The Universe of Discourse


Mon, 13 Feb 2006

Accidental Features
Last week I was away in Chicago as part of my Red Flags world tour, about which more in an upcoming entry. I mentioned before that I always learn something surprising from giving these classes, and this time was no exception. I was reviewing a medium-sized piece of code for a client and I encountered the following astonishing construction:

        for $i qw(foo bar baz) {
          # do something with $i
        }
If you're not familiar with Perl, you aren't going to see what is wrong with this. You may not see it for a while even if you are familiar with Perl.

Normally, the syntax for this sort of construction is:

        for $VAR (LIST) {
          ...
        }
where the parentheses around the LIST are not optional. And the qw(...) construction generates a list. So it should be:
        for $i (qw(foo bar baz)) {
          # do something with $i
        }
Leave off the outer parentheses, and you should get a syntax error. So when I saw this code, I said, "Oh, that's a syntax error."

Except, of course, it wasn't. The code I was reviewing was in production on the client's web site, and was processing millions of dollars of orders. If it had bee a syntax error, they would have known right away. And indeed, it worked when I tried it.

But it's not documented anywhere, and I have never seen it before. It is certainly little-known. I believe it is an accidental feature.

This is not the first time that someone has discovered an unintended but useful feature in Perl. Another example that comes to mind is the s/.../.../ee trick. Normally, you give s/.../.../ a pattern and a replacement string, and it performs a substitution, locating some string that matches the pattern and replacing it with the replacement. For example, s/\d+/carrots/ locates a a numeral (a sequence of digits) and replaces it with carrots. But you might want to do something tricky; say you want to replace each numeral with its double. Then you can use s/(\d+)/$1 * 2/e. The $1 is replaced with the digits, so that if the digits are 123, you get 123 * 2, and the trailing e means that this is a Perl expression, rather than a literal string, and should be evaluated. Without the e, 123 becomes 123 * 2; with e, it becomes 246. (Other modifiers were permitted: for example, i made the pattern match case-insensitive.)

The implemenation was to scan the modifiers left-to-right, and to call Perl's expression evaluator if an e was seen. The accidental feature was that the implementation therefore allowed one to put two es to force two evaluations. The typical use of this was to write something like s/(\$\w+)/$1/ee, which finds $foo and replaces it with the value of the variable $foo. The first e evaluates $1 to $foo, and the second evaluates $foo to its value. I believe this was discovered by Tom Christiansen. It's rarely used now, since there are more straightforward ways to accomplish the same thing, but it still appears in three different places in the documentation.

I don't think it says anything good about Perl that this sort of accidental feature shows up in the language. I can't think of anything comparable in any other language. (Except APL.) C has the 4[ptr] syntax, but that's not a feature; it's just a joke. The for $i qw(...) { ... } syntax in Perl seems genuinely valuable.

Accordingly, I dropped a note to the Perl development mailing list, asking if anyone had ever noticed this before, and whether people thought we should commit to supporting it in the future. (It works in all versions of Perl that support qw().) I was surprised that there was no reply. So I suppose that the next step is to prepare a documentation patch describing it and some tests in the test suite for it. If everyone is still asleep when I post those, they may go in without comment, and once a feature is put in, it is very difficult to remove it.


[Other articles in category /prs] permanent link