Message-ID: <008901c5ac9b$31bb8b50$0600a8c0@Wilbur>
From: "Julian Scarfe" <julian@avbrief.com>
To: <pgsql-sql@postgresql.org>
References: <001a01c5ac84$496fc750$1102a8c0@dream>
Subject: Negative lookbehind assertions in regexs
Date: Mon, 29 Aug 2005 14:11:37 +0100
MIME-Version: 1.0
Content-Type: text/plain; format=flowed; charset="iso-8859-1";
	reply-type=response
Content-Transfer-Encoding: 7bit

I'd like a regex that matches 'CD' but not 'ABCD' in any part of the regex.

In Perl I'd use a negative lookbehind assertion (?<!AB)CD to do the job:

$ cat lb.pl
print "CD: ",   'CD'   =~ /(?<!AB)CD/, "\n";
print "XYCD: ", 'XYCD' =~ /(?<!AB)CD/, "\n";
print "ABCD: ", 'ABCD' =~ /(?<!AB)CD/, "\n";

$ perl lb.pl
CD: 1
XYCD: 1
ABCD:

But Postgresql (7.4) doesn't seem to like that:

# select 'ABCD' ~ '(?<!AB)CD';
ERROR:  invalid regular expression: quantifier operand invalid

Is there a workaround that allows me to do this as a single regex?

I know I could and together a ~ and !~ like this

# select ('CD' ~ 'CD') and ('CD' !~ 'ABCD');
 ?column?
----------
 t
(1 row)

# select ('ABCD' ~ 'CD') and ('ABCD' !~ 'ABCD');
 ?column?
----------
 f
(1 row)

but changing the SQL would break the existing paradigm.

TIA

Julian