information about OOP in Poplog and in particular the OBJECTCLASS
package which extends Pop-11 with sophisticated object oriented
programming facilities -- partly inspired by the Common Lisp Object
System (CLOS), and partly different because of the way OBJECTCLASS
makes use of the special features of Pop-11.

This file also mentions a variety of experimental OOP libraries previously
implemented within Pop-11.

Pop-11 (often written as Pop11, or pop11) is a powerful programming
language which is the core part of the POPLOG multi-language software
development environment. (See Wikipedia links below.)

Note added: 22 Jun 2015
I have produced this revised version using html with internal links to
improve navigation. The original navigation was done by the Poplog/Pop11 editor
Ved, long before html existed. For background see these Wikipedia entries

CONTENTS

-- History ------------------------------------------------------------

I'll start with an incomplete history. Don't treat this as
authoritative: it is based only on what I've learnt since around 1970. A
more complete history would require substantial research.

The first language that combined several OOP ideas was SIMULA-67,
described in a book called Simula Begin, by Nygaard et al. Simula was
very influential though many of those who were indirectly influenced by
its ideas were unaware of the fact.

SMALLTALK was developed by AI researchers at Xerox Parc in Palo Alto
during the 1970s and was used among other things for implementing one of
the early mouse and windows interfaces on a machine developed at XEROX.
Some of those ideas were later taken over by Apple and used in the Apple
Macintosh.

An extension to Lisp called "FLAVORS" was produced at MIT in the late
70s and early 80s. This was later replaced by CLOS after Common Lisp had
been adopted as a standard by (some!) of the Lisp user community.

In the mid 80s a library called FLAVOURS, partly modelled on Lisp
FLAVORS was added to Pop-11. In the early 90s a sub-language called
OBJECTCLASS more closely related to CLOS was added to Pop-11. (See
further information below.)

A different strand of development arose in the early 80s out of attempts
to add OO features to conventional non-interactive languages, such as C
and Pascal, e.g. Objective-C, C++, Object Pascal, and many other
languages were developed in parallel. C++ became the most widely used of
these, probably because it was so close to C and compilers were
developed for a wide range of computers and operating systems.

Eiffel was a language developed around the same time from what some
people regard as a cleaner theoretical base. ADA was for a time used
widely by professional software engineers especially in military
contexts. It was designed to support concurrency, with some object
oriented features, along with a mathematically precise specification,
designed to support safety critical applications and military
applications where there was no room for error.

There were also various "object oriented" extensions to Prolog, and to
database systems.

During the 1980s there were also many software applications developed
using ideas associated with "object orientation" in a variety of
interpretations. For instance the X window system which provides the
basis of all the graphics and mouse-based interaction on Unix systems
was explicitly designed with an object oriented methodology, especially
the use of inheritance (explained below), even though the language used
to implement most X facilities is C.

During the 1980s much confusion (discussed below) reigned regarding what
"object oriented" meant, and that confusion survives today. In fact
there was a collection of more or less independent ideas, and different
people focused on different subsets of the ideas, according to their
interests and their prior knowledge. (Most computer scientists and
software engineers have only a limited knowledge of the variety of
languages and language features in existence.)

In the early 90s various attempts were made to start again with powerful
new languages designed entirely within an Object Oriented framework
(though not everyone agreed on what that meant). A language called
"Dylan" developed by Apple and Harlequin was largely inspired by
Scheme, a lisp-like functional language.

Java, developed at the same time by Sun, spread much faster than Dylan
and eventually completely eclipsed it, although many people think Dylan
was inherently a much better (cleaner, more powerful, language). Java
was designed specifically to support remote access to computers in
connection with World Wide Web browsers, and that was one of the
features which led to its rapid spread.

One problem is that the different developers do not use "object" or
"object oriented" in the same way, so that there is an incredible amount
of confusion and hype (hyperbole)!

There is a useful potted history of OO ideas in Chapter 2 of Booch.
It also contains a good bibliography, though such things easily get out
of date, and you may need to use bibliographic search mechanisms to get
important recent reports surveys.


