home | O'Reilly's CD bookshelfs | FreeBSD | Linux | Cisco | Cisco Exam  


Perl CookbookPerl CookbookSearch this book

1.12. Expanding Variables in User Input

1.12.3. Discussion

The first technique is basically to find what looks like a variable name, then use symbolic dereferencing to interpolate its contents. If $1 contains the string somevar, ${$1} will be whatever $somevar contains. This won't work if the use strict 'refs' pragma is in effect because that bans symbolic dereferencing.

Here's an example:

our ($rows, $cols);
no strict 'refs';                   # for ${$1}/g below
my $text;

($rows, $cols) = (24, 80);
$text = q(I am $rows high and $cols long);  # like single quotes!
$text =~ s/\$(\w+)/${$1}/g;
print $text;
I am 24 high and 80 long

You may have seen the /e substitution modifier used to evaluate the replacement as code rather than as a string. It's designed for situations where you don't know the exact replacement value, but you do know how to calculate it. For example, doubling every whole number in a string:

$text = "I am 17 years old";
$text =~ s/(\d+)/2 * $1/eg;

When Perl is compiling your program and sees a /e on a substitute, it compiles the code in the replacement block along with the rest of your program, long before the substitution actually happens. When a substitution is made, $1 is replaced with the string that matched. The code to evaluate would then be something like:

2 * 17

If we tried saying:

$text = 'I am $AGE years old';      # note single quotes
$text =~ s/(\$\w+)/$1/eg;           # WRONG

assuming $text held a mention of the variable $AGE, Perl would dutifully replace $1 with $AGE and then evaluate code that looked like:

'$AGE'

which just yields us our original string back again. We need to evaluate the result again to get the value of the variable. To do that, just add another /e:

$text =~ s/(\$\w+)/$1/eeg;          # finds my( ) variables

Yes, you can have as many /e modifiers as you'd like. Only the first one is compiled and syntax-checked with the rest of your program. This makes it work like the eval {BLOCK} construct, except that it doesn't trap exceptions. Think of it more as a do {BLOCK} instead.

Subsequent /e modifiers are quite different. They're more like the eval "STRING" construct. They don't get compiled until runtime. A small advantage of this scheme is that it doesn't require a no strict 'refs' pragma for the block. A tremendous advantage is that unlike symbolic dereferencing, this mechanism finds lexical variables created with my, something symbolic references can never do.

The following example uses the /x modifier to enable whitespace and comments in the pattern part of the substitute and /e to evaluate the righthand side as code. The /e modifier gives more control over what happens in case of error or other extenuating circumstances, as we have here:

# expand variables in $text, but put an error message in
# if the variable isn't defined
$text =~ s{
     \$                         # find a literal dollar sign
    (\w+)                       # find a "word" and store it in $1
}{
    no strict 'refs';           # for $$1 below
    if (defined ${$1}) {
        ${$1};                    # expand global variables only
    } else {
        "[NO VARIABLE: \$$1]";  # error msg
    }
}egx;

Once upon a time, long ago and far away, $$1 used to mean ${$}1 when it occurred within a string; that is, the $$ variable followed by a 1. This was grandfathered to work that way so you could more readily expand the $$ variable as your process ID to compose temporary filenames. It now always means ${$1}, i.e., dereference the contents of the $1 variable. We have written it the more explicit way for clarity, not correctness.



Library Navigation Links

Copyright © 2003 O'Reilly & Associates. All rights reserved.