Programming Perl, 4th Edition

Recursive Patterns

When you reference a capture group from within the pattern, whatever was actually captured by that group is what gets used for the backreference.

/\b (        \p{alpha}+ ) \s+    \1    \b /x   # numbered backref
/\b (        \p{alpha}+ ) \s+  \g{1}   \b /x   # alternate syntax
/\b (        \p{alpha}+ ) \s+ \g{–1}   \b /x   # relative backref
/\b (?<word> \p{alpha}+ ) \s+ \k<word> \b /x   # named    backref

Those four examples all use backreferences to match the same word twice in a row. But sometimes you want to match two different words with the same pattern.

For that, a different syntax is used: (?NUMBER) calls back into the pattern of a numbered group, whereas (?&NAME) does so for a named group. (The latter is reminiscent of the & form of subroutine calls.)

/\b (        \p{alpha}+ ) \s+ (?–1)    \b /x   # "call" numbered group
/\b (?<word> \p{alpha}+ ) \s+ (?&word) \b /x   # "call" named group

With the NUMBER form, a leading minus sign counts groups right to left from the current location, so –1 means to call the previous group; you don’t have to know the absolute position of it. On the other hand, if the number has a plus sign in front, (?+NUMBER), you count forward that many groups to the right of where you are.

You are even allowed to call a group that you’re already in the middle of, causing the matching engine to recurse on itself. This is a normal thing to want to do. Here’s one way to match balanced parentheses:

/ ( \( (?: [^()]++ | (?1) )*+ \) )/x

Because the entire pattern is enclosed by capture parentheses, you ...