-- What is object oriented programming? --------------------------

-- -- Confusions about OOP

For a while there were many people who thought that object-oriented
systems are systems associated with windows and pointing devices. This
was simply a confusion arising from the fact that object oriented
systems provide excellent mechanisms for implementing window/mouse
mechanism.

Another persistent confusion is that object oriented programming has
something to do with thinking about objects or using objects in the
machine.

But that notion is too vague to be of any interest: many programming
styles and programming languages use data-structures which are objects,
such as lists, records, vectors and arrays, which may include static
information and have associated procedures that can be run to access,
update, or make use of the information.

Moreover, all sorts of programs include things that represent objects
outside the machine, for instance, databases of information about
employees in a firm, simulation models of a factory or mechanism. This
is not a feature unique to object oriented programming.

-- -- OOP and "message sending"

A related confusion is that object oriented programming has something to
do a distinction between

(a) applying a procedure (function) to a object

and

(b) sending a message to an object, and allowing the object to
    decide what to do.

This may be a useful idea in a distributed system where objects can reside on
remote machines, requiring something like messages to be sent across a network,
but in other contexts, message sending can simply be seen as "syntactic sugar"
for a special type of procedure call (sometimes referred to as "method
invocation"). Moreover it is a RESTRICTIVE syntax because it enforces an
asymmetry (between object sent, and recipient) which is sometimes unnatural.

E.g. The following syntax might be used for sending a message to a
graphical object O1 saying, in effect, "link yourself to objects O2
and O3":

    O1 <- link_to(O2, O3)

A variant of this notation might use a dot rather an an arrow to
represent sending a message:

    O1.link_to(O2, O3)

Since this is really invoking a procedure to do something involving O1,
O2 and O3, this can be seen as syntactic sugar for something which could
just as well be expressed as:

    link_to(O1, O2, O3)

or perhaps, sending a message to all three saying "link yourselves to
one another":

    (O1, O2, O3) <- link_to

The first form, which treats O1 as special, might be used in cases where
the link_to method is specifically defined in the context of objects of
the type to which O1 belongs. The asymmetric syntax shows that O1 is
treated differently from O2 and O3, and that might be useful in some
cases. But does it justify a special syntax? We often use procedures
(functions) in connection with two or more objects which are treated
differently, e.g.

    member(item, list)

Why should we be required to express this as:

    item <- member(list)
    "Item, tell me, are you in list?"
or
    list <- member(item)
    "List, tell me, is item in you?"


-- -- Message sending considered harmful

Some people who have used most of the features of object-oriented
languages have found the kind of enforced asymmetry required in message
sending unhelpful and restrictive.

For example, a vision program given two line segments and required to
find where their intersection is, is inherently symmetric, so that being
forced to write

    line1 <- intersect(line2)

rather than

    intersect(line1, line2)

is just a source of confusion and unnecessary difficulty. Likewise, if
you have a method for finding where a line intersects a circle, why
should you have to write

    line1 <- intersect(circ1)

rather than

    circ1 <- intersect(line1)

Why not simply define a method which takes a line and a circle and can
be invoked as

    intersect(line1, circ1)

returning a list of intersection points?

It is often supposed that something like the message sending syntax is
needed because it allows the behaviour of the "link_to" method to depend
on the TYPE of thing O1 is. I.e. O1 is in control of the action
produced.

However there are object oriented languages (E.g. CLOS, the Common Lisp
Object System, and OBJECTCLASS in Pop-11) which allow method invocation
to use the second form, AND allow the type of behaviour to depend on the
types of ALL the objects, not just ONE of them. Thus you could have an
intersect method which behaves appropriately when given two lines, a
line and a circle, a circle and a line, two circles, etc. A different
definition of the method is given for each combination of types.

These are therefore not ordinary procedures, which can only have ONE
definition, and so they are sometimes referred to as "generic
functions", or simply "methods".

This ability to use the same procedure name in connection with different
procedures for handling different combinations of types of arguments is
sometimes called "overloading", and sometimes "polymorphism". It is
sometimes available in languages which have nothing to do with object
oriented programming, such as Algol68.

