From: Robert Klemme <bob.news@...>
Date: 2004-05-12T16:48:53+09:00
Subject: Re: Strange regexp behaviour in gsub


"Kristof Bastiaensen" <kristof@vleeuwen.org> schrieb im Newsbeitrag
news:pan.2004.05.12.01.24.42.706400@vleeuwen.org...
> On Wed, 12 May 2004 02:12:19 +0200, Florian Gross wrote:
>
> > Kristof Bastiaensen wrote:
> >
> >> Yes, that would clarify the situation, but is it the correct
> >> behaviour?  I would think that (?!a)a doesn't mean the same
> >> character, but consecutive ones.  Because it doesn't consume
> >> the character, it effectively is the character 'before' the
> >> match (if any).  The other behaviour wouldn't make sense,
> >> because (?!a)b is then exactly the same as b.
> >
> > I think that it's the intended behavior. Just use /(?!a).b/ if you
want
> > to consume the character.
>
> Hi,
> You are right, I looked it up in the manual, and there it was.  The
> term zero-width-look-ahead pretty much says it all.  I must have
> gotten the definition all wrong.
>
> >
> > Thinking about this, it is indeed possible to implement fixed-width
> > look-behind -- interesting.
>
> I was thinking more about something like variable-width look-between :-)
> Meaning for example a(?^\w+)b would match any a(.*)b if (.*) is
> not equal to (\w+)

IMHO that's not generally possible with regular expressions.  You'll
always have to define positively things that should match.  Exclusion
character classes are just a means of convenience but this does not extend
to complete (sub) expressions.

For example: to match a.*a where the part in the middle does not contain
only b's (i.e. matches /b+/) you can do:

/a(.*[^b].*)?a/

irb(main):004:0> rx=/a(.*[^b].*)?a/
=> /a(.*[^b].*)?a/
irb(main):005:0> rx === "aa"
=> true
irb(main):006:0> rx === "aba"
=> false
irb(main):007:0> rx === "acba"
=> true

Regards

    robert