Ruby 1.9 Regex Lookbehind Assertion & Anchors

Question

Ruby 1.9 regex supports lookbehind assertion but I seem to have difficulty when passing anchors in the pattern. When anchors are passed in the lookahead assertion it runs just fine.

"well substring! "[/(?<=^|\A|\s|\b)substring!(?=$|\Z|\s|\b)/] #=> RegexpError: invalid pattern in look-behind: /(?<=^|\A|\s|\b)substring(?=$|\Z|\s|\b)/

Does anybody know how to make anchors work in lookbehind assertions as it does in lookahead?

Is there a special escape sequence or grouping that is required for lookbehind?

I have tested this behavior in 1.9.1-p243, p376 and 1.9.2-preview3 just in case it was patched.

Answer 1

Looks like the interpretation of the lookbehind is that of a range [] and not a group () like lookahead assertions. That possibly means \b is an invalid backspace character and not a word boundary.

"well substring! "[/(?<=^|\A|\s|[^\B])substring!(?=$|\Z|\s|\b)/]  #=> substring!
"well substring! "[/(?<=^|\A|\s|[^\B])substring(?=$|\Z|\s|\b)/]   #=> substring
"well !substring! "[/(?<=^|\A|\s|[^\B])substring(?=$|\Z|\s|\b)/]  #=> substring
"well !substring! "[/(?<=^|\A|\s|[^\B])!substring(?=$|\Z|\s|\b)/] #=> !substring

When all else fails... use a double negative!

Answer 2

Looks like you're right: \b works as expected in a lookahead, but in a lookbehind it's treated as a syntax error.

It doesn't really matter in this case: if (?<=^|\A|\s|\b) would have yielded the desired result, \b is all you needed anyway. The character following the assertion has to be s--a word character--so \b means either (1) the previous character is not a word character, or (2) there is no previous character. That being the case, ^, \A and \s are all redundant.

However, if the string starts with ! it's a different story. ^ and \A still match the beginning of the string, before the !, but \b matches after it. If you want to match !substring! as a complete string you have to use /\A!substring!\Z/, but if you only want to match the whole word substring you have to use /\bsubstring\b/.

As for [^\B], that just matches any character except B. Like \b, \B is a zero-width assertion, and a character class has to match exactly one character. Some regex flavors would throw an exception for the invalid escape sequence \B, but Ruby (or Oniguruma, more likely) lets it slide.