Re (tilly) 1: Lisp is More Evaluatable

[tilly]

Since the Perl has been commented on enough, I will comment on the Lisp.

First of all I suspect that delay is not doing what you think it is doing. For a start in, say, Common Lisp your code crashes and burns because delay is a special form in Scheme, but not Common Lisp. But beyond that delay is meant to be used with force, which you are not doing. If you look at this description you will find that when you use a delay without force the result is implementation dependent, but may be indistinguishable from the immediate value. Even if it is not indistinguishable, the code will run at most once and be memoized.

Therefore your Lisp code is more portably and less deceptively written as:

(setq closure (setq z (+ 2 2)))
(eval closure) ;;; works

because the code that defined closure is now evaluated by having been used, and further "evals" of it should no longer affect the value of z. (ie The promise set up by delay in your code was redeemed in handing a 4 to eval. Further evals will continue to eval 4, but will not run the promise again.)

Furthermore if you try to make the Lisp more like the Perl, for instance by making closure an anonymous function, you will find that the call semantics no longer work through an eval. As people have pointed out, that is because your Perl appears to reflect a deep misunderstanding about what all of these programming constructs should be doing.

So while I believe that Lisp is more evaluatable than Perl, I do not think this is an example that illustrates the principle.

BTW further stylistic points. In Lisp it is considered bad style to write code with side effects. As the advice goes, pretend that operations like setq have a penalty for use, and try to avoid them. Your pretend example of trying to set up a closure that will modify a global at a later date is indescribably bad - I would shudder to face code like that. In any language.

Secondly in Lisp it is almost never a good idea to actually use eval. Use macros instead. Just as much power delivered in a more structured form.

What both of these points get at is that writing debuggable code is a good idea. You seem to evaluate constructs on the basis of their ability to get very complex. You seem to entirely discount any notions that lead to debuggable code. I am coming to the conclusion that if the general programming world judged like you seem to, then goto would have never fallen out of favour...

[hding]

It's really not bad style to write code with side effects (depending, I suppose, on your precise definition thereof). Of course, it's (usually) bad style to write code with random, unpredictable, or unclear side effects, just as in any language. But look in any well-regarded Lisp book like Norvig or one of Graham's and you'll see side effects, setfs, etc. It'd be hard to use CLOS, for example, without them. If one's intent is to use primarily the functional features of Lisp, then I guess I agree, but in practice, one uses the imperative and OO features just as much, it seems.

You're spot on about eval, though.

Oh, one other thing. For the code to do what (I guess) is actually desired, one needs to quote the form to be evaled:

(setq closure '(setq z (+ 2 2)))
(eval closure) ;;; works

As originally written, both closure and z would be set to 4, and the eval form would simply eval 4 to get 4

[tilly]

Perhaps "bad style" is too strongly put.

But it is certainly considered good style to avoid side-effects. And the advice of thinking of setq as having a penalty associated is straight from On Lisp by Graham. It is not a religious rule. But it is worth thinking about. In fact I think about it in Perl as well, though I am not as extreme because Perl cannot optimize tail recursion.

As for the other point, perhaps I was unclearly verbose. But that is exactly what I was getting at with the phrase "...should no longer affect the value of z."

[hding]

Probably we more or less agree, but have different ways of expressing what we mean. Graham is good, but I think it's dangerous to rely too much on what he says. Just as an example, he seems to have a disdain for the loop macro that just isn't shared by the community at large, and if unduly under his influence one might neglect to use it even in situations where it's clearly the easiest and best solution. But (like Perl) there's room enough in Lisp for everyone, at least IMHO. :-)

One should be careful about relying on tail recursion being optimized in Common Lisp, too. The spec doesn't require it (unlike Scheme which does). I think most implementations will do it under certain but not all circumstances. For example, in the one I use, you need to have debug set below 3 in the code to be compiled. I don't have a good enough sense of history to speculate whether this is because of or the cause of Common Lispers being a lot more willing to iterate than Schemers.