Comment by gruseom
16 years ago
A classic piece which contains one of the best things ever written about design:
Take the hardest and most profound thing you need to do, make it great, and then build every easier thing out of it.
It's interesting to me that the idea that inspired Kay to this insight was FEXPRs: letting functions control the evaluation of their arguments. FEXPRs have long been banished from the Lisp world, which makes a sharp distinction between functions and macros (only the latter get to control evaluation). I still don't quite understand the reasons for this, and it seems like Kay doesn't either -- or at least didn't when he was inventing Smalltalk.
There's no real difference between a FEXPR, and a macro that calls eval or apply during expansion. The real meaning of "banishing" FEXPRs is that we actually get something extra: guaranteed-non-macro functions that behave according to a predictable pattern (evaluate arguments, pass them in, return a value.) Functions are basically a codified Lisp design-pattern, while macros (or FEXPRs, as they may be) are the real core of Lisp's abstraction mechanism, and really what make Lisp, Lisp.
It's just the same with Smalltalk: functions take lambdas as arguments, not primitive or composite values. "Receives values" is simply a design pattern.
That's a pretty good explanation; if it's accurate, I wonder why no one else explains it that way.
In Smalltalk, I thought arguments are always evaluated unless they're blocks. Is that wrong? Or is it that, at a lower level, everything's a block, and functions are implemented on top of that as something that always evaluates args?
No, you're right—arguments are always evaluated unless they're blocks in the full Smalltalk design we have today. However, like I said, this is a codified design pattern—the language could get away without doing it while still retaining the "core Smalltalk conceit" of everything being a message sent to an object. This is basically what the io language does—it's basically a Smalltalk at its core, but with less cruft.
There's a bit of disagreement on the benefits of FEXPRs in the comments of this earlier HN discussion: http://lambda-the-ultimate.org/node/3640
Interesting, there's a researcher by the name of John Shutt who in said "I suspect them of offering fundamentally greater abstractive power than any other programming language feature yet devised" in 2009. http://lambda-the-ultimate.org/node/3640#comment-51563
This is his language which supports FEXPRs: http://web.cs.wpi.edu/~jshutt/kernel.html