bug#18577: Regexp I-search: [(error Stack overflow in regexp matcher)]

[Stefan Kangas <stefan@marxist.se>]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

>>> Is this a defect in my regexp or in the regexp engine?
>> It is fundamental to the way regexp matching works.
>
> To clarify: it is fundamental to the way *our* regexp engine works.
>
> As long as the regexp doesn't use backrefs, it can be matched
> efficiently, without backtracking.  Of course using \(..\) (as opposed
> to using \(?:..\)) can also make the problem harder since the various
> different (but largely equivalent) ways to match might need to be
> distinguishable via match-data.
>
> But even tho your regexp doesn't use backrefs, and even if you replace
> all \(..\) with \(?:..\), your regexp will still cause problems because
> our regexp engine does not try to optimize these kinds of cases.
>
> So you have to do it by hand.
>
>>> If the former, how could I rewrite the regexp so that it would not hit
>>> these problems?
>
> Maybe something like:
>
> /\*\(<insidecomment>\)*\*+/
>
> where <insidecomment> is something like
>
>    [^'*]\|\*+\([^/'*]\|'<afterquote>\)\|'<afterquote>
>
> where <afterquote> is something like
>
>    \([^'*]\|\*+[^/'*]\)*'
>
> Tho this will still push a backtrack point for every character.
> Maybe better would be something like
>
> /\*[^'*]*\(<insidecomment>\)*\*+/
>
> where <insidecomment> is something like
>
>    \(\*+[^/'*]\|\**'<afterquote>\)[^'*]*
>
> where <afterquote> is still something like
>
>    \([^'*]\|\*+[^/'*]\)*'
>
> so that we should only push a backtrace point when we see a * or a ' in
> the comment.

Should we do anything about this, like document it in etc/PROBLEMS, or
should this bug just be closed?