The Artima Developer Community
Sponsored Link

Weblogs Forum
The Adventures of a Pythonista in Schemeland/9

6 replies on 1 page. Most recent reply: Nov 3, 2008 9:06 PM by Michele Simionato

Welcome Guest
  Sign In

Go back to the topic listing  Back to Topic List Click to reply to this topic  Reply to this Topic Click to search messages in this forum  Search Forum Click for a threaded view of the topic  Threaded View   
Previous Topic   Next Topic
Flat View: This topic has 6 replies on 1 page
Michele Simionato

Posts: 222
Nickname: micheles
Registered: Jun, 2008

The Adventures of a Pythonista in Schemeland/9 (View in Weblogs)
Posted: Oct 30, 2008 11:24 PM
Reply to this message Reply
Summary
This episode is entirely devoted to Scheme macros from a personal point of view. Pattern matching is introduced as the fundamental mechanism on which macros are built.
Advertisement

A minimal introduction to Scheme macros

Scheme macros have many faces. You can see them as a general mechanism to extend the syntax of base Scheme, and also as a mechanism to reduce boilerplate. On the other hand, if you focus your attention on the fact that they work at compile time, you can see them as a mechanism to perform arbitrary computations at compile time, including compile time checks.

I think the correct way of looking at macros is to see them as a general facility to write compilers for micro-languages - or Domain Specific Languages, DSL - embedded in the Scheme language. The languages defined through macros can be very different from Scheme; for instance you can define object oriented languages (object systems such as TinyCLOS or Swindle are typical examples) or even languages with static typing (the new language Typed Scheme, built on top of PLT Scheme, is such an example).

In order to address such use cases, Scheme macros have to be extremely advanced, much more than Lisp macros and any other kind of macros I know of; as a consequence, they also have a reputation for complexity. Unfortunately, on top of the intrinsic complexity, Scheme macros also suffers from accidental complexity, due to history (there are too many standard macro systems) and to the tradition of not caring about readability (many Scheme constructs are made to look more complex than they actually are).

Scheme has two macro systems included in the de jure standard - syntax-rules, which allows to define hygienic macros only, and syntax-case, which has the full power of Lisp macros and much more - plus a de facto standard based on define-macro system, which is available in all implementations and it is well known to everybody because of its strict similarity to Common Lisp defmacro system.

Which macrology should I teach?

Since there are so many macro systems it is difficult to decide from where to start in a pedagogical paper or tutorial. If you look at the original Italian version of this paper, you will see that I did talk about syntax-rules macros first. However, after a lot of thinking, I have decided to go my own way in this English series of the Adventures. Here I will not discuss syntax-rules, nor I will discuss syntax-case: instead, I will discuss my own version of Scheme macros, which I called sweet-macros.

Why I am doing that? After all, why my readers should study my own version of macros when they surely will be better served off by learning the standard macrology used by everybody? I have spent years debating with myself this very question, but at the end I have decided to go this way for a series of reasons:

  1. I regard the existence of two separate macro systems in the same standard as a wart of Scheme and as a mistake made by the R6RS editors: they should have included in the language syntax-case only, leaving syntax-rules as a compatibility library built on top of syntax-case;
  2. I really don't like the syntax-case syntax, it is by far too verbose and unreadable; I find there is a strong need for some sugar on top of it and that is what sweet-macros are for;
  3. sweet-macros are very close to syntax-case macros, so once you understand them you will understand syntax-case too; from there, understanding syntax-rules is a breeze;
  4. starting from sweet-macros is much better from pedagogical purposes, especially for readers with a Common Lisp background, since it is easy to explain the relation with defmacro and the hygiene issue;
  5. my target readers are programmers coming from the scripting languages (Perl/Python/Ruby) world. For this kind of public, with no previous exposition to Scheme, bare syntax-case is just too hard, so I needed to dress it in nice clothes to make it palatable;
  6. sweet-macros are intended to easier to use than syntax-case macros, but they are also more powerful, since they provide introspection and debugging capabilities as well as guarded patterns, so they should look attractive to experienced users too; however, this is a nice side effect and not the main motivation for the library;
  7. sweet-macros were written expressely for this series of papers, since I did not want to litter my explanation of Scheme macros with endless rants. So, I took action and I wrote my own library of macros made "right": this is also a tribute to the power of Scheme macros, since you can "fix" them from within the standard macro framework in fifty lines of code.

