The Universe of Discourse
           
Wed, 20 Sep 2006

The world's worst macro preprocessor
Last week I added another plugin to my Blosxom installation. As I wrote before, the sole benefit of Blosxom is that it's incredibly simple and lightweight. So when I write plugins for it, I try to keep them incredibly simple and lightweight, lest I spoil the single major benefit of Blosxom. Sometimes I'm more successful, sometimes less so. This time I think I did a good job.

The goal last time was a macro processor. I write a lot of math articles. I get tired of writing <sup>2</sup> every time I want a superscript 2. Even if I bind a function key to that sequence of characters, it's hard to read. But now, with my new Blosxom macro processor, I just insert a line into my article that says:

  #define ^2 <sup>2</sup>

and for the rest of the article, ^2 is expanded to <sup>2</sup>.

This has turned out really well, and I'm using it for all sorts of stuff. I use it for math notations, such as for making -> an abbreviation for &rarr; (→), and for making ~ an abbreviation for &not; (¬).

But I've also used it to #define Godel G&ouml;del. I've used it to #define KK <b>K</b> and #define SS <b>S</b>, which makes an article I'm writing about combinatory logic readable, where it wasn't readable before. In my recent article about job hunting, I used it to #define CV r&eacute;sum&eacute;, which saved me from having to interrupt my train of thought several times in the article.

There are some important points about the design that I think I got right on the first try. Whenever you write a macro system, you have to ask about escape sequences: what do you do if you don't want a macro expanded? For example, in the combinatory logic article I defined a macro SS. This meant that if I had written MOUSSE in the article somewhere, it would have turned into MOUSE. How should I prevent that kind of error?

Answer: I don't. I'm unlikely to do that. But if I do, I'll pick it up during the article proofreading phase. If I can't avoid writing MOUSSE, I have two choices: I can change the name of the SS macro to something easier to avoid—like S*, say, or I can define a second macro: #define !MOUSSE MOUSSE. But so far, it hasn't come up.

One alternative solution is to say that macros are expanded only in certain contexts. For example, SS might only be expanded when it is a complete word, not when it is in the middle of a word, as MOUSSE. I resisted this solution. It is much simpler to remember that every macro is expanded everywhere. And it it is much easier to fix the problem of a macro being expanded when I don't want it than it is to fix the problem of a macro not being expanded when I do want it. So every macro is expanded no matter where it appears.

Related to the unintentional-expansion issue is that each article has its own private macro set. I don't have to worry that by defining a macro named -> in one article that I might be sabotaging my opportunity to actually write -> in some unknown future article. Each set of macros can be totally ad hoc. I don't have to worry about global tradeoffs. Do I #define --- &mdash;, knowing that that will foreclose my opportunity to use --- in any other way? I can make the decision based on simple, local information.

It would have been tempting to over-engineer the system and add all sorts of complex escape facilities. I think I made the right choice here by not doing any of that.

Another escaping issue: What if I want to write something that looks like a definition but isn't? Here I avoided the problem by choosing a definition syntax that I was unlikely to write in any other context: #define in the leftmost column indicates a definition. In this article, I had to write some similar text. It was no trouble to indent it a couple of spaces, disabling the special meaning. But HTML is already full of escape mechanisms, and it would have been no trouble to write &#35;define instead of #define if for some reason I had really needed it to appear in the leftmost column. (Unlikely anyway, since HTML has no column semantics.)

Another right choice I think I made was not to parametrize the macros. An article on algebra might well have:

  #define ^2 <sup>2</sup>
  #define ^3 <sup>3</sup>
and it might be oh-so-tempting to try to eliminate the duplication à la C:

  #define ^(\w+) <sup>$1</sup>
I did not do this. It would have complicated the processing substantially. It would also have complicated the use of the package substantially: I would have to worry a lot more than I do about invoking macros unintentionally. And it is not needed. Not so far, anyway. Because macro definitions only last for the duration of the article, there is no pressure to make a complete or consistent set of definitions. If an article happens to use the notations 2, i, and N, I can define macros for those and only those notations.

Also tempting is to extend the macro system to support something like this:

  #define BF(.*) <b>$1</b>
I have so far resisted this. My feeling is that if I want to do anything like this, I should take it as a sign that I should be writing the articles in some markup system other than HTML. Choice of that markup system should be made carefully, and not organically as an ad-hoc overburdening of the macro system.

I did run into one trouble with the macro system. Originally, it was invoked before some of my other plugins and after others. The earlier plugins automatically inserted certain text into the article that sometimes accidentally triggered my macros. I have not had any trouble with this since I changed the plugin order to invoke the macro processor before any of the other plugins.

The macro-processing code is about 19 lines long, of which three are diagnostic. It is the world's worst macro system. It has exactly one feature. It is, I think the simplest thing that could possibly work, and so a good companion to Blosxom. For this application, the world's worst macro system is the world's best.

[ Addendum 20071004: There's now a one-year retrospective analysis. ]


[Other articles in category /prog] permanent link