How to reset entire match at the end of a capturing group?

Question

Suppose I have the following text:

Yes: [x]
Yes: [  x]
Yes: [x  ]
Yes: [  x  ]
No: [x
No: x]

I am interested in a regex expression with two capturing groups as follows:

• group $1: match the brackets [ and ] that contain an x in between with a variable amount of horizontal space around x. I can use The fourth bird's answer based on a branch reset group for this:

  • regex: (?|(\[)(?=\h*x\h*])|(?<=\[)\h*x\h*(]))

  • which results in:

    

• group $2: should match the x that is contained in the brackets [ and ]. For this, I can use a non-capturing group with a combination of positive lookbehind and lookahead assertions:

  • regex: (?:(?<=\[)\h*(x)(?=\h*]))

  • which results in:

    

The problem is that when I try to join the expressions by OR, the second expression does not match anything. For example:

(?|(\[)(?=\h*x\h*])|(?<=\[)\h*x\h*(]))|(?:(?<=\[)\h*(x)(?=\h*]))

Results in (i.e., see demo with extended flag enabled for clarity):

My intuition (i.e., perhaps not correct) is that there is no x left to match for the second expression because the x is matched in the first expression (i.e., group $0). For example, simplifying the second expression to (?:(x)) (i.e., see demo) will match just fine the x not comprised in brackets as shown below.

Therefore, I thought I should somehow reset the group $0 matches from the first expression. So I tried adding the \K meta escape to the first expression before (]), but that did not solve anything.

Also, as much as possible, I want to stick to the format (?|regex)|(?:regex)|... because I want to be able to extend the expression further with other groups. I am using Oniguruma regular expressions and the PCRE flavor. Do you have any ideas on how this can be accomplished?

P.S. Apologies if the title of the question is not entirely accurate.

Answer 1

The main issue is that x is already consumed with \h*x\h*(]) part in the first alternative, and \h*(x) in the second alternative cannot re-match what has already been consumed.

If you put the second alternation in the branch reset group inside a lookahead you can "free up" the x for the second alternative to catch it:

(?|
  (\[) (?= \h* x \h* ] ) | (?<= \[ )(?= \h* x \h* (])) # <--- here
)
|
(?:
  (?<= \[ ) \h* (x) (?= \h* ] )
)

See the regex demo. Pay attenation to the (?=\h*x\h*(])) part: it is now a positive lookahead that only checks for its pattern match immediately on the right, but does not put the matched text to the match value buffer and does not advance the regex index, so that subsequent subpatterns can try to match their patterns against this text.

To accommodate for further alternatives, make sure you use this technique: try to match as close to the start of string as possible, and only consume text that does not have to be rematched, else, use positive lookaheads with capturing groups in them.