Front page | perl.perl5.porters |
Postings from September 1999
On Tue, Sep 07, 1999 at 07:34:38PM -0700, horos wrote:
> > $brackets = qr{
> > {
> > (?:
> > (?> [^{}]*) |
> > (?p{ $qr })
> > )*
> > }
> > }x;
> : I do not follow this \G stuff. Where do you want to put it?
>
> Well, take the above regular expression. If there was a 'no backtracking'
> operator,
??? There is 'no backtracking' operator, and you were using it.
> you could match the following relatively quickly:
>
> "{this{ fails { very } slowly } without no backtrack op" =~ m"\G$brackets";
It is slow only because you asked it to be slow. It becomes twice as
quick if you use (?> [^{}]+ ) instead of (?> [^{}]* ). It becomes six
times as quick if you use your idea of
$brackets = qr{
{
(?:
(?> [^{}]+ | (?p{ $qr }) )
)*
}
}x;
[ I will investigate why one cannot drop these (?:). ]
And probably you want m{\G(?p{ $brackets })}, which would avoid
recompile of the REx.
> Come to think of it, perhaps 'all or nothing' is a better term for the operator
> - ie, if something fails to match, the engine rewinds all the way back
> to the beginning, bumps up its position by 1 from the beginning point
> and then starts again.
Why this cannot be implemented by nested (?>) ?
> So matching the above without the \G modifier would look like:
>
> tries '{.... (goes to end without success)
> tries 't' (fails immediately)
> tries 'h' (fails immediately)
> tries 'i' (fails immediately)
> tries 's' (fails immediately)
> tries '{........ slowly } (succeeds.)
Why do you think it is not doing it like this?
Ilya