These languages simply generalise the fact, which is common to most
programming languages, and very widely accepted, namely that infix
arithmetic functions like:

    +  *  /

and also printing procedures and equality tests, all have to do
different things depending on the types of their arguments.

E.g. in a language which allows "+" to be used with combinations of
integers and reals, there have to be four different types of behaviours,
to cope with all these cases:

    3 + 4,  3.0 + 4,  3 + 4.0,  3.0 + 4.0

In languages like Pop-11 and Common Lisp, which include ratios, big
integers and complex numbers even more combinations are required!

E.g. in Pop-11 all the following are legal and will produce a result:

    3 + 3  =>
    3.5 + 3  =>
    3 + 3.5  =>
    3 + 7/2  =>
    3/4 + 77.5  =>
    3/4 + sqrt(-1) =>


Likewise equality testing and printing often use a common syntax,
involving the same form of procedure invocation, but require different
behaviours depending on the types of the objects involved. Printing an
integer, a real, a string, and a list involve very different procedures.

I.e. these arithmetic, printing, and equality functions are already
generic functions (overloaded functions).

In most languages there are some built in overloaded functions, but
USERS of the language are not allowed to define new ones. OOP languages
generally do allow that (though in a restricted format where message
sending is used). However this is a feature of data-abstraction, rather
than something specific to OOP.

For these reasons I regard it as a serious mistake to attempt to
DEFINE OOP in terms of the use of a "message sending" syntax, which
enforces an asymmetry between the different sorts of objects to which a
method is applied. Rather message sending is just one form of syntax
that can be used in connection with overloading and data-encapsulation
(defined below).

Many of the well known OOP languages (and the Pop-11 Flavours system) do
use some form of message sending syntax, but it is also the case that
some very powerful OOP languages, including the Common Lisp Object
System (CLOS) and Pop-11 Objectclass do not do so.

So "message sending" is merely an OPTIONAL FEATURE of an object oriented
language, found in some OOP languages, but not in others.


-- -- Data encapsulation and abstract data-types

Another common idea is that OOP is about data-encapsulation: having
data-structures (objects) that make use of some internal representation
that is not "publicised" and which cannot be used directly from outside
the code that defines those objects. Instead the data in an object
always has to be accessed by means of procedures that provide an
interface to the objects, or by sending a message to the object.

This is a very important programming idea: it allows different subsets
of a program, or packages, to be developed and used in a way that allows
the developers of each package to make various kinds of "hidden" changes
and improvements without those who use the packages having to change
their code.

For instance, a class of lists could be defined in which there is an
empty list nil (or [] in Pop-11) two main list constructors :: and <>
where :: creates a new list by adding an item to an old list:

    x :: list

and <> creates a new list by concatenating two old lists:

    list1 <> list2

and two list-accessing procedures hd (head) and tl (tail) which can be
applied to a non-empty list, with hd(list) returning the first item in
the list and tl(list) returning a (possibly empty) list containing
everything except the first item.

The idea in data-encapsulation is that the programmer who develops lists
and procedures for constructing and manipulating lists, such as ::, <>,
hd, tl, need not specify how lists are represented in the machine nor
how the procedures work in detail. For instance they may invoke "hidden"
store management procedures for creating new lists when needed.

Making such a distinction between "public" information about a type of
data with associated procedures and the "hidden" or "private" details
is also sometimes described as "data abstraction".

A class of data can be defined entirely in terms of how the instances of
the class behave "from outside," without any mention of how the data are
implemented.

For example, lists and the main procedures which operate on them could
be defined by equations specifying how the list-accessing and list
constructing functions behave, e.g.

1. The head of a list created by adding x to the front of a list is x:

                hd(x :: list) = x

2. The tail of a list created by adding x to the front of a list is that
list:

                tl(x :: list) = list

3. Applying HD or TL to the empty list produces an error:

                hd(nil)   = error
                tl(nil)   = error

4. Concatenating list with the empty list, or the empty list with list
returns the original list:

                list <> nil   = list
                 nil <> list  = list

5. If the result of adding x to the front of list1 is concatenated with
list2, then that produces the same list as adding x to the result of
concatenating list1 with list2:

        (x :: list1) <> list2   =  x :: (list1 <> list2)

and so on.

