Logged In: YES 
user_id=7887

Greg, what I meant is that I *think* that every SRE_MATCH()
recursion with *failing* possibilities should be protected
by SAVE/RESTORE(), as it can potentially change
lastmark/lastindex before returning, depending on what is
being executed.

Do you belive the first part of MAX_UNTIL shouldn't be
protected, even though it could match a subpattern including
marks and fail? If so, can you explain?

OTOH, I agree with you that MAX_UNTIL and MIN_UNTIL are not
coherent. I'll fix that in the safe side for now, including
mark protection in both. If we end up discovering it was not
necessary, we simply drop it.