Optional Feature: backtracking is sub-calls

[User4martin]

  IsMatching('Recursive', '^(?''A''(?:b[^b]+(?&A)?x+))', 'baaabaaaaaxx', [1,12,  1,12]);

https://regex101.com/r/g3pDSW/1

^(?'A'
  (?:
    b[^b]+
    (?&A)?  ## This sub-call will match ALL the "x", but when the outer match fails, it backtracks, and matches only ONE "x"
     x+
  )
)

TRegExpr fails to match, because it does not backtrack the sub-call.

[User4martin]

It appears that recursion (rather than sub-call) (?R) or (?0)` does not backtrack.

https://regex101.com/r/T2Il6J/1
(?:b[^bx]+((?R))?x+) on `baaabaaaxx'

If it did backtrack, then IMHO it should match the whole pattern, as it does if the + after the x+ is removed: https://regex101.com/r/uEG8kF/1

There is a mention of "recursion being atomic" here https://www.rexegg.com/regex-recursion.html

[User4martin]

Actually, changing the regex a bit ^(?'A'(?:b[^bx]+(?&A)?x+))
https://regex101.com/r/wwuP8H/1

• Then in PCRE >= 7.3 the sub-call does back-tracking.
• But in PCRE < 7.3 it does not. (As can be seen by removing the + of the x+

[User4martin]

I would suggest to keep that as a reminder for an optional feature.

[Alexey-T]

Martin, I am out of code knowledge here. so do the change as you wish. please test it.

[User4martin]

If it is ok, for now just keep the issue open. Needs more research how different engines handle it, and if there is documentation. Then maybe a property controlled behaviour: current or as pcre7.3.

Solving this, would need changes that would also remove the 20 levels of recursion.

I don't plan any immediate work on it... But might later go for it.