Tuesday, July 08, 2008

Lisp: Hackable by Default

I'm watching the MIT lectures for the Structure and Interpretation of Computer Programs which my buddy Mike Cheponis was nice enough to give to me. I noticed a few interesting things.

In C, it's common to write "x->a". x is a pointer to a struct with a member a. In Lisp, you'd write (a x). a is a function that operates on some object x in order to return a value. That means that, by default, "a" has some "wiggle room" to do something special because it's a function. That's why in Java, the norm is to use getters and setters for everything. It provides "wiggle room" which I also call "hackability". In languages like Python, Ruby, and C#, "x.a" might or might not be an actual function call because those languages support properties. Depending on the details, in languages with inheritance, you might even say there's an extra dimension of hackability. For instance, in Java, x.getA() might be extended by subclasses.

One of the lectures covered symbolic manipulation of algebraic expressions. That means writing a system that can recognize that "(+ x x)" is the same as "(* x 2)". The professor had a procedure make-sum. "(make-sum x y)" is pretty straightforward. Under the covers, it returns "(list '+ x y)". However, he showed that there's no reason make-sum shouldn't be smarter. For instance "(make-sum x 0)" should really just return x. Internally, I was thinking of make-sum as a Java-like constructor for some class named Sum. However, a Java constructor can't just return x. It has to return an instance of the Sum class. Hence, if you were coding this in Java, you wouldn't use a constructor. You would use a static method that returns an Expression, where Expression is some interface. My point is that in Lisp make-sum is just a constructor, and constructors in Lisp by default have the freedom to return an instance of whatever they want. That means they're hackable by default.

I have one more thing I'd like to mention. Look at the image. This was filmed in 1986. First of all, it's amazing to me that something filmed 22 years ago is still very interesting and relevant to me today. It's even more amazing to consider that Lisp started in 1958. However, what's intriguing is that there are several women in the class. Not just one or two, but several. Where'd all the women go? There are so few female programmers these days.


swheatley said...
This comment has been removed by the author.
swheatley said...

(reposted with a working link)

Thought I'd post a link to the video series:


Anonymous said...

Yes. There are much more women in computer science classes than working in the software industry... You can check http://www.devchix.com

Shannon -jj Behrens said...

> Thought I'd post a link to the video series:

Very helpful. Thanks!

Edin Salković said...

There are also video lectures that cover the newer edition of the book at the ArsDigita University website.

Anonymous said...

When i was watching these classes it happened frequently that individuals of the audience looked a bit "displaced". Not interested, not able to follow, maybe just tired. After all it was however produced for internal company use originally (HP IIRC). No company video without propaganda.. therefore it would make sense to include as many minorities of that time in this specific industry. And then we have it.. ;-)

Mark said...

>However, a Java constructor can't just return x. It has to return an instance of the Sum class. Hence, if you were coding this in Java, you wouldn't use a constructor.

Bjarne Stroustrup discusses a similar scenario for implementing efficient matrix multiplication in C++ in section 22.4.7 of "The C++ Programming Language, Third Edition". He winds up defining overloaded operators that return auxilary classes representing computations to be performed. He sums it up by saying:

"This technique is based on the idea of using compile-time analysis and closure objects to transfer evaluation of subexpression into an object representing a composite operation. It can be applied to a variety of problems with the common attribute that several pieces of information need to be gathered into one function before evaluation can take place. I refer to the objects generated to defer evaluation as composition closure objects or simply compositors."