From: Eric Mahurin <eric.mahurin@...>
Date: 2008-02-05T00:57:09+09:00
Subject: Re: Parsing JSON (#155)

------=_Part_7248_712601.1202140636600
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

On Feb 4, 2008 4:29 AM, Paolo Bonzini <bonzini@gnu.org> wrote:

> On Feb 4, 11:01 am, Paolo Bonzini <bonz...@gnu.org> wrote:
> > Here is my try using regexes.  I use the "copy-on-write trick" from
> > the suffix tree quiz: the regex is always anchored to the beginning of
> > the string using \A, and the matched text is discarded using
> > post_match.  In some places where I don't want to discard I use (?
> > =...).
> >
> > Using Eric's benchmark I get 36kb/sec, but I haven't benchmarked any
> > other solution.
> >
> > http://pastie.caboo.se/147201
>
> For what it's worth, I get 144kb/sec from tho_mica_l's solution, after
> converting it to Ruby 1.8 like this:
>
>  class JSONParser
>
>    RXE = /
>        \[|\]|
>        \{|\}|
>        (:)|
>        (,\s*[}\]])|
>        ,|
>        ("(?>[^"\\]+|\\(?:u[0-9a-fA-F]{4}|[bfnrt"\/\\]))*")|
>        -?(?=\d)(?>0|[1-9]\d*)(?>\.\d+)?(?>[Ee][+-]?\d+)?(?=\D|$)|
>        true|
>        false|
>        (null)|
>        (?>[[:space:][:cntrl:]]+)|
>        ((?>.+))
>        /xmu
>
>    def parse(json)
>        ruby = json.gsub(RXE) do |t|
>            if !$5.nil?||!$2.nil?       then invalid($5.nil? ? $2 :
> $5)
>            elsif !$4.nil?              then 'nil'
>            elsif !$1.nil?              then '=>'
>            elsif !$3.nil?              then $3.gsub(/#/, '\\\\#')
>            else
>                t
>            end
>        end
>        begin
>            return eval(ruby)
>        rescue Exception => e
>            invalid(json)
>        end
>    end
>
>    def invalid(string)
>        raise RuntimeError, 'Invalid JSON: %s' % string
>    end
>
> end
>


Thank you.  Now we can compare apples to apples  with this on 1.8.6.  I are
the results with these two new benchmarks on my machine:

ch/s    author/gem
----    ----------
-       Pawel Radecki (RE, mismatch)
3214    Justin Ethier (RE lexer + ruby eval, fixed number parsing)
4054    Eric Mahurin (Grammar0, no lexer, no parser generation)
4078    Eric I (Treetop, unicode broken)
6534    oksteev (Treetop, mismatches in benchmark)
8313    Clifford Heath (Treetop, had to remove handling of "\/")
54586   Eric Mahurin (Grammar, no lexer, v0.5)
137989  Paolo Bonzini (RE)
166041  Thomas Link (RE lexer + ruby eval, ruby 1.9 results)
220289  json
223486  Eric Mahurin (Grammar, no lexer, unreleased)
224823  fjson (uses C extensions)
333368  Thomas Link & Paolo Bonzini (RE + eval, unicode broken)
553081  Eric Mahurin (Grammar, no lexer, unreleased, w/ ruby2cext)
1522250 json (w/ C extensions)


Looks like the above solution is fastest pure-ruby solution out there.  It's
interesting that the performance is better in 1.8.6 than 1.9.  The "eval"
solution is a nice trick as it uses ruby's C parser to act as the parser.
But, this isn't viable for most other languages you might parse.  Maybe fine
for this JSON case though.  The "eval" does seem a bit dangerous though.

------=_Part_7248_712601.1202140636600--