Forum: Ruby a regular expression problem

Announcement (2017-05-07): www.ruby-forum.com is now read-only since I unfortunately do not have the time to support and maintain the forum any more. Please see rubyonrails.org/community and ruby-lang.org/en/community for other Rails- und Ruby-related community platforms.
E5d5e03402b10c64cdaf44084343db80?d=identicon&s=25 unknown (Guest)
on 2007-03-05 05:10
(Received via mailing list)
hi,it seems that there's some problem with the regeular expression
processing in ruby,when I ran the code below, it happened that it loop
for ever and would not get out.
$value = "20053301831294544645645645634556354"
if $value =~ /\b(?:\d{1,3}[,\.]?)+\d{3}.{0,20}\b(?:pics|pictures|
images|photos|movies)/i
  puts "hit"
else
  puts "miss"
end
any idea? Thank u!
4feed660d3728526797edeb4f0467384?d=identicon&s=25 Bill Kelly (Guest)
on 2007-03-05 07:28
(Received via mailing list)
From: <cppasm@gmail.com>
> any idea? Thank u!
You can try this in irb, like:

>> value = "20053301831294544645645645634556354"
=> "20053301831294544645645645634556354"
>> value =~ /\b(?:\d{1,3}[,\.]?)+\d{3}.{0,20}\b(?:pics|pictures|images|photos|movies)/i
(takes a long time)

If you shorten value a bit:

>> value = "2005330183129454464564"
=> "2005330183129454464564"
>> value =~ /\b(?:\d{1,3}[,\.]?)+\d{3}.{0,20}\b(?:pics|pictures|images|photos|movies)/i
=> nil

You'll see that it takes a second or two, then finishes.

The problem is, the regexp involves nested variable-length matches,
so when no match is found, the engine keeps backtracking and trying
to find other possible matches, across all combinations of
(?:\d{1,3}[,\.]?)+\d{3}.{0,20}
...of which there are exponential possibilities within a string of
digits...

Hopefully, it may be possible to rewrite your regexp to fail more
quickly on bad data.If you could tell us more about the exact
specification of the data you're trying to match, people might
be able to help further.


Regards,

Bill
E5d5e03402b10c64cdaf44084343db80?d=identicon&s=25 unknown (Guest)
on 2007-03-05 07:55
(Received via mailing list)
On 3ÔÂ5ÈÕ, ÏÂÎç2ʱ27·Ö, "Bill Kelly" <b...@cts.com> wrote:
> > puts "miss"
>
> The problem is, the regexp involves nested variable-length matches,
>
> Regards,
>
> Bill

Thank you so much!
I'm trying to rewrite SpamAssassin in ruby just for fun :-). The
regexp mentioned is just a rule in SpamAssassin's rule files.It's used
to match against every line of a mail message.In fact I tested using
the value and regexp mentioned above in Java(JDK, oro and regexp of
Apache),perl,python and ruby.It finished very soon in perl and
python,while the same thing happened in Java(whichever library I
use,jdk,oro or regexp) as in ruby,so I guess maybe the underlying
matchings mechanism between them are quite different?
4feed660d3728526797edeb4f0467384?d=identicon&s=25 Bill Kelly (Guest)
on 2007-03-05 08:08
(Received via mailing list)
From: <cppasm@gmail.com>
On 3ÔÂ5ÈÕ, ÏÂÎç2ʱ27·Ö, "Bill Kelly" <b...@cts.com> wrote:
> > be able to help further.
>
> Thank you so much!
> I'm trying to rewrite SpamAssassin in ruby just for fun :-). The
> regexp mentioned is just a rule in SpamAssassin's rule files.It's used
> to match against every line of a mail message.In fact I tested using
> the value and regexp mentioned above in Java(JDK, oro and regexp of
> Apache),perl,python and ruby.It finished very soon in perl and
> python,while the same thing happened in Java(whichever library I
> use,jdk,oro or regexp) as in ruby,so I guess maybe the underlying
> matchings mechanism between them are quite different?

Yes, although not so much the underlying engine, but some
regex matchers are optimized to try to short-circuit certain
patterns to fail fast.  From what I've read, a lot of code has
been added to the Perl regexp matcher over the years to make it
smarter at failing fast on various nested variable-length
expressions.

