Re: Perl Complains of Nested Quantifiers

[AnomalousMonk]

in reply to Perl Complains of Nested Quantifiers

The 'atomic' extended pattern  "(?>pattern)" (available in 5.8) will give you possessiveness around the entire pattern or any sub-pattern of your choosing. See Extended Patterns in perlre. (The following example doesn't really illustrate atomic/possessive matching; for a good discussion and pertinent examples, see the docs.)

>perl -wMstrict -le
"print qq{perl ver. $]};
 my $rx = qr{ ((?> \d+)) }xms;
 'abc1234def' =~ $rx;
 print qq{'$1'};
"
perl ver. 5.008009
'1234'

[Kenosis]

The Extended Patterns documentation mentions that "(?>pattern) does not disable backtracking altogether once it has matched." A concern was expressed about "catastrophic backtracking." I'm curious about the nature of the 'catastrophe,' and whether your solution would satisfactorily avert it.

[demerphq]

Well, what this means is that a pattern like /foo(?:baz)++baz/ will never match. If you had a string like "foobazbazbaz" it would fail because (?:baz)++ will "gobble up" all of the "baz" in the string, and then refuse to give anything back. What it means by "does not disable backtracking" is that "foobazbazbaz foobazbazbaz" will *attempt* the /(?:baz)++/ twice, once after each "foo". Neither will match of course, and if the RE was _really_ smart it would know it could never match and wouldn't try at all, but it isn't. :-)

[AnomalousMonk]

A concern was expressed about "catastrophic backtracking." I'm curious about ... whether your solution would satisfactorily avert it.

It would not. AFAIU, possessive quantifiers like  ++ *+ ?+ {n,m}+ are just special, limited cases of the general  (?>...) atomic grouping. See example below.

>perl -wMstrict -le
"my $s = 'foobazbazbaz foobazbazbaz';
 print 'match 1'
   if $s =~ m{ foo (?> (?:baz)+) baz }xms;
 print 'match 2'
   if $s =~ m{ foo (?> .* baz) baz }xms;
"

(no output)