RegEx - negative look behind failing

[iCode]

Why is the negative look behind group failing in this example?

$str = '<a href="http://example.com">link</a>' ; should NOT match
;$str = 'http://example.com' ; should match

$str = StringRegExpReplace($str, _
    '(?i)(?<!href=")(https?://[a-z0-9][a-z0-9.-]*[a-z0-9]\.[a-z]{2,4}|[a-z0-9][a-z0-9.-]*[a-z0-9]\.(?:biz|com|edu|eu|gov|info|mil|org|net|uk|us))(/\S*)*(?=\s|[[:punct:]]|$)', _
    '<a href="$1$2">$1$2</a>')

ConsoleWrite($str & @LF)

[jchd]

In PCRE lookbehind assertions must have a fixed length. Sometimes K may be used to overcome this limitation.

[iCode]

Hi jchd!

I understand that it must have a fixed length and it does... doesn't it?

(?<!href=")(https?

If i understand roughly how PCRE works, the 'matching point' is 'h' in the 'htttps?' - from there it should look at the previous 6 chars for 'href="'

I got done what i needed to do by adding a look-ahead group at the end and removing the neg. look-behind, but i still don't quite understand why the original failed to work, other than perhaps it has something to do with the anchor/matching point

$str = '<a href="http://example.com">link</a>' ; should NOT match
;$str = 'http://example.com' ; should match

$str = StringRegExpReplace($str, '(?i)(https?://[a-z0-9][a-z0-9.-]*[a-z0-9]\.[a-z]{2,4}|[a-z0-9][a-z0-9.-]*[a-z0-9]\.(?:biz|com|edu|eu|gov|info|mil|org|net|uk|us))(/\S*)*(?=\s|[[:punct:]]|$)(?!")', _
    '<a href="$1$2">$1$2</a>')

ConsoleWrite($str & @LF)

Edited December 19, 2013 by iCode

[jchd]

Ouch! Youre right of course. That is by far the most common reason for lookbehind to fail that I replied without looking enough.

What's your input?

[mikell]

When oversimplifying the regex the lookbehind seems to work

$str0 = '<a href="http://example.com">link</a>' ; should NOT match
$str = StringRegExpReplace($str0, '(?i)(?<!href=")(https?://\w+\.\w+)', '<a href="$1">$1</a>')
Msgbox(0,"", $str0 & @crlf & $str)

$str1 = 'http://example.com' ; should match
$str = StringRegExpReplace($str1, '(?i)(?<!href=")(https?://\w+\.\w+)', '<a href="$1">$1</a>')
Msgbox(0,"", $str1 & @crlf & $str)

Edit

Got it. It's a problem with the first OR ( "|" ) which allows the match for its 2nd altenative, exactly like if your regex was

$str = StringRegExpReplace($str, '(?i)([a-z0-9][a-z0-9.-]*[a-z0-9]\.(?:biz|com|edu|eu|gov|info|mil|org|net|uk|us))(/\S*)*(?=\s|[[:punct:]]|$)',  '<a href="$1$2">$1$2</a>')

This should be corrected by the use of grouping parentheses to define precisely the alternatives for this OR

$str = StringRegExpReplace($str, _
    '(?i)(?<!href=")(https?://(([a-z0-9][a-z0-9.-]*[a-z0-9]\.[a-z]{2,4})|([a-z0-9][a-z0-9.-]*[a-z0-9]\.(?:biz|com|edu|eu|gov|info|mil|org|net|uk|us))))(/\S*)*(?=\s|[[:punct:]]|$)', _
    '<a href="$1">$1</a>')

Edited December 19, 2013 by mikell

[iCode]

Hi mikell

I must need a brain enlargement because i don't get it

So when the engine comes upon the "OR", are you saying content outside of the group is considered?

Essentially it is grouped like...

(?<!a)(b|c)(d)*(?=e)

So nothing should match if 'b' OR 'c' is preceded by 'a' and 'b' OR 'c' is not followed by 'e'

I'm not understanding how changing that to the following changes anything...

(?<!a)((b)|(c))(d)*(?=e)

This PCRE stuff is hard

Edited December 19, 2013 by iCode

[jchd]

iCode,

Take a minute to look at the result of a StringRegExp (not *Replace) to see that this is part c of the example which gets selected. You seem to rely on the anchoring of  the a part, but there is no anchor in your pattern.

[mikell]

Be careful also about what is 'matched' and what is 'replaced'

$str = 'abcde'
$str = StringRegExpReplace($str, '(?<!a)(\w|\D)(\w)*(?=e)', "=$1=" &@crlf& "=$2=" &@crlf& "=$3=")
msgbox(0,"", $str)

$str = 'abcde'
$str = StringRegExpReplace($str, '(?<!a)((\w)|(\D))(\w)*(?=e)', "=$1=" &@crlf& "=$2=" &@crlf& "=$3=" &@crlf& "=$4=")
msgbox(0,"", $str)

[iCode]

Thanks gentlemen

@jchd - I began to suspect it was an anchor problem after reading more of the doc linked to in your sig

@mikell - that's not the actual replace expression i'm using, but yes, forgetting about captured groups has nearly bitten me before