From: Ara.T.Howard@...
Date: 2005-06-02T11:40:23+09:00
Subject: Re: Interesting discovery...

On Wed, 1 Jun 2005, Austin Ziegler wrote:

> On 5/31/05, Ara.T.Howard <Ara.T.Howard@noaa.gov> wrote:
>> On Wed, 1 Jun 2005, Austin Ziegler wrote:
>>> I had a user report a slowdown in PDF::Writer. I'm pretty certain I know
>>> where the problem is, but I found something very interesting. About 3% of
>>> the profiled performance of the manual generation (with Shugo Maeda's new
>>> profiler) -- and near 20,000,000 (yes twenty MILLION) calls -- are calls
>>> to Kernel.respond_to?.  =20 A little digging later, and it looks like 98%
>>> of these calls -- or more -- are in Marshal.dump. That's ... amazing.
>> wow. can you confirm with somthing like
>
> I've done quite a bit of confirmation. Much of this has to do with the
> unique nature of the object graph in PDF::Writer documents -- there's quite
> a bit of depth to the object graph.
>
> I was originally investigating Transaction::Simple as the culprit of my
> number of calls, but that didn't mesh with what I *did* find. My final test
> was to do:
>
>  def Kernel.respond_to?(*a)
>    puts caller[0]
>    old_respond_to(*a)
>  end
>
> After a few moments, I literally had tens of *thousands* of calls indicating
> that the caller was Marshal.dump.
>
> My guess -- and I haven't looked at the Marshal.dump code to confirm -- is
> that what's beeing looked for is a custom dumper, and that search is faster
> than assuming the dumper is there (because in most cases, it won't be, I
> don't think).

great.  thanks for the follow-up.  i run into things like that alot and always
wonder if calling a method and rescuing NoMethodError is slower or faster than
checking with respond_to?

   harp:~ > cat a.rb
   require 'benchmark'

   class C
     def c; end
     def rescue_test n
       n.times do
         begin
           self.a
         rescue NoMethodError
           begin
             self.b
           rescue NoMethodError
             self.c
           end
         end
       end
     end
     def respond_to_test n
       n.times do
         if respond_to? 'a'
           self.a
         elsif respond_to? 'b'
           self.b
         else
           self.c
         end
       end
     end
   end

   c = C::new
   n = 2 ** 16
   GC::disable

   Benchmark::bm do |b|
     %w(rescue_test respond_to_test).map{|m| b.report(m){c.send m,n}}
   end


   harp:~ > ruby a.rb
         user     system      total        real
   rescue_test  2.940000   0.320000   3.260000 (  3.264758)
   respond_to_test  0.160000   0.050000   0.210000 (  0.208268)


looks like respond_to? is indeed faster.

cheers.

-a
-- 
===============================================================================
| email :: ara [dot] t [dot] howard [at] noaa [dot] gov
| phone :: 303.497.6469
| My religion is very simple.  My religion is kindness.
| --Tenzin Gyatso
===============================================================================