nospam-abuse at bloodgate
Nov 22, 2006, 11:59 AM
Post #12 of 32
(918 views)
Permalink
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Moin,
offlist because I accidentily replied to myself, then resent it but forgot
to add p5p. So, repost to list:
On Wednesday 22 November 2006 19:23, you wrote:
> On 11/22/06, Tels <nospam-abuse[at]bloodgate.com> wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > Moin,
> > On Wednesday 22 November 2006 17:55, you wrote:
> > > On 11/22/06, Tels <nospam-abuse[at]bloodgate.com> wrote:
> > > > * I want to spur Yves into looking into why qr// is slower than it
> > > > should, the slowdown makes assembling regexps from dynamically
> > > > created parts much less usefull than it could be.
> > >
> > > Its because there is no easy way to merge a compiled regex into one
> > > that is being compiled. Its not impossible its just not easy. I dont
> > > see this happening in time for 5.10 if we want 5.10 coming out around
> >
> > Okay on the time frame. Just some small followup question for me to
> > understand:
> >
> > As far as I understood Dave's message,
> >
> > $qr_foo = qr/foo/; $qr_foobar = qr/bar$qr_foo/;
> >
> > results in $qr_foobar merely containing:
> >
> > (?-xism:bar(?-xism:foo))
> >
> > as a string. Why aren't the "inner" (?-xism:...) cut away when the
> > content of the xism doesn't contain any variables or is "simple" or
> > whatever? (I know that probably sounds much easier than it is :)
>
> They have to be wrapped in (?: ) so they are self contained.
>
> Think about:
>
> $p=qr/a|b|c/;
> $x=qr/foo$p/;
>
> you dont want $x to be: fooa|b|c. You want it to be (?:foo(?:a|b|c)).
Good point. (That was exactly the point I was waiting for since I couldn't
come up with an example).
So
qr/foo/ => foo
qr/a|b/ => (?:a|b)
qr/(foo)/ => (foo)
etc.
Maybe a simple heuristic is that if it contans only [a-z0-9\[\]],
interpolate as it is, otherwise interpolate as "(?:" $qr .")", unless it
already has "(" at the start. Hm, but the regexp engine would surely know
much better which case could be handled as which.
> > I want the result to be:
> >
> > (?-xism:barfoo)
> >
> > and actually not a bunch of pointers to the little parts.
>
> That would be nice. I dont know if its that easy to do. We have some
> of the information on hand to make it happen tho.
>
> > The reason is
> > because it can become unwieldingly large and complicated:
> >
> > perl -le '$f = qr/f/; for (0..3) { $f = qr/$f$f/; } print
> > $f,"\n"'
> >
> > I already have written a parser that nestes about 3 layers of multiple
> > qr// objects and a simple debug printout of the resulting regexp start
> > to spam the screen with xisms :-)
>
> Yeah, well you could post process the regex with a regex. :-) remove
> all the unnecessary wrappers. Or you could use something like the
> tools in the latest version of the re module that allow you to get
> access to the pattern unwrapped.
Well, if I had some free time lying around here somewhere :/
> > Plus, everytime you use that string to match
> > something, the regexp compiler seems to goes crazy.
>
> IN what way?
Parsing & processing all these (?-xism:) every match takes time :) Not much,
but needless time anyway.
> > Maybe someone should write RegExp::Optimize::qr? :-D
> Heh.
> BTW, why offlist?
>
> Cheers,
> Yves
- --
Signed on Wed Nov 22 20:54:46 2006 with key 0x93B84C15.
Visit my photo gallery at http://bloodgate.com/photos/
PGP key on http://bloodgate.com/tels.asc or per email.
This email violates U.S. patent #5,978,791 <http://tinyurl.com/5t6ft>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
iQEVAwUBRWSsEHcLPEOTuEwVAQLQyQf9HPq78BzY68q5xLdsF5tS499q0PWk+dB2
EW4k9DuFpWk59h86igOmfsDjJH3Fl/GPc6WiyIVmjEPr0GAB7OSHlCa6cIul4BGR
CF+0mCzmEPODGim9TCLhprtb19l7qfmiG4PT6zsP+qpeF3yVHianjjdHWfLfLRp9
MjO/96EulOI3/GRBfFdAA01SAAJDanqwbz9TPVIWUfQltq/9i8G10cTU3fi8XnMY
Q572lOfjM6TYLfAyAsTE1XcHJ5gfSOJGazgMKtt84JFViocIZ2WdnWXwfCp7GYAp
XoeI79qkJ4nEjItJUgDPmWovZcJz2TMQnu5PY4xKDhw4AmMeLlXxpg==
=gf83
-----END PGP SIGNATURE-----
|
1
