Why {} is much faster than Hash.new ?

[Iñaki Baz Castillo]

Hi, a simple benchmark:

Benchmark.realtime { 1000000.times { @m =3D Hash.new } }
=3D> 0.6552023887634277

Benchmark.realtime { 1000000.times { @m =3D {} } }
=3D> 0.17668819427490234

I would like to know why {} is so much faster. I can understand that
Hash.initialize method must inspect is an argument is given and so, is
just due to it?

Thanks.


--=20
I=C3=B1aki Baz Castillo
<[email protected]>

 

[Robert Klemme]

Hi, a simple benchmark:

Benchmark.realtime { 1000000.times { @m =3D Hash.new } }
=3D> 0.6552023887634277

Benchmark.realtime { 1000000.times { @m =3D {} } }
=3D> 0.17668819427490234

I would like to know why {} is so much faster. I can understand that
Hash.initialize method must inspect is an argument is given and so, is
just due to it?

I can imagine that's about right. Another benchmark:

13:10:40 Temp$ ruby19 bm.rb
Rehearsal --------------------------------------------------
Hash.new 1.297000 0.000000 1.297000 ( 1.293000)
Hash 0.125000 0.000000 0.125000 ( 0.127000)
Hash.nil? 0.187000 0.000000 0.187000 ( 0.186000)
{} 0.188000 0.000000 0.188000 ( 0.191000)
----------------------------------------- total: 1.797000sec

user system total real
Hash.new 1.296000 0.000000 1.296000 ( 1.292000)
Hash 0.125000 0.000000 0.125000 ( 0.127000)
Hash.nil? 0.188000 0.000000 0.188000 ( 0.185000)
{} 0.187000 0.000000 0.187000 ( 0.191000)
13:10:50 Temp$ cat bm.rb
require 'benchmark'

R =3D 1_000_000

Benchmark.bmbm 15 do |b|
b.report "Hash.new" do
R.times { Hash.new }
end

b.report "Hash" do
R.times { Hash }
end

b.report "Hash.nil?" do
R.times { Hash.nil? }
end

b.report "{}" do
R.times { {} }
end
end
13:10:52 Temp$

Cheers

robert

--=20
remember.guy do |as, often| as.you_can - without end
http://blog.rubybestpractices.com/

 

[Jonas Pfenniger (zimbatm)]

I think it's the cost of the method call with options parsing.

Benchmark.realtime do 1_000_000.times do @m =3D Hash.new end end
=3D> 1.1976041793823242

def blank_method; end
Benchmark.realtime do 1_000_000.times do @m =3D blank_method end end
=3D> 0.1965808868408203

Benchmark.realtime do 1_000_000.times do @m =3D {} end end
=3D> 0.2893531322479248

Benchmark.realtime do 1_000_000.times do @m =3D Hash.allocate end end
=3D> 0.3925323486328125

You see that Hash.allocate's time is about the same as ({} +
blank_method). The rest comes from option parsing I guess.

 

[Iñaki Baz Castillo]

I think it's the cost of the method call with options parsing.

Benchmark.realtime do 1_000_000.times do @m =3D Hash.new end end
=C2=A0=3D> 1.1976041793823242

def blank_method; end
Benchmark.realtime do 1_000_000.times do @m =3D blank_method end end
=C2=A0=3D> 0.1965808868408203

Benchmark.realtime do 1_000_000.times do @m =3D {} end end
=C2=A0=3D> 0.2893531322479248

Benchmark.realtime do 1_000_000.times do @m =3D Hash.allocate end end
=C2=A0=3D> 0.3925323486328125

You see that Hash.allocate's time is about the same as ({} +
blank_method). The rest comes from option parsing I guess.

Good point

Thanks a lot.




--=20
I=C3=B1aki Baz Castillo
<[email protected]>

 

[Ryan Davis]

Benchmark.realtime { 1000000.times { @m =3D Hash.new } }
=3D> 0.6552023887634277
=20
Benchmark.realtime { 1000000.times { @m =3D {} } }
=3D> 0.17668819427490234

% echo "{}" | parse_tree_show
shash)

% echo "Hash.new" | parse_tree_show
scall, sconst, :Hash), :new, sarglist))

 

[Roger Pack]

I think it's the cost of the method call with options parsing.

You'd think they'd be able to optimize that , if that's the case...

-r