Date: Wed, 03 Oct 2001 17:24:27 -0400
From: Benjamin Goldberg <goldbb2@earthlink.net>
Subject: Re: RegExp /g shall substitute more often
Message-Id: <3BBB820B.B9ABE17F@earthlink.net>

Markus Dehmann wrote:
> 
> The two solutions are nice, but because actually my problem is more
> complex they don't help.

Your fault for not describing the actual problem.

> Look-behind works only for a fixed length but my pattern that stands
> before B has in fact a variable length! This is my real problem that
> is more complex than just AAA:
> 
> Suppose you have this string:
> $_ = "bla-R carter-R bush-R clinton-R dunnowman-R clinton-R bla-R";
> 
> There is always a name "tagged" with 'R'. Now my rule says: Always tag
> clinton with a D if there a two names tagged with R before him. My
> output shall be:
>
> bla-R carter-R bush-R clinton-D dunnowman-R clinton-D bla-R

I see... so what you *want* is to transform Rafael Garcia-Suarez's
solution:
	s/(?<=A)A/B/g;
into:
	s/(?<=\w+-R \w+-R clinton-)R/D/g;
[This won't work because perl doesn't do variable length lookbehind]

> But s/(\w+-R \w+-R clinton-)R/$1D/; delivers, of course, only
> bla-R carter-R bush-R clinton-D dunnowman-R clinton-R bla-R

That doesn't work for the same reason why:
	s/(A)A/$1B/g;
doesn't work.

> The rule doesn't apply to the second clinton because he has not two
> names tagged with R before him anymore.

Erm, no, that's not the reason.  It's because in yours, it starts
looking for two Rs starting *after* the substitution... yours has to
match and capture those Rs.

> This:
> s/(?<=\w+-R \w+-R clinton-)R/$1D/;
> produces an error: "variable length lookbehind not implemented"

Right.  And even if it was implemented, it would be wrong, since you've
got that stupid $1 in the replacement.  If you understood what
lookaheads and lookbehinds did, you would know that you don't need it.

> What can I do?

On solution is to change it to a lookahead, as Jeff Pinyan did.

$_ = "bla-R carter-R bush-R clinton-R dunnowman-R clinton-R bla-R";
$_ = reverse $_;
s/R(?=-notnilc R-\w+ R-\w+)/D/g;
$_ = reverse $_;

Another is to not use s///, but do it in a more 'naive' way:
my @names = split;
for( my $i = $#names; $i >= 2; --$i ) {
    if( $names[$i] eq "clinton-R" &&
        $names[$i-1] =~ /-R$/ && $names[$i-2] =~ /-R$/ ) {
        $names[$i] = "clinton-D";
    }
}
print "@names\n";

Whether this is sufficiently fast or not, I don't know, but it's almost
certainly the least obfuscated version.


-- 
"I think not," said Descartes, and promptly disappeared.