Those "axioms" define a collection of operations on lists without
specifying how the lists are implemented in the machine. (They are often
implemented as "chained pairs" as depicted in TEACH BOXES, but
alternative implementations are possible.)

By using the abstract specification of lists in terms of HOW THEY
BEHAVE, we give the programmer all the information that is needed for
writing programs that create and manipulate lists.

(Sometimes there are issues to do with maximising efficiency, which may
depend on having additional information about internal representations.)

This use of abstract data types is very important for ensuring that
programs are correct (relative to their specification) and that
different people working on the same large project know exactly what
they can expect from one another.

However, it does not require any special language support: a really good
programmer can program like this in any language. It is a matter of
programming discipline, rather than a type of language, though some
languages may provide better support for it than others.

For example, the defclass mechanism in Pop-11 (see REF * DEFCLASS),
which can be used to define new record classes or vector classes (as
described in HELP * RECORDCLASS, HELP * VECTORCLASS) provides some
support for data abstraction, as does DEFSTRUCT in Common Lisp.

Related techniques exist in other languages, such as ML (described
briefly in HELP * PML). But Pop-11, LISP and ML are not normally
regarded as object oriented languages simply because they provide data
abstraction/encapsulation in this sense.


-- What is OOP really? (A silly question) -----------------------------

It should be clear now that this is a silly question, because it has no
answer!

The phrase "object oriented programming" has come to connote a
collection of ideas, and arguing about which of these ideas is somehow
correct, or THE defining condition of object orientedness, is just
silly. (Such silliness is often displayed in discussions in internet
news groups such as comp.object, comp.lang.clos, comp.lang.dylan.)

We could ask: which are the most important ideas typically associated
with object oriented programming languages. Different people would give
different answers.

My personal answer is: support for inheritance (described below) and
overloading.

That's because after many years of programming in a variety of styles I
have found that the combination of these two features (especially with
multiple inheritance) can provide very powerful support for production
of complex software packages which can be extended in different ways in
different contexts without the risks involved in having to extend them
by re-writing existing code.

The extensions supported include using packages written for different
purposes in a new application combining those purposes: that is where
multiple inheritance (described below) is particularly useful.

E.g. in the Pop-11 RCLIB package there is a collection of processes for
creating and manipulating pictures on the screen and using the mouse to
manipulate them. In the Pop-11 SIM_AGENT toolkit there are classes and
methods concerned with managing processes involving a collection of
interacting simulated agents and objects. These two sets of classes and
methods can be combined to produce a simulation with a graphical
interface by introducing new classes and methods which inherit structure
and behaviours from the two pre-existing packages.

The other features normally associated with OOP, e.g. data abstraction,
and data-encapsulation, are also important, and perhaps more
fundamental, but they are much older, very widely used outside the
context of OOP, and do not to the same extent require special features
to be built into the language. However inheritance and overloading
require the programming language to include special mechanisms (at
compile time and at run time.)


-- Questions worth asking about OOP -----------------------------------

Instead of having "religious" debates about which are the correct ideas,
or what OOP "really" is, what we need to do is understand that there is
an important collection of ideas which have been developed in the
context of a variety of programming languages since computers first
became available. Different subsets of those ideas can be combined for
different purposes.  We need to know:

o what the full collection of features in various object oriented
  languages is

o how they work

o when they are useful and when they are not useful in the production
  of software

o which combinations of the features fit well in different programming
  contexts

o what sorts of tools are needed to make them available

o which syntactic forms are available for these features

o what the pros and cons of different syntactic forms are

o what sorts of development environments support the use of the various
  features.


