Scheme has a way of defining macros that's supposed to be better than old-fashioned defmacro. These "hygienic macros" seemed to a lot of people (though not me) to be a good idea when they were invented. They got into the Scheme standard then. I think people don't like them so much now, but once something gets into a standard it's impossible to get it out.

However (a) all the Scheme implementations I know of have implemented classic defmacro macros as well, and (b) if one hadn't, you could easily write it yourself. So in practice there's not much of a decision to make.

The Arc implementation you're using to read this is written on top of Scheme. Currently Mzscheme.

Hygienic macros do seem to be a good idea: "Hygienic macros are macros whose expansion is guaranteed not to cause collisions with existing symbol definitions." (from Wikipedia, http://en.wikipedia.org/wiki/Hygienic_macro)

What are the disadvantages? Is there any reason to dislike defining syntax rules?

One reason I've heard is that sometimes you actually want those collisions (I haven't come across a good example though - anyone know of one?).

For me the reason is just that defmacro is easier to read.

One example is the magical introduction of special variables. Here's "aif", implemented in PLT Scheme:

 (require (lib "defmacro.ss"))

 (define-macro (aif a b c)
   `(let ((it ,a))
      (if it ,b ,c)))

  (aif (* 10 10) (+ it 1) 'oops) ==> 101

The "aif" form makes it so you don't have to create a temporary variable name first -- the temp var is just assumed to be named "it".

If you're thinking, "whatever, I don't need that" you are wrong. :) Somehow this turns into the most wonderful little line-saver. At my last company, we used it all over the place.

Yup, I've used aif a lot, but had never got around to implementing it for myself and so hadn't made the connection with variable capture.

here's one:

 (defmacro aif (test then &optional else)
   `(let ((it ,test))
      (if it ,then ,else)))

In Scheme, the cases where one would use anaphoric IF can be handled using a COND clause with a "=>":

    (cond ((assoc 'key collection)
           => (lambda (it) (do-something-with it)))
          (else
              (alternative)))

Alternatively, if you really wanted to create an AIF macro, you could use SYNTAX-RULES to create something that sugared up the above thusly:

    (aif it (assoc 'key collection)
         (do-something-with it)
         (alternative))

Or you could, going outside the standard, use SYNTAX-CASE (ugh, in R6RS) or one of the other non R5RS approaches to macros e.g. syntactic closures[1] that don't throw out the macro hygiene baby with the occasional slight inconvenience bath-water.

I'm a bit of a reformed macrologist; if I get an idea for a cool macro, I try to see how far I can take a non-macrotic version and compare that to what the syntactically-sugared version would be like. Most of the time, I wind up not bothering with a macro. I put macros in the same category as EVAL: If you think you need to use a macro, think again.

[1]: (http://community.schemewiki.org/?syntactic-closures)

You are right about cond, but don't forget that cond and other Scheme syntax can be implemented as hygienic macros:

http://www.schemers.org/Documents/Standards/R5RS/HTML/r5rs-Z...

Macros are a build-your-own-language construct, but to a certain extent so are functions. I agree that one should try functions first, but would not go so far as to put macros in the same category as using eval.

I thought a lot of Scheme implementations define IF in terms of COND. Is there some reason you're pointing this out?

I should have been more detailed and noted that on the "by all means use it" vs. "if you think you need it, think again" continuum, EVAL requires IMO more justification than a macro. The source of my concern over macros is that when designing them, care needs to be taken to clearly communicate their semantics, which is best done IMO by following the established conventions of the language and whatever other macros the user is expected to be familiar with.

You pointed out one could achieve much the same result as the aif macro by using cond. I wanted to point out that cond itself is an example of the power of macros.

My point was more than one isn't usually missing much by being limited to writing hygienic macros. I don't think anyone is disputing the power of macros; it's the power/cost of hygiene violation that people are discussing.

Someone could give a "trust the programmer" argument for making unhygienic macros available in a language, but that nostrum is problematic because there are two programmers: the macro writer and the macro user. It's harder than you think to write a macro using DEFMACRO that always behaves the way a user might expect it to behave, and as a user of a macro, you need to wonder how hard the macro writer worked to protect you -- something that I don't like as a fan of black boxes.

I'm not familiar with the arguments against using eval. Can you enlighten me as to why it would be a bad practice?

Because EVAL is an atomic bomb, and there's no need to use it when a garrote would more elegantly perform the task at hand.

I'm not much into macros yet, but while syntax-case is a R6RS feature, I thought there is a portable implementation that runs on top of R5RS? At least Chicken 2.6 provides that, and its implementor is viscerally against R6RS.

Yes, SYNTAX-CASE has been around for quite a while, and there is a portable implementation that has just a few prerequisites beyond what R5RS provides.

http://www.cs.indiana.edu/chezscheme/syntax-case/

Funny, because I just started reading On Lisp and it's right there, page 191. Thanks for the example.

Ah, of course - thanks!

Thanks for the info! I learned Scheme with a rather obscure implementation (LMU Scheme, from my university), so I wasn't aware of the Macro capabilities of Scheme. Edit: I just checked, LMU Scheme is based on R4RS and indeed has no Macros.

Yes, but both are widely considered inferior to Visual Forth ++.

R5RS hygienic macros qualify as LISP goodness, but a lot of Scheme implementations also implement defmacro (like CL) and/or syntax-case.

Use of macros makes the code reflect the programmer's thinking about the problem domain. Whether this is readable or not depends on the programmer.

Macros produce kickass code: http://neverfriday.com/blog/?p=10#more-10

If I didn't create a string-case= macro, I would have to write the final expression over and over again. With multiple string-case matches I would have to write string=? many many times:

 (cond ((or (string=? "hello" my-string)
            (string=? "world" my-string))
        (print "match")))

To do that with the string-case= macro:

 (string-case= my-string
   (("hello" "world")
    (print "match")))

Saves quite a bit of typing and now I have a good example of why macros are good to have around :D

edit: formatting ftw.

If you put spaces in front of your code, news.yc will indent nicely.

 (cond ((or (string=? "hello" my-string)
	    (string=? "world" my-string))
	(print "match")))

 (string-case= my-string
   (("hello" "world") (print "match")))

That only saves you one quote and one lambda for each clause, right?

  (define (string-case= my-string . clauses)
    (cond ((null? clauses) 'whatever)
          ((string-member? my-string (caar clauses))
             ((cadar clauses)))
          (else (apply string-case= 
                       my-string 
                       (cdr clauses)))))

Where string-member? is trivial to define as a procedure too.

Then you use it like

  (string-case=
    ('("hello" "world")
       (lambda () (print "match"))))

I myself wouldn't use a macro for this little gain.

[Edit: I had completely misunderstood the macro.]

That should possible with a higher-order-function (and a lambda), too.