From: Robert Klemme <bob.news@...>
Date: 2005-09-03T00:36:25+09:00
Subject: Re: Ruby-specific performance heuristics?

------=_NextPart_000_00D4_01C5AFE4.5216D700
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

Ara.T.Howard wrote:
> On Sat, 3 Sep 2005, Robert Klemme wrote:
>
>>> here's some things i think of:
>>
>> <lots of good stuff deleted>
>>
>>>    - anchor every single regular expression : performance degrades
>>>      exponentially otherwise
>>
>> That's not possible - especially with String#scan.  Sometimes you
>> want all occurrences of something.
>
> true.  however there nearly always __some__ anchor one can use:

Ah, yes, I overlooked that one.

>    harp:~ > irb
>    irb(main):001:0> "foo bar".scan %r/ \b\w+ /x
>    => ["foo", "bar"]

But does this really make such a big difference?  Or does this depend on
input sequence and rx?  My quick hack test didn't show significant
differences.

17:29:48 [ruby]: ruby bm-rx-anchor.rb
Rehearsal -------------------------------------------------------
no anchor             4.469000   0.000000   4.469000 (  4.465000)
1 anchor              4.515000   0.000000   4.515000 (  4.525000)
2 anchors             4.500000   0.000000   4.500000 (  4.520000)
--------------------------------------------- total: 13.484000sec

                          user     system      total        real
no anchor             4.500000   0.000000   4.500000 (  4.509000)
1 anchor              4.484000   0.000000   4.484000 (  4.494000)
2 anchors             4.531000   0.000000   4.531000 (  4.531000)


> a better thing would be to say "use an anchor unless you know exactly
> why you shouldn't."

Sounds good.

>>>    - compile regular expressions only once unless they require
>>>      variable substitution
>>
>> I guess you're talking about flag /o or storage of a regexp object
>> somewhere which is really only needed if the regexp contains
>> interpolation that you want to do only once.  Otherwise a regexp
>> *is* compiled only once and inline regexps are roughly as performant
>> as a stored regexp and might be easier to read (not always).
>
> you are quite right - guy pointed this out to me once before - for
> some reason my brain hasn't retained it though ;-)

Well, repetition helps. :-))

Kind regards

    robert

------=_NextPart_000_00D4_01C5AFE4.5216D700
Content-Type: application/octet-stream;
	name="bm-rx-anchor.rb"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
	filename="bm-rx-anchor.rb"

=0A=
require 'benchmark'=0A=
=0A=
s =3D "foo bar;baz.snorkel"=0A=
REP =3D 100000=0A=
=0A=
Benchmark.bmbm(20) do |bm|=0A=
=0A=
  bm.report "no anchor" do=0A=
    REP.times { s.scan(/\w+/) {|m|} }=0A=
  end=0A=
=0A=
  bm.report "1 anchor" do=0A=
    REP.times { s.scan(/\b\w+/) {|m|} }=0A=
  end=0A=
=0A=
  bm.report "2 anchors" do=0A=
    REP.times { s.scan(/\b\w+\b/) {|m|} }=0A=
  end=0A=
=0A=
end=0A=

------=_NextPart_000_00D4_01C5AFE4.5216D700--