-- Typical features of OOP Languages ----------------------------------

OOP Systems generally provide some subset of the following.

o A means of defining classes of objects, where a class is an abstract
specification which can have many instances. (In some languages a class
is itself an instance of a higher order metaclass.)

o A means of creating instances of classes, where instances of the same
class will generally have a structure inherited from the class
definition, and may contain (initially) dome default information also
inherited from the class definition. This is instance-inheritance.
Instances of classes are often referred to as "objects". (In some
OOP languages classes are themselves just a special type of object.
In others there is a sharp distinction.)

o Provisions for defining a new subclass of an existing class (its
superclass). The subclass will "inherit" various features from its
superclass, including a structure (a set of slots or fields capable of
having values) some default values (for slots), and some methods
defined for the superclass.

Example:
If class C2 (e.g. accountant) is a subclass of class C1 (e.g. employee)
then instances of C2 will inherit both from the superclass C1 (e.g. all
employees have a salary and an employer) and from the subclass C2 (e.g.
accountants have a minimum initial salary).

The inheritance by C2 of some of C1's features can be called
class-inheritance. If a1 is an instance of C2, i.e. an actual accountant
object, then its inheritance of features from the classes C2 and
(indirectly C1) can be called instance-inheritance.

o Some OOP languages allow a class to have more than one superclass,
i.e. they support "multiple inheritance".


-- -- Single and Multiple inheritance

o 'Single-inheritance' systems allow each class to have only one
super-class, so that from any object or sub-class, there is a single
chain of 'upward' links through classes.

o 'Multiple-inheritance' systems allow a class to be defined as a
sub-class of more than one super-class, so that the upward links branch
out in tree-like fashion, requiring a searching operation to find
information in super-classes. With single inheritance finding
information requires simply looking up a non-branching chain of
superclasses.

NOTE: OBJECTCLASS provides multiple inheritance, so that you can define
the class 'teacher' as a sub-class of both 'person' and 'worker'.
For examples see TEACH * OBJECTCLASS_EXAMPLE


-- -- Mixins

o Mixins are a kind of class that can have no instances in themselves
but can be combined with other classes.

Thus instead of having employee as a stand-alone class we could have it
as a "mixin" which can be combined with other classes such as plumber,
carpenter, builder, to produce new subclasses called employed_plumber,
employed_carpenter etc.

Mixins work only in systems that allow multiple inheritance, since the
combined class has to inherit from both the mixin and its other
superclass or superclasses. Mixins are often specially useful because of
the methods associated with them.


-- -- Methods

o OOP languages typically provide some means of defining "methods"
associated with a class (and automatically inherited by all its
sub-classes). Thus the class person might have a method called birthday,
which when applied to an instance of a person increases that person's
age by 1. If employee is a subclass of person then the birthday method
will also be applicable to all instances of employee (i.e. all
employees). Likewise if teacher is a subclass of employee then all
teachers will have the birthday method. However it may be possible to
define a specialised version of the birthday method for a particular
subclass. E.g. for the class sailor the method might post a greeting
for the sailor to pick up at the next port of call.

Multiple inheritance systems support more complex ways of inheriting
methods. If a method name (e.g. M) is used to define methods for two
classes C1 and C2 (i.e. different methods with the same name), and then
a new class C3 is defined which inherits from both C1 and C2, a decision
has to be taken regarding which version of M will be inherited by
instances of C3. Typically that will depend on whether C1 was listed
before C2 or C2 before C1 in specifying the definition of class C3.
Some languages with multiple inheritance may provide syntax for
specifying the superclass from which the method should be selected,
e.g.
    M(c: as C1)     apply to c, the version of M defined for C1

o Method invocation may use a message sending or a function application
syntax, as explained above.

E.g. running the birthday method may look like one or other of
these (or some variant):

    john <-- birthday

    birthday(john)

In either case what happens depends on the class(es) that john belongs
to, and how they have defined the birthday method. I.e. invoking
apparently the same method can produce different sorts of behaviour for
different sorts of objects. This is sometimes also called "overloading".

In systems that use the second form of syntax (e.g. OBJECTCLASS, CLOS)
methods are often referred to as generic functions, or multi-methods,
because the method selection is based on the classes of all the
arguments, not just the first one.


