Rails core - Jun 2006 - 22-30% faster and added to_xml functionality

I''ve been working on improving the performance and functionality of
to_xml.

While putting together this patch, I discovered that the recent upgrade of
Builder::XMLMarkup to 2.0 slows down to_xml by 80%, since it now encodes
attributes and seems to do a better job, that''s more expensive, of
encoding all
values.

http://dev.rubyonrails.org/changeset/4260

Could somebody take a look at this patch and give some feedback.

http://dev.rubyonrails.org/ticket/4989

The major benefits of this patch are (more improvements are listed in the
ticket).

1) Allow ActiveRecord::Base subclasses to override to_xml.  Currently, all AR
instances are turned into a Hash and then have to_xml run on the hash, which
prevents subclassing of AR instances through an :include.

2) Speeds up to_xml generation by 22-30% depending upon the model.  This is done
by using the fact that we know the SQL column type and this allows us to skip
the XML encoding if the type doesn''t need encoding, such as booleans,
floats,
integers.  I''ve attached a small script you can use to time the to_xml
before
and after the patch.

3) Allows the :include option to be arbitrary deep, previously, only one level
of :include worked.

4) Binary columns are Base64 encoded and have an encoding="base64"
attribute
added to it. It turns out that it is 80 times faster to do builder <<
Base64.encode(value) then builder.tag!(tag, value).

Outstanding issues are

1) The XML output is not at nice looking with the optimization of using

   builder.tag!(tag) do
     builder << value.to_s
   end

instead of builder.tag!(tag, value), but that seems like a small price to pay.

   <?xml version="1.0" encoding="UTF-8"?>
   <node>
     <node-id type="integer">
   1234  </node-id>
   </node>

Regards,
Blair

-- 
Blair Zajac, Ph.D.
<blair@orcaware.com>
Subversion training, consulting and support
http://www.orcaware.com/svn/




#!/usr/local/bin/ruby

# This assumes that the script is run from the top of the Ruby on
# Rails project.
print "Loading Ruby on Rails..."
$stdout.flush
require ''config/environment''
print "\n"

require ''benchmark''
include Benchmark

N = 1000

subclasses = ActiveRecord::Base.send(:subclasses)
subclasses = subclasses.reject { |s|
''CGI::Session::ActiveRecordStore::Session''
== s.to_s }
subclasses = subclasses.sort_by { |s| s.to_s }

max_length = subclasses.collect { |s| s.to_s.length }.max


bm(max_length + 4) do |x|
   subclasses.each do |subclass|
     obj = subclass.find(:first, :order => subclass.primary_key)
     s = sprintf(''%s %2d %s'',
                 subclass,
                 subclass.column_names.length,
                 '' ''*(max_length - subclass.to_s.length))
     x.report(s) { N.times { obj.to_xml } }
   end
end

On 2-jun-2006, at 23:44, Blair Zajac wrote:
> o_xml generation by 22-30% depending upon the model.  This is done
> by using the fact that we know the SQL column type and this allows  
> us to skip
> the XML encoding if the type doesn''t need encoding, such as  
> booleans, floats,
> integers.  I''ve attached a small script you can use to time the  
> to_xml before
> and after the patch.
>
> 3) Allows the :include option to be arbitrary deep, previously,  
> only one level
> of :include worked.
>
> 4) Binary columns are Base64 encoded and have an
encoding="base64"
> attribute
> added to it. It turns out that it is 80 times faster to do builder <<
> Base64.encode(value) then builder.tag!(tag, value).
>
> Outstanding issues are
>
> 1) The XML output is not at nice looking with the optimization of  
> using
>
>   builder.tag!(tag) do
>     builder << value.to_s
>   end
The new Builder "escape all" approach is a bit absurd, especially  
considering it''s doing packs/unpacks to secure all Unicode values  
outside of ASCII. I tried to devise a patch to that but the stuff is  
buried too deep and Jim didn''t seem keen on the idea.

--
Julian ''Julik'' Tarkhanov
please send all personal mail to
me at julik.nl

> I''ve been working on improving the performance and functionality
of to_xml.
Good stuff, Blair. I''ve taken the liberty to massively refactor the
implementation, though. Having a method going on 100+ lines of code
was a sure tell sign that it needed some love. And it never was a good
fit for base.rb anyway, so now it all sits in xml_serialization.rb and
is mixed in.
-- 
David Heinemeier Hansson
http://www.loudthinking.com -- Broadcasting Brain
http://www.basecamphq.com   -- Online project management
http://www.backpackit.com   -- Personal information manager
http://www.rubyonrails.com  -- Web-application framework

David Heinemeier Hansson wrote:
>> I''ve been working on improving the performance and
functionality of
>> to_xml.
> 
> 
> Good stuff, Blair. I''ve taken the liberty to massively refactor
the
> implementation, though. Having a method going on 100+ lines of code
> was a sure tell sign that it needed some love. And it never was a good
> fit for base.rb anyway, so now it all sits in xml_serialization.rb and
> is mixed in.
Thanks David.

Regards,
Blair