Still, it is often possible to translate an expression into
something that will fail fast without requiring engine-specific
optimizations.

If you HAVE to parse SpamAssassin rule files without modification,
then, you will likely run into more situations like this where a
pattern that failed fast due to optimizations in the Perl matcher,
runs too slowly in a regexp engine without the equivalent shortcut
optimization.  :-(


Hope this helps,

Bill
E0d864d9677f3c1482a20152b7cac0e2?d=identicon&s=25 Robert Klemme (Guest)
on 2007-03-05 09:41
(Received via mailing list)
On 05.03.2007 08:08, Bill Kelly wrote:
>> > quickly on bad data.If you could tell us more about the exact
>> use,jdk,oro or regexp) as in ruby,so I guess maybe the underlying
> something that will fail fast without requiring engine-specific
> optimizations.
>
> If you HAVE to parse SpamAssassin rule files without modification,
> then, you will likely run into more situations like this where a
> pattern that failed fast due to optimizations in the Perl matcher,
> runs too slowly in a regexp engine without the equivalent shortcut
> optimization.  :-(

Hm, one quick manual optimization is to use

value =~ /\b(?:pics|pictures|images|photos|movies)/i &&
   value =~
/\b(?:\d{1,3}[,\.]?)+\d{3}.{0,20}\b(?:pics|pictures|images|photos|movies)/i

i.e. precheck the texts and use short circuit evaluation.  That at least
reduces the number of patterns dramatically where the exponential
runtime hits you.

Ah, I found a better one:

value =~
/\b\d{1,3}(?:[,.]?\d{3})*.{0,20}\b(?:pics|pictures|images|photos|movies)/i

That one's really fast on your value because it involves much less
backtracking.

I don't have my "Mastering Regular Expressions" handy but there you can
also find techniques for "loop unrolling" which should help in this
case.

Btw, you don't need to escape the dot in the character class.  It does
not have special meaning there.

Kind regards

  robert
E5d5e03402b10c64cdaf44084343db80?d=identicon&s=25 unknown (Guest)
on 2007-03-05 11:16
(Received via mailing list)
On 3ÔÂ5ÈÕ, ÏÂÎç4ʱ13·Ö, Robert Klemme <shortcut...@googlemail.com> wrote:
> >> > ..of which there are exponential possibilities within a string of
> >> to match against every line of a mail message.In fact I tested using
> > smarter at failing fast on various nested variable-length
> > optimization.  :-(
>
>
> Btw, you don't need to escape the dot in the character class.  It does
> not have special meaning there.
>
> Kind regards
>
>         robert

Thank you for your optimization,but I think there's a little problem
with it .For example,when the value is
'2.efgabc pics' it won't match the original one but will match the
optimized one.I think adding '\d{3}' after the asteroid would be ok.

Regards,
         cppasm
E0d864d9677f3c1482a20152b7cac0e2?d=identicon&s=25 Robert Klemme (Guest)
on 2007-03-05 12:11
(Received via mailing list)
On 05.03.2007 11:13, cppasm@gmail.com wrote:
>>>>> (?:\d{1,3}[,\.]?)+\d{3}.{0,20}
>>>> the value and regexp mentioned above in Java(JDK, oro and regexp of
>>> Still, it is often possible to translate an expression into
>>    value =~
>>
>>
>>         robert
>
> Thank you for your optimization,but I think there's a little problem
> with it .For example,when the value is
> '2.efgabc pics' it won't match the original one but will match the
> optimized one.I think adding '\d{3}' after the asteroid would be ok.

What asteroid?  Is an impact imminent?  Do we have to duck and cover?

SCNR

Seriously: I thought with your initial portion of the RX you wanted to
match something like "1,000,000", "1,000" and "10" (plus "." as
separator as well as no separator) although I realize that your original
does also match "1,0,0,00,0" etc.  Your suggested "fix" won't make the
two RX more similar with regard to matching because mine always matches
exactly three digits between a separator.  Since I don't know what you
are actually trying to match I'm out of suggestions at the moment.

Kind regards

  robert
This topic is locked and can not be replied to.