-- -- Static vs dynamic methods

o In some languages every constant and every variable, and every complex
expression built out of procedure calls, variables, and constants has a
fixed "type" that can be determined at compile time.

For instance in many languages if a variable X has been declared as of
type "integer" then no legal program can ever assign to it a decimal
number or ratio. Thus if an overloaded method is applied to X, the
compiler can work out at compile time which version of the method is to
be used, and the program will run fast.

In other languages, such as Pop-11 and Lisp and Prolog, there is more
flexibility, since a variable can have different types of values at
different times during the running of the program. That means that
decisions about which methods are appropriate have to be made at run
time, not compile time.

Methods that allow run-time selection on the basis of the actual types
of objects can be called dynamic methods. They work only when the
implementation ensures that each and every object carries information
about its type with it. So this gives greater flexibility at the cost of
extra space (usually) and reduced speed because of the run time checks.

This disadvantage is sometimes worth putting up with because of the
extra flexibility, especially in AI programming environments. Where
either maximum speed or a compile-time guarantee that the program will
never have a "method failure" is required, static methods may be
preferred.



-- -- Single vs multi methods

o Some languages that support the generic function style of OOP allow
the behaviour of the method to depend not only on the the type of the
first argument but on the combination of types of all the arguments.

For example a language for designing graphical systems may allow you to
specify that a link should be drawn between two objects, but allow the
method definitions to be different depending whether it is a link
between a square and a square, a square and a circle, a circle and a
circle, and so on. (This was illustrated above, in the discussion of
restrictions of message sending.)

For a language without multi-methods you have to specify within the
methods for each class how instances of that class should be linked to
instances of other classes.

E.g. you specify for the class circle a method that checks whether it is
being asked to link itself to a square, a triangle or another circle and
then takes appropriate action. For some applications this is quite
natural, whereas for many applications it is very arbitrary to put
that "how to link" information with just ONE of the classes, instead of
defining a method that is explicitly defined to handle TWO classes, in a
symmetric way.

(This is one of those topics that can easily generate stupid theological
debates about which is "right". It all depends on the problem, or
perhaps on the style the programmer is used to, or perhaps on how the
program has to interface with other programs.)



-- Inheritable features -----------------------------------------------

An OOP system may provide, associated with each class, data features and
procedural features, as follows.


-- -- Data Features associated with a class:

D1. A set of 'slots' or 'fields' with particular names, to be found in
    each instance of the class.

D2. Default values or 'fillers' for these slots.

D3. A set of data associated globally with the class.

-- -- Procedural Features associated with a class:

P1. Actions to be performed when a new subclass is created.

P2. Actions to be performed when a new instance of the class is created.

P3. Actions to be performed when an instance (object) is sent a message.

P4. Actions to be performed when a slot or field in an instance is
    accessed or updated.

P5. Actions to be performed when information associated globally with
    the class is accessed or updated.

P6. Actions to be performed when an instance is destroyed.

All of these features, except perhaps D3 and P5 should be inherited
from superclasses by their sub-classes.


-- Additional features ------------------------------------------------

There are additional features of some object oriented systems which
are not found in all, e.g.

1. The ability to have classes which are also objects. This would enable
a class to be both an instance of a more abstract type of class (a
METACLASS) and also a subclass of a more general class at the same level
of abstraction. This can add considerably to the flexibility of the
system. (Pop-11 Flavours has meta-flavours, but Objectclass does not
support this.)

2. The ability to alter a class after some instances have been created,
e.g. by adding a new information slot to the class, and have the
instances automatically feel the effects. This can be useful during
development and debugging but complicates the implementation and reduces
efficiency.

This is another way in which Pop-11 Flavours is more flexible than
Objectclass. However both of them allow you to redefine methods after
instances have been created: the new versions of the methods will work
on pre-existing agents.


