Teaching SELF

[dbc at cs.brown.edu]

Mario Wolczko writes:
 >    I am planning to use Self in the practical part of an introductory
 >    course on OOP. 
 >    Who has experience in teaching Self and can provide me with related
 >    tips and/or material for such a course? What is the difference to
 >    teaching class-based languages?

 > Your message comes at an ideal time -- the Self group are currently
 > putting together some ideas and projects to encourage the use of Self
 > in the initial teaching of OOP.

 > We would be delighted to include you, and anyone else considering
 > using Self for teaching, in our plans and discussions.  To get the
 > ball moving, perhaps you could get in touch with me by email, so that
 > we can find out more about our respective plans?  I'm replying to the
 > whole list; anyone with similar plans is encouraged to mail me.

Andy van Dam and I talked to Randy earlier this year about this some.
I've included the body of this at the end.

 > If there is sufficient interest, we may consider running a workshop or
 > BOF on the teaching of Self at OOPSLA (or ECOOP, too, I suppose) --
 > anyone who would consider attending such an event, please mail me.

I'd be interested.


Brook

ps -- here's the mail -- how we've been thinking about Self for
graphics and for introductory programming


  Randy> 1]  Picked up some rumour (Brad Myeres?) that you have been thinking
  Randy> about the utility of Self. Are you in fact thinking Selfish thoughts,
  Randy> and if so, what thoughts are you thinking?

For our own work in interactive 3D graphics, there's two ways we find
Self interesting.  The first and most fundamental way is that we feel
prototype/delegation object systems match the needs of interactive
graphics much better than class-instance systems.  As systems jocks,
and not programming language designers, we implemented what turned out
to be very similar to a delegation-based object system -- we just
didn't know it was called that.  That system was described in a 1991
SIGGRAPH paper by Zeleznik et al.  We later investigated why we might
want to use a delegation system, and why we feel it is better for
graphics. The short answer to why we feel this way is that objects
provided by graphics systems end up implementing some sort of dynamic,
per-object sharing (using the terms of the "Treaty of Orlando").  A
more detailed version of this argument can be found in Conner and van
Dam, "Sharing Between Graphical Objects Using Delegation", from the
1992 Eurographics Workshop on Object-Oriented Graphics (which we can
provide in PostScript form if you are interested).

The second way Self is interesting is as a possible replacement to our
own home-brew programming language.  This possibility takes the burden
of supporting a language off of our shoulders.  Unfortunately, we run
on a variety of platforms, including Suns, HPs, DECs, IBMs and SGIs.
Self is as yet a Sun-only system, and, from what we gathered from
people in the Self group (Urs Holzle in particular), porting the
compiler would be a major undertaking.

  Randy> 2]  We've been thinking about using Self as a way to teach the
  Randy> fundamental concepts of object oriented programming  (things like
  Randy> message, object, method, inheritance, poymorphism, overriding, etc.).
  Randy> Since Self is minimalist in semantics, it should be a good medium for
  Randy> learning this stuff. And a good graphical interface could make these
  Randy> things very clear.
  Randy>    Self could also be used to teach about itself - (all
  Randy> of the above plus Self-specific semantics and syntax).  This could make
  Randy> it a nice environment for undergraduate projects.
  Randy>    What are your reactions to this general line of work?

In general, we're very excited by the possiblities Self shows in this
regard.

In particular, Self seems to have the potential to do for teaching
object-oriented programming what Abelson and Sussman does for teaching
procedural or functional programming.  As youpoint out, the minimalist
semantics is the key part here.  One could begin by teaching a single
object in isolation (as we do in our new first-semester course in
programming, which is object-oriented from day 1).  Adding multiple
objects is easy.  Adding more and more objects leads one to want to
factor commonality.  Hence we can introduce inheritance.  Classes fall
out naturally as soon as you notice that lots of objects have similar
behavior but different state.  As you need new functionality, you can
build it.  Need a conditional?  Build it.  And so forth.  Instead of
new language syntax and semantics, we simply use particular idioms.

So that's our thinking (in a nutshell) about how Self might be used
for teaching object-oriented programming.

We're not using Self in our first-semester programming course.  Why
not?  Principally because we felt we needed to minimize the cost of
switching the curriculum to object-oriented programming and we felt
that we should not stray too far from what is being used in industry
(the industry we're preparing our students for), namely C++.

  The shift to teaching object-oriented programming in the first
semester was (and continues to be) a big change.  We felt the need to
mimimize that change by sticking with a language pretty close to what
people were used to -- namely object extensions to Pascal based on
Turbo Pascal.  This means the language is similar in feel to C++
(which we use in later courses), and not too alien to students (and
faculty and TAs) who have been using Pascal up until now.
Unfortunately, we get all the pain of a complex hybrid language that
you might expect.

At this point, we've just finished the first semester of our
object-oriented first course -- a draft of our textbook is with
reviewers (we can send you a copy of this, too, if you'd like).  The
semester has been rough -- we've had to redesign the entire text,
syllabus, and every single assignment.  All of our TAs had to be
retrained for the new course.  It was, frankly, quite a lot of work.

As a consequence, for the next couple of years, we simply want to
smooth out the course.  We don't want to completely revamp the course
yet again right away -- it needs some stability.

However, that doesn't eliminate our interest in Self as a programming
language for a first course.  It is something we plan to continue
working on.  Many of the current TA staff, as well as the grad
students and members of the graphics group are interested in Self.
Interest is high enough that we're considering running a class in it,
perhaps this spring, investigating its usefulness both for graphics
and for a first course.