If you are an advanced reader, i.e. a Schemer knowing syntax-case/syntax-rules or a Lisper knowing defmacro, I a am sure you will ask yourself what are the differences of sweet-macros with respect to the system you know. I will make a comparison of the various systems in the future, in episode #12 and later on. For the moment, you will have to wait. I do not want to confuse my primary target of readers by discussing other macro systems right now. I also defer to episode #12 the delicate question are macros a good idea?. For the moment, focus on what macros are and how you can use them. Then you will decide if they are a good idea or not.

Enter sweet-macros

My sweet-macros library is a small wrapper around the syntax-case macro system. I release it under a liberal BSD licence. You can do whatever you want with it, just keep the attribution right.

http://www.phyast.pitt.edu/~micheles/scheme/pastine-alle-mandorle.jpg

The primary goal of sweet-macros is semplicity, so it only exports three macros, def-syntax, syntax-match and syntax-expand:

  • def-syntax is a macro used to define simple macros, which is similar to defmacro, but simpler and strictly more powerful.
  • syntax-match is a macro used to define complex macro transformers. It is implemented as a thin layer of sugar on top of syntax-case.
  • syntax-expand is a macro which acts as a debugging facility to expand macros defined via def-syntax or syntax-match.

It should be mentioned that standard Scheme macros do not provide debugging and/or introspection facilities and that every implementation provides different means of debugging macros. This is unfortunate, since debugging macros is usually difficult and it is done often, since it is uncommon to get a macro right the first time, even if you are an experienced developer.

I wanted to provide my readers with the tools to understand what they are doing, without relying on the details of the implementation they are using. Therefore macros defined via syntax-match (and that includes macros defined via def-syntax) provide out of the box introspection and debugging features.

Of course, readers who want to rely on the debugging tools of their implementation can do so; for instance I hear that DrScheme has a pretty good macro stepper but I have not tried it since I am an Emacs-addict.

First of all, you should download and install the right sweet-macros library. Unfortunately the R6RS module system does not really solve the portability issue (to my endless frustration) so I had to write different versions of the same library :-( I you are using Ikarus you should download the single file version of the library

$ wget http://www.phyast.pitt.edu/~micheles/scheme/sweet-macros.sls

and put it everywhere in you IKARUS_LIBRARY_PATH. If you are using PLT Scheme (you need a version of PLT newer than 4.0 for R6RS support) you must download the zip file version

$ wget http://www.phyast.pitt.edu/~micheles/scheme/sweet-macros.zip
$ unzip sweet-macros.zip
$ mv sweet-macros <your collects directory>

and install it in your collects directory, which on my machine is $HOME/.plt-scheme/4.0/collects.

Actually, the multifile version of the library works also with Ikarus if you have a recent enough version (right now I am using the trunk, version 0.0.3+, revision 1654). I have not tried the library on Larceny; I have tried it in Ypsilon Scheme which however has a small bug so that it does not run there (the bug is already fixed in the trunk). You should always keep in mind than R6RS implementations are pretty young and that implementors are still working to make them really compatible. I have also prepared an R5RS version which should work in Chicken Scheme, at least in the interpreter:

$ wget http://www.phyast.pitt.edu/~micheles/scheme/sweet-macros.scm

However I have developed and tested sweet-macros in Ikarus only (caveat emptor!). Still, since the title of this blog is The Explorer, I think it is fine if we deal with exploratory code.

You can check that the installation went well by importing the library:

$ ikarus
> (import (sweet-macros))

and by trying to define a macro.

An example: multi-define

Here is a multi-define binding construct which allows to define many identifiers at once:

(def-syntax (multi-define (name ...) (value ...))
  #'(begin (define name value) ...))

As you see, Scheme macros are based on pattern matching: we are giving instructions to the compiler, specifying how it must acts when it sees certain patterns. In our example, when the compiler sees a multi-define expression followed by two sequences with zero o more arguments, it must replace it with a begin expression containing a sequence of zero or more definitions. You can check that this is exactly what happens by means of syntax-expand:

> (syntax-expand (multi-define (a b) (1 2)))
(begin (define a 1) (define b 2))

Notice that (multi-define () ()) is valid code expanding to a do-nothing (begin) expression; if you want to reject this corner case, you should write your macro as

(def-syntax (multi-define (name1 name2 ...) (value1 value2 ...))
  #'(begin (define name1 value1) (define name2 value2) ...))

so that multi-define requires one or more arguments. However, it is often useful to accept degenerate corner cases, because they may simplify automatic code generation (i.e. multi-define could appear in the expansion of another macro).

multi-define works as you would expect:

> (multi-define (a b) (1 2)) ; introduce the bindings a=1 and b=2
> a
1
> b
2

I have just scratched the surface of Scheme macros here: I leave the rest for the next episode, don't miss it!


