A positive lookahead (?=123) asserts the text is followed by the given pattern, without including the pattern in the match. Similarly, a positive lookbehind (?<=123) asserts the text is preceded by the given pattern. Replacing the = with ! negates the assertion.

---

Input: 123456

• 123(?=456) matches 123 (positive lookahead)
• (?<=123)456 matches 456 (positive lookbehind)
• 123(?!456) fails (negative lookahead)
• (?<!123)456 fails (negative lookbehind)

---

Input: 456

• 123(?=456) fails
• (?<=123)456 fails
• 123(?!456) fails
• (?<!123)456 matches 456

A lookbehind can be used at the end of a pattern to ensure it ends or not in a certain way.

([a-z ]+|[A-Z ]+)(?<! ) matches sequences of only lowercase or only uppercase words while excluding trailing whitespace.

Some regex flavors (Perl, PCRE, Oniguruma, Boost) only support fixed-length lookbehinds, but offer the \K feature, which can be used to simulate variable-length lookbehind at the start of a pattern. Upon encountering a \K, the matched text up to this point is discarded, and only the text matching the part of the pattern following \K is kept in the final result.

ab+\Kc

Is equivalent to:

(?<=ab+)c

---

In general, a pattern of the form:

(subpattern A)\K(subpattern B)

Ends up being similar to:

(?<=subpattern A)(subpattern B)

Except when the B subpattern can match the same text as the A subpattern - you could end up with subtly different results, because the A subpattern still consumes the text, unlike a true lookbehind.