On Wed, Sep 10, 2008 at 7:24 PM, Iñáki Baz Castillo <ibc / aliax.net> wrote:
> So using "instance_eval" is ~  40% slower than running directly the code.

well, we can try again :)

botp@jedi-hopeful:~$ cat test.rb
require 'benchmark'
n=100_000
Benchmark.bmbm do |x|
  x.report("normal") { n.times { a=nil;a=123; a2=a*2; a3=a*3 }}
  x.report("string") { n.times {instance_eval 'a=nil;a=123; a2=a*2; a3=a*3' }}
  x.report("block") { n.times {instance_eval {a=nil;a=123; a2=a*2; a3=a*3} }}
  x.report("block-broken") { n.times {a=nil;instance_eval{a=123};
                                      instance_eval{ a2=a*2};
                                      instance_eval{a3=a*3}
                                      }
                            }
end




botp@jedi-hopeful:~$ ruby test.rb
Rehearsal ------------------------------------------------
normal         0.140000   0.000000   0.140000 (  0.247775)
string         2.420000   0.010000   2.430000 (  3.438391)
block          0.250000   0.000000   0.250000 (  0.375656)
block-broken   0.470000   0.010000   0.480000 (  0.698093)
--------------------------------------- total: 3.300000sec

                   user     system      total        real
normal         0.120000   0.000000   0.120000 (  0.216789)
string         2.480000   0.020000   2.500000 (  3.291342)
block          0.300000   0.000000   0.300000 (  0.423327)
block-broken   0.490000   0.000000   0.490000 (  0.703046)
botp@jedi-hopeful:~$


slower but not slow (in my own mileage :)
or optimistically, direct code is faster than a code block