Subject: Re: Macro-writing in CL
From: Erik Naggum <erik@naggum.net>
Date: Thu, 07 Jun 2001 14:57:21 GMT
Newsgroups: comp.lang.lisp
Message-ID: <3200914634726937@naggum.net>

* "Biep @ http://www.biep.org/" <reply-via@my-web-site.com>
> Let me start by clarifying that I am NOT complaining about anything [...]

  Hmmm.

> I mean that people walk up to me, even though they are decent
> programmers, to have me get wrinkles out of their macro.  Of if they
> don't walk up to me, I can often construct a context in which their macro
> doesn't do what they had intended it to do.

  Seems to me you are indeed complaining instead of solving the problem.
  If you know this stuff so well (which I have yet to see evidence of),
  either your communication skills towards the people you are helping is
  lacking, or they fail to learn enough from you.  Write a paper on proper
  macro-writing and publish it.  From what I have seen so far, you should
  expect criticism of your narrow view of their uses and purposes.

> I know what a macro is, and how to write them.

  I am a fairly good marksman, and after having seen little improvement in
  my shooting in the past couple months, I worried to a fellow marksman,
  much, much better than me, that I had reached the apex of my skills.  He
  looked amused and said he had experienced the same thing several times
  and always needed advice from better marksmen to change some of the
  things he did not think needed improvement, some of them rather dramatic.
  One of the great things about truly useful skills is that there is always
  more to learn.  Optimization has the drawback that if you optimize the
  wrong way, for the wrong purpose, you end up being good at something that
  is simply _wrong_ to be good at.  If you still insist on improving, you
  may in fact get worse to an objective eye, since you fail to reach the
  goal.  (This part is fortunately impossible in shooting.  It is such an
  easy way to get feedback.)

> Of course.  A lot of research is going on (more in the Scheme community
> than in the CL community as far as I am aware), but I don't think the
> final solution has been found yet.  Obviously, until that happens we all
> happily go on hacking.

  In my view, the final solution on macros has already been found.  Scheme
  folks are continuing to "research" down the wrong path, as they have for
  many, many years.  They are like the "doctors" of the past who thought
  illness was in the blood, and go on with their "research" despite the
  discovery of penicillin elsewhere.

> About someone calling your macro in the context of an (flet ((list ..)) ..).

  The problems in Scheme are not the problems in Common Lisp.  If a Scheme
  jerk does that in Common Lisp, he breaks the rules, written and unwritten
  and should not blame the language for his ignorance.

  "Doctor, doctor, it hurts when I make this contorted movement."
  "So don't do it."

> CLtL2 saves you here

  You know, one of the symptoms of your lack of sufficient effort is that
  you think CLtL2 is the authoritative reference.  You may not be aware of
  this, but ANSI X3.226 Common Lisp has been issued, in fact was issued on
  1994-12-15.  It is now the authoritative reference on Common Lisp.  You
  can find hypertext versions on the Net if you too cheap to buy the real
  thing from ANSI.

> but if instead of 'list' it were a function you had defined yourself, you
> would still be bitten.

  But why should the macro be using it?  Obviously, the macro references an
  environment known to it, and draws on functions that are well-defined in
  it.  If you change those with flets, you should be fully aware of the
  consequences.  Incidentally, if you feel that this is such a drag, the
  rule that you should not redefine functions in the common-lisp package is
  not an arbitrary restriction.  It is good advice never to redefine a
  function locally that has a global, _published_ definition.  If you _do_
  do so, it is for the obvious side effect that you blame "gurus" for.

  Does this mean that the application programmer should be mortally afraid
  of screwing with somebody's names, like he would be in a one-namespace,
  no-packages language like Scheme?  Not at all.  Packages have published
  and unpublished symbols, mind you, so if you take care to review the
  published symbols in a package you have used, you follow the same rule
  for _any_ package you use as you do for the common-lisp package.  Only
  the standard could not have mandated the same thing for your symbols as
  it does for the standard symbols.