Michele Simionato

Posts: 222
Nickname: micheles
Registered: Jun, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 1, 2008 10:20 PM
Reply to this message Reply
Not many comments yet ...
How is it going, folks? Do you still think the pace is too slow (yes, I am referring to you, Merriodoc! ;)?
The episodes you are reading now are somewhat of an experiment in teaching macrology: I am presenting Scheme macros the way *I* think is most pedagogic, but I may be wrong, so some feedback from people who never had previous exposure to Scheme macrology would be welcome.
What would you like to see? More examples, more motivation, more code, less code, who knows? There are two more episodes on macros, and then a detour about functional programming in Scheme. Stay tuned!

Michele

Grant Rettke

Posts: 23
Nickname: grettke
Registered: Nov, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 2, 2008 8:48 AM
Reply to this message Reply
Your directions work for PLT Scheme. Though, syntax-expand isn't happy (not sure if this is user-error (aka me), it could very well be!)


#!r6rs

(import (rnrs)
(sweet-macros))

(def-syntax (multi-define (name ...) (value ...))
#'(begin (define name value) ...))

; (syntax-expand (multi-define (a b) (1 2)))
; C:\Documents and Settings\grettke\Application Data\PLT Scheme\4.1.1\collects\sweet-macros\helper1.sls:56:11: compile: bad syntax; reference to top-level identifier is not allowed, because no #%top syntax transformer is bound in: syntax-match

(multi-define (a b) (1 2))

(display a) (newline)

(display b) (newline)

Grant Rettke

Posts: 23
Nickname: grettke
Registered: Nov, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 2, 2008 8:50 AM
Reply to this message Reply
> Not many comments yet ...
> How is it going, folks?

I only starting reading your series early this morning :).

> Do you still think the pace is too
> slow (yes, I am referring to you, Merriodoc! ;)?

I wish you hadn't skipped syntax-rules. It is easy to make the argument to go from syntax-rules to syntax-case:

1. You need error checking and reporting.
2. You need to break hygience.

Once you get syntax-rules, syntax-case isn't so hard. I will trust you claim that sweet-macros will make learning syntax-case easier.

> The episodes you are reading now are somewhat of an
> experiment in teaching macrology: I am presenting Scheme
> macros the way *I* think is most pedagogic, but I may be
> wrong, so some feedback from people who never had previous
> exposure to Scheme macrology would be welcome.

That is not me.

> What would you like to see? More examples, more
> motivation, more code, less code, who knows? There are two
> more episodes on macros, and then a detour about
> functional programming in Scheme.

All of the above.

Michele Simionato

Posts: 222
Nickname: micheles
Registered: Jun, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 2, 2008 8:59 AM
Reply to this message Reply
> Your directions work for PLT Scheme. Though, syntax-expand
> isn't happy

I can't say I am surprised. We need some PLT expert to step on and explain what's happening. sweet-macros are stretching the limits of R6RS portability. They also are quite sensitive to bugs in syntax-case implementations, but in this case I think it is yet another issue related to phase separation.

Michele Simionato

Posts: 222
Nickname: micheles
Registered: Jun, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 2, 2008 9:35 AM
Reply to this message Reply
> I wish you hadn't skipped syntax-rules.

By now you should have realized that syntax-rules macros are a special case of sweet-macros (syntactically you get syntax-rules macros by removing the #' characters from a sweet macro) so by teaching sweet-macros I am also teaching syntax-rules (and much more).

Michele Simionato

Posts: 222
Nickname: micheles
Registered: Jun, 2008

Re: The Adventures of a Pythonista in Schemeland/9 Posted: Nov 3, 2008 9:06 PM
Reply to this message Reply
> > Your directions work for PLT Scheme. Though,
> syntax-expand
> > isn't happy
>
> I can't say I am surprised. We need some PLT expert to
> step on and explain what's happening. sweet-macros are
> stretching the limits of R6RS portability. They also are
> quite sensitive to bugs in syntax-case implementations,
> but in this case I think it is yet another issue related
> to phase separation.

It was, and Abdul Aziz Ghuloum figured out how to fix it. If you download sweet-macros.zip again it should work now.

Flat View: This topic has 6 replies on 1 page
Topic: New Control Structures for Java Previous Topic   Next Topic Topic: The Adventures of a Pythonista in Schemeland/7

Sponsored Links



Google
  Web Artima.com   

Copyright © 1996-2019 Artima, Inc. All Rights Reserved. - Privacy Policy - Terms of Use