The Universe of Discourse


Wed, 25 Jan 2006

Red Flags world tour: New York City
My wife came up with a brilliant plan to help me make regular progress on my current book. The idea of the book is that I show how to take typical programs and repair and refurbish them. The result usually has between one-third and one-half less code, is usually a little faster, and sometimes has fewer bugs. Lorrie's idea was that I should schedule a series of talks for Perl Mongers groups. Before each talk, I would solicit groups to send me code to review; then I'd write up and deliver the talk, and then afterward I could turn the talk notes into a book chapter. The talks provide built-in, inflexible deadlines, and I love giving talks, so the plan will help keep me happy while writing the book.

The first of these talks was on Monday, in my home town of New York.

'It makes no odds whether a man has a thousand pound, or nothing, there. Particular in New York, I'm told, where Ned landed.'

'New York, was it?' asked Martin, thoughtfully.

'Yes,' said Bill. 'New York. I know that, because he sent word home that it brought Old York to his mind, quite wivid, in consequence of being so exactly unlike it in every respect.'

(Charles Dickens, Martin Chuzzlewit, about which more in some future entry, perhaps.)

The New Yorkers gave me a wonderful welcome, and generously paid my expenses afterward. The only major hitch was that I accidentally wrote my talk about a submission that had come from London. Oops! I must be more careful in the future.

Each time I look at a new program it teaches me something new. Some people, perhaps, seem to be able to reason from general principles to specifics: if you tell them that common code in both branches of a conditional can be factored out, they will immediately see what you mean. Or so they would have you believe; I have my doubts. Anyway, whether they are telling the truth or not, I have almost none of that ability myself. I frequently tell people that I have very little capacity for abstract thought. They sometimes think I'm joking, but I'm not. What I mean is that I can't identify, remember, or understand general principles except as generalizations of specific examples. Whenever I want to study some problem, my approach is always to select a few typical-seeming examples and study them minutely to try to understand what they might have in common. Some people seem to be able to go from abstract properties to conclusions; I can only go from examples.

So my approach to understanding how to improve programs is to collect a bunch of programs, repair them, take notes, and see what sorts of repairs come up frequently, what techniques seem to apply to multiple programs, what techniques work on one program and fail on another, and why, and so on. Probably someone smarter than me would come up with a brilliant general theory about what makes bad programs bad, but that's not how my brain works. My brain is good at coming up with a body of technique. It's a limitation, but it's not all bad.


The goal of generalization had become so fashionable that a generation of mathematicians had become unable to relish beauty in the particular, to enjoy the challenge of solving quantitative problems, or to appreciate the value of technique.

(Ronald L. Graham, Donald E. Knuth, Oren Patashnik, Concrete Mathematics.)

So anyway, here's something I learned from this program. I have this idea now that you should generally avoid the Perl . (string concatenation) operator, because there's almost always a better alternative. The typical use of the . operator looks like this:

          $html = "<a href='".$url."'>".$hot_text."</a>";
It's hard to see here what is code and what is data. You pretty much have to run the Perl lexer algorithm in your head. But Perl has another notation for concatenating strings: "$a$b" concatenates strings $a and $b. If you use this interpolation notation to rewrite the example above, it gets much easier to read:

          $html = "<a href='$url'>$hot_text</a>";
So when I do these classes, I always suggest that whenever you're going to use the . operator, you try writing it as an interpolation too and see which you like better.

This frequently brings on a question about what to do in cases like this:

          $tmpfilealrt = "alert_$daynum" . "_$day" . "_$mon.log" ;
Here you can't eliminate the . operators in this way, because you would get:
          $tmpfilealrt = "alert_$daynum_$day_$mon.log" ;
This fails because it wants to interpolate $daynum_ and $day_, rather than $daynum and $day. Perl has an escape hatch for this situation:
          $tmpfilealrt = "alert_${daynum}_${day}_$mon.log" ;
But it's not clear to me that that is an improvement on the version that used the . operator. The punctuation is only slightly reduced, and you've used an obscure notation that a lot of people won't recognize and that is visually similar to, but entirely unconnected with, hash notation.

Anyway, when this question would come up, I'd discuss it, and say that yeah, in that case it didn't seem to me that the . operator was inferior to the alternatives. But since my review of the program I talked about in New York on Monday, I know a better alternative. The author of that program wrote it like this:

          $tmpfilealrt = "alert_$daynum\_$day\_$mon.log" ;
When I saw it, I said "Duh! Why didn't I think of that?"


[Other articles in category /prs] permanent link