  I think the rule is really, really simple: If the symbol is not in your
  own package, be very cautious.  Now, there are some "problems" with the
  package symbol in that it would have been lovely if the various values of
  a symbol could be treated differently, but this quickly leads to serious
  problems, too.

  Some other languages require serious ownership protocols for objects.
  Common Lisp requires an ownership protocol for symbols, instead.  The
  former are runtime concerns..  The latter are read-time concerns and
  should be covered by interprogrammer communication, a.k.a. documentation.

> Lexical scoping would have saved you here, would have allowed you to
> write a protected lexical scope that would also have saved you from
> redefinitions.

  Really?  Are you sure you understand what lexical and dynamic scope are
  all about?  It sure looks like you have skipped a few classes on this.

> If doing something turns out to go wrong, and it can be fixed, I consider
> it a weakness simply to forbid doing it.

  An amazing attitude.  How is this reflected in your view of society in
  general?  Or are you one of those bondage-and-discipline programmers who
  want static type checking, class-owned methods, etc, too?  If so, why do
  you bother with Common Lisp at all?

> It is a bit of brittleness in an otherwise sturdy language.

  It is called "flexibility and responsibility".  It can be bent, but if
  you bend it out of shape or break it, that is your responsibility.  If
  you set up the rules so that you need never bend them, somebody else will
  feel imprisoned in your language.  I can say right now that I would not
  use the language you think is so much better than Common Lisp.

> I don't want to see CL turn into a "you cannot do that" language -- we
> have already enough of those.

  Precisely, but you argue very strongly in favor of just that.  Exactly
  like a politician who loves freedom as long as everybody does what he
  thinks is best for them.  Freedom in a society is not a question of what
  you can do within the rules, it is a question of what happens to those
  who want to break or change the rules because they think the rules are
  wrong.  Macros in Common Lisp makes it possible for freedom fighters to
  write their _own_ rules.  Some people are incredibly emotionally opposed
  to such a concept, calling it "anarchy" and all sorts of negative things.
  I read you to be such a person.

> By wrapping a scope around your functions that protects the original
> definition of list.

  This may be done with a code-walking macro.  Go write it.

> > > If your macro references a variable X, it just grabs whatever X
> > > happens to be available at the location where the macro is used.
> >
> > Yes. And thats what it is supposed to do (although I don't like the
> > term "grabs").
> 
> ..and that's dynamic scoping.

  Bzzzzzt!  Wrong!

> An X in a lambda doesn't behave that way: it always refers to the X in
> the context in which the lambda was defined.

  Which lambda are you talking about now?  Are you perchance unaware of the
  effect of special declarations, which do in fact make real dynamic scope?

> I have been way too much in the discussion of dynamic vs. lexical scoping
> for lambda's, and endlessly heard exactly the phrase you use "And that's
> what it is supposed to do", and don't have the energy any more to go into
> that discussion, sorry.  Nothing personal.  You might want to check
> http://www.acm.org/pubs/citations/proceedings/lfp/62678/p86-bawden/ if
> you are an ACM member.  (If you are not, you might find a working version
> of that paper at ftp://altdorf.ai.mit.edu/pub/scheme-reports/synclo.ps).

  Precisely.  You want syntactic closures.  You can have that in addition
  to macros.  You know how you can get that, do you not?  If not, let me
  remind you of how to write macros: Use copy-symbol for all symbols
  referenced in the macro body, and refer to the uninterned symbols you
  have created.  Pretty easy to do.  I suggest you show us how to do it,
  since you (1) know how to write macros, and (2) need this.  If you do
  not, I just conclude that you are one of those people who have a Scheme
  brain that cannot deal with Common Lisp.  There are a lot of those in the
  Scheme community.  It is one of the reasons I want such Scheme people to
  go to hell.  Fortunately, they are doing it on their own, anyway.  It is
  also one of the reasons I no longer consider Scheme a Lisp.

#:Erik
-- 
  Travel is a meat thing.