From: Robert Dober <robert.dober@...>
Date: 2008-07-31T07:12:09+09:00
Subject: Re: Suggestions for improving a trivial tag parser

On Wed, Jul 30, 2008 at 10:43 PM, Gregory Brown
<gregory.t.brown@gmail.com> wrote:

> It does a double pass through the segments rather than a single pass,
Yes but these two passes are quite fast, see below.
> However, I think it'll be okay for my purposes (PDF inline styling),
> unless I missed some other concern Rolando had.
Well trying to be useful I checked for some larger texts, I omitted
the conditional #first at the end of the parsing method for clarity.

Turns out that split is still much better than scanning for a string
of a size over one megabyte, but it is not
really fast either:
539/39 > cat split.rb && ruby split.rb

require 'benchmark'

def split_select txt
  txt.split(%r{(</?[bi]>)}).select{|x| ! x.empty? }
end

def split_delete txt
  txt.split(%r{(</?[bi]>)}).delete_if{|x| x.empty? }
end

def use_scan txt
  r=[]
  txt.scan(%r{(.*?)(</?[ib]>)}) do | pr,po | r << pr unless pr.empty?;
r << po end
  r << $' unless $'.empty?
end

N = 400_000;
text = "<b>bold<i>italicbold<i>bold<b>normal" * N;

Benchmark.bmbm do | bm |
  bm.report("split_select") do
    split_select text
  end
  bm.report("split_delete") do
    split_delete text
  end
  bm.report("use_scan") do
    use_scan text
  end
end
Rehearsal ------------------------------------------------
split_select   6.844000   0.094000   6.938000 (  7.063000)
split_delete   7.687000   0.109000   7.796000 (  7.953000)
use_scan      20.063000   0.203000  20.266000 ( 20.109000)
-------------------------------------- total: 35.000000sec

                   user     system      total        real
split_select   6.344000   0.031000   6.375000 (  6.485000)
split_delete   6.500000   0.109000   6.609000 (  6.359000)
use_scan      16.625000   0.265000  16.890000 ( 16.906000)

HTH
Robert






-- 
http://ruby-smalltalk.blogspot.com/

There's no one thing that's true. It's all true.
--
Ernest Hemingway