ZERO_LENGTH match

sh1tn has asked for the wisdom of the Perl Monks concerning the following question:

Fellow Monks,

From perlre:

The lower-level loops are interrupted (that is, the loop is broken) wh
+en Perl detects that a repeated expression matched a zero-length subs
+tring. Thus:
m{ (?: NON_ZERO_LENGTH | ZERO_LENGTH )* }x;

   is made equivalent to

m{   (?: NON_ZERO_LENGTH )*
    |
     (?: ZERO_LENGTH )?
}x;
[download]

But in fact:

$_ = 'aaabbb';

/((?: a | c? )*)/x;       # 'c' matches once with zero-length
                          # which is not the same as
/(( a )* | ( c? )?)/x;    # where 'c' does not match at all
[download]

My question - is this behaviour expected? Thank you for the help!

Update: code sample given below - Re^4: ZERO_LENGTH match

Comment on ZERO_LENGTH match Select or Download Code

Replies are listed 'Best First'.
Re: ZERO_LENGTH match by hv (Prior) on Aug 01, 2005 at 16:15 UTC
I think you are misinterpreting the intent of the documentation: when it refers to 'ZERO_LENGTH' in the equivalence it is talking about actual zero length (ie matching an empty string) rather than potential zero length, such as matching `/c?/`. The implication is therefore that, for example: `/( (?: c? )* )/x` [download] is treated as equivalent to: `/( (?: c* \| ) )/x` [download] to break the loop. Similarly for a more complex zero-length expression such as a lookahead: `/( (?: a \| (?=c) ) )/x` [download] is treated as equivalent to: `/( (?: a \| (?=c) ) )/x` [download] Hope this helps, Hugo	[reply] [d/l] [select]
Re^2: ZERO_LENGTH match by Anonymous Monk on Aug 01, 2005 at 17:01 UTC
/( (?: a* \| (?=c) ) )/x This doesn't make any sense - a* cannot fail so you'll never end up in the position to try the second alternative. /( (?: a \| (?=c) ) )*/x This means - match as many 'a'-s as you can and when this becomes impossible try the second alternative - if it machtes (a zero-width) you'll face an infinite loop that you want to break - and you does this by allowing only one such zero-width match to happen.	[reply]
Re^3: ZERO_LENGTH match by hv (Prior) on Aug 02, 2005 at 01:15 UTC
Yes, I was thinking as I wrote my reply that it would make more sense to break it as: `/( (?: a+ \| (?=c) ) )/x` [download] But I didn't want to introduce unnecessary complications for the OP, and I wasn't entirely sure there was no deep reason I was missing as to why the docs show `/a*/` rather than `/a+/` for this. Hugo	[reply] [d/l] [select]
Re^3: ZERO_LENGTH match by Anonymous Monk on Aug 01, 2005 at 17:16 UTC
ok, the first paragraph is wrong - if you backtrack you _can_ use the second alternative but fact is that you'll try matching (?=c) at position 0 not at some position after the last 'a' in the string which is the case in /( (?: a \| (?=c) ) )*/x. The breaking of the infinite loop only forces (?=c) to be tried only once but doesn't redefine the position at which this happens.	[reply]
Re^4: ZERO_LENGTH match by fro (Novice) on Aug 02, 2005 at 08:09 UTC
Re: ZERO_LENGTH match by Anonymous Monk on Aug 01, 2005 at 15:26 UTC
What do you with the matches not being the same? `$_ = 'aaabbb'; while (/((?: a \| c? ))/xg) { printf "1: '%s' '%d'\n", $&, pos; } while (/(( a ) \| ( c? )?)/xg) { printf "2: '%s' '%d'\n", $&, pos; } __END__ 1: 'aaa' '3' 1: '' '3' 1: '' '4' 1: '' '5' 1: '' '6' 2: 'aaa' '3' 2: '' '3' 2: '' '4' 2: '' '5' 2: '' '6'` [download]	[reply] [d/l]
Re^2: ZERO_LENGTH match by sh1tn (Priest) on Aug 01, 2005 at 15:52 UTC
Without 'g' modifier: `1.>perl -e 'use re(debug); $_ = 'aaabbb'; /(( a \| c? ))/x' 2.>perl -e 'use re(debug); $_ = 'aaabbb'; /(( a ) \| ( c? )?)/x'` [download]	[reply] [d/l]