3. Some systems, e.g. Smalltalk-80, provide powerful mouse and window
browsing tools. Objectclass has a limited browsing capability
described in REF * OBJECTCLASS/Browsing


-- OOP and AI ---------------------------------------------------------

The relevance of OOP to AI is debatable. During the 1970s and 1980s a
considerable amount of work was done on developing the notion of a
Frame-based knowledge representation formalism. Examples included FRL,
KL1, KLONE, KRYPTON, and then later systems like KEE, ART,
Knowledgecraft, CLASSIC and more besides. You can find discussions of
some of these in:

Brachman, R.J. and Levesque, H.J. (eds). 1985 Readings in knowledge
representation} Los Altos, CA: Morgan Kaufmann 1985.

Frame systems tend to be similar to Object Oriented Programming
languages with additional features, e.g. extra methods of inference
besides "isa" inheritance from superclasses, extra constraints and
checking on values that are put in slots, more abstract ways of
specifying classes, and so on.

When these languages were first introduced they generated heated debates
about whether frames actually did anything that could not be done with
logic alone. See especially the paper on the logic of Frames by Pat
Hayes in the Brachman and Levesque collection.

At least we can say that for some purposes, an OOP system may be useful
for knowledge which has to be 'compiled' into an efficient, special
purpose, representation.


-- Previous OOP extensions to Pop-11 ----------------------------------

At various times a number of Pop-11 users have experimented with object
oriented extensions to Pop-11.


-- . LIB OBJ         (circa 1983)

LIB * OBJ (available at Sussex and Birmingham) was a package for object
oriented programming, implemented by Aaron Sloman using POP-11 processes
(HELP * PROCESS) to represent objects, around May 1983.


-- . LIB NEWOBJ      (circa 1985)


LIB * NEWOBJ (distributed with Poplog) extends this to provide multiple
inheritance and some other extensions and bug-fixes, including a
separation between 'start_actions' and 'default_actions', previously
confused. For more information see HELP * NEWOBJ


-- . LIB FRAMES      (1985, or earlier)

LIB * FRAMES (available at Sussex and Birmingham) provides a
transcription to Pop-11 by Rudi Lutz, around 1985, of the Frame system
described in the LISP book by P.H. Winston and B. Horn, For details see
TEACH * FRAMES. This is an implementation (or partial implementation?)
of the language known as KRL (Knowledge Representation Language), partly
inspired by Marvin Minsky's paper on Frames as a representation for
knowledge, published around 1975 and represented in several collections of
AI papers in the 1970s.


-- . LIB FLAVOURS    (circa 1985)

LIB * FLAVOURS (distributed with Poplog) was an ambitious system
implemented by Mark Rubinstein, modelled loosely on the ZetaLisp FLAVORS
system, developed at MIT within the MIT Lisp environment and used on
LISP Machines. The key ideas are introduced in TEACH * FLAVOURS.

The FLAVOURS system was quite widely used by a number of projects based
on Pop-11, as it is very flexible, though it suffered from a number of
defects

    (a) It introduces a type of syntax which is very different from
        that of Pop-11, unlike Objectclass, which fits very well with
        the Pop-11 syntax

    (b) It is fairly slow: that is a consequence of the flexibility,
        which allows class definitions to be changed after instances
        have been defined.

    (c) Where a method involving two objects is inherently symmetric
        (E.g. a method to link two objects on the screen) you are forced
        to treat it as asymmetric because it is invoked by sending a
        message to one of the objects. The message can mention the
        other. This is a defect common to all OOP systems based on message
        sending, as opposed to treating method invocation as a
        generalisation of procedure application.

OBJECTCLASS overcomes these limitations, but is not quite so flexible.
E.g. it does not allow a class definition to be changed after instances
have been created, though methods can be redefined after instances have
been created.

For more information on FLAVOURS see the following documentation:

HELP * FLAVOURS lists available documentation
REF  * FLAVOURS is a reference file for the flavours system.
HELP * FLAVOUR_LIBRARY lists the flavours and messages provided in
    the autoloadable library.
HELP * METAFLAVOURS describes the instances that represent flavours.


