For starters, perlop says that the qr operator looks like this: qr/STRING/imosx - g isn't allowed.

Using $stream =~ /$regex/g still only compiles the regex once as the way that perl interpolates the $regex into a match operator should handle this for free (as in CPU time).

OK, that seems to make sense... So presumably the following idiom is the right way to go about things...

while ($stream =~ /$regex/g) {
    print "token: $1\n";
}
[download]

A compiled regex (qr//) does not take the place of the pattern matching operation. It merely holds the regex.

If you want to know when qr// is useful, that's another question that I'll answer (if you ask).

OK, consider it asked. I presume that the most useful thing about qr// is that it allows you to pass regular expressions around as arguments to functions and such, but honestly my lack of experience with it leaves me wondering.

When you have variables in a regex, Perl examines the contents of those variables to see if the overall representation of the regex has changed:

for ("ab", "cd") {
  if ($str =~ /$_/) { ... }
}
[download]

In the above code, the regex 'ab' is compiled and executed, and then the regex 'cd' is compiled and executed. Compare that with:

($x, $y) = ("ab", "c");
for (1, 2) {
  if ($str =~ /$x$y/) { ... }
  ($x, $y) = ("a", "bc");
}
[download]

Here, even though $x and $y change, the ACTUAL regex ('abc') does not change, so the regex is compiled only once. The process that Perl does internally is this:
1. take regex at this opcode
2. interpolate any variables
3. compare with previous value of the regex at this opcode
4. compile if different
5. execute this regex
When you use the /o modifier, it tells Perl that after it has compiled the regex, it should SKIP steps 2-4 of this process, meaning that the regex at this opcode will NEVER change.

So what is qr// good for? Consider this:

my @strings = make_10_strings();
for (@strings) {
  for my $p ('x+', 'yz?y', 'xz+y') {
    if ($_ =~ $p) { handle($_) }
  }
}
[download]

This code compiles a grand total of 30 regexes. Why? Because for each string in @strings we've got three patterns to execute, and because each time the $_ =~ $p is encountered the contents of $p has changed, the regex is compared and recompiled each time. Now sure, you could reverse the order of the loops, but that will result in the calls to handle() happening in a different order.

So enter the qr//.

When Perl sees a regex comprised solely of a single variable, Perl checks to see if that variable is a Regexp object (what qr// returns). If it is, Perl knows that the regex has already been compiled, so it simply uses the compiled regex in the place of the regex. That means doing:

my @strings = make_10_strings();
for (@strings) {
  for my $p (qr/x+/, qr/yz?y/, qr/xz+y/) {
    if ($_ =~ $p) { handle($_) }
  }
}
[download]

is considerably faster. There is no additional compilation happening. It's probably even better to move the qr// values into an array, but that might be moot since they're made of constant strings in this example. The point is, the use of qr// in a looping construct is the primary benefit it offers. Yes, it helps break a regex up into pieces too, but that's just a matter of convenience.

Be warned that the benefit of qr// objects is lost if there is additional text in the pattern match. I mean that $foo =~ /^$rx_obj$/ suffers from the same problem as $foo =~ /^$text$/.

_____________________________________________________
Jeff japhy Pinyan, P.L., P.M., P.O.D, X.S.: Perl, regex, and perl hacker
How can we ever be the sold short or the cheated, we who for every service have long ago been overpaid? ~~ Meister Eckhart

Another use for qr// is to break up unmanageably complex regular expressions into simpler, named, self-contained pieces. (There's a direct parallel here with subs, which do the same for 'ordinary' Perl code. In fact, you can consider a named regex to be just a function written with a funny-looking syntax: its input is a string and its output is either a Boolean value or one or more strings, depending on whether it captures anything.)

Here's an example from a code-filtering assertions module (yes, another one) that's not yet tested thoroughly enough to submit to CPAN:

Read more... (4 kB)

The full spec for qr in perlop is qr/STRING/imosx, so what you observed is the expected behavior.

Hm. I was looking in the Camel book and I didn't see that. I'll have to remember to look at the man pages as well in the future.