-- . LIB OBJECTCLASS (circa 1991)

LIB * OBJECTCLASS (distributed with Poplog) was designed by Steve Knight
(now Leach) at Hewlett Packard Research Laboratories, Bristol, then
integrated with Poplog by Rob Duncan at Sussex University. It maps OOP
ideas onto existing Pop-11 concepts. The correspondence of ideas is as
follows (extracted from HELP * OBJECTCLASS):

    methods:
        procedures, which act according to the type of their arguments

    instances:
        records, the instance variables of which are simply the fields
        of the record and are accessed by appropriate methods

    classes:
        keys, although the keys that  play the role of classes are given
        some extra fields (via hidden properties) so that they can
        provide inheritance of behaviour
        See REF * KEYS, for more on this. Also HELP * CLASSES

Because this is a natural mapping, the overhead of a method call is
relatively low.  This means that the objectclass package provides an
efficient means of doing object-oriented programming.

TEACH * OBJECTCLASS_EXAMPLE
    provides a gentle introduction to OOP, based on OBJECTCLASS syntax.

TEACH * ADVENT_OBJECTCLASS
    shows how to implement the beginnings of an adventure game.

HELP * OBJECTCLASS
    gives a more complete overview.

REF * OBJECTCLASS
    gives full gory details


-- Further reading (Packages using Objectclass) -----------------------

LIB * SIM_AGENT

    Describes a package which combines the object oriented features of
    Objectclass with some of the "rule-based system" toolkit
    LIB POPRULEBASE. The package is intended to define a framework for
    exploring architectures for intelligent interacting agents.
    The inheritance mechanism of Objectclass provides many convenient
    features, allowing aspects of the package to be tailored differently
    for different classes of objects. See this introduction to
    poprulebase with some simple examples: TEACH RULEBASE

    More detail: TEACH POPRULEBASE

    Even more (technical) detail: HELP POPRULEBASE

HELP * RCLIB
    (Available at Birmingham) Describes an alternative extension to
    RC_GRAPHIC, based on Objectclass providing a range of useful tools
    including control panels, movable objects, etc.

The above two packages are available in the Birmingham Poplog directory
    http://www.cs.bham.ac.uk/research/projects/poplog/

The above two packages are also included as part of Poplog, which is available
here
  For linux:
    http://www.cs.bham.ac.uk/research/projects/poplog/latest-poplog/
  For windows (less complete and less well supported):
    http://www.cs.bham.ac.uk/research/projects/poplog/winpop/
  System overview and history
    http://www.cs.bham.ac.uk/research/projects/poplog/freepoplog.html

HELP * GO
    Describes the Graphical Objects library, which combines objectclass
    and the rc_graphic system (HELP * RC_GRAPHIC) along with additional
    X window facilities (e.g. event handlers), in a sophisticated
    graphical package. (GO is distributed with Poplog.)


-- Object Oriented Design Methodologies -------------------------------

In addition to OOP languages people have tried to develop OO methods,
OO design methods. For example, see

Grady Booch: Object Oriented Design, The Benjamin Cummings Publishing
Company Inc. 1991.

Another such method, developed by researchers at Hewlett Packard, called
the "fusion method", is described in:

Derek Coleman, et. al.,  "O-O Development - The Fusion Method."
          Prentice-Hall Object-Oriented Series. ISBN 0-13-338823-9

For a collaborative project on user interface design environments I
wrote some notes on "Structure Sharing Ontology Oriented Design"
available online at Birmingham in the file ~axs/pap/ssood.90.11.03
(Available on request from outside Birmingham.)

NOTE: At Birmingham there is a file TEACH * OOPS.ORIG, which is the
original teach file circulated by Steve Knight (Hewlett Packard
Research Labs), who designed and implemented the OBJECTCLASS library.

For more on poplog see
    http://www.cs.bham.ac.uk/research/projects/poplog/freepoplog.html

DOWNLOADS
    http://www.cs.bham.ac.uk/research/projects/poplog/latest-poplog
    (report problems to Aaron Sloman: a.sloman@cs.bham.ac.uk)

--- $usepop/pop/teach/oop
--- [NOTE: the copyright notice is now the same as the XFREE86 copyright notice.]
    http://www.cs.bham.ac.uk/research/projects/poplog/copyright.html