Ruby Dynamic Method Invocation Performance

Ruby is a very flexible and expressive language. A recent question posted by a Ruby newbie got me looking through my IRC logs for a discussion about the performance of various dynamic method invocation approaches, so I thought I’d share some performance results.

Ruby is currently my primary programming language, and I like it a lot; however, the conciseness (except for blocks) and performance of its higher order function capabilities could be improved. Scheme and Haskell beat it in this regard.

Consider three versions of a higher order function, caller, and associated ways of providing information to it to allow invoking a method dynamically given a pre-existing object and argument:

1
2
3
4
5
6

# 1. lambda
def caller fn
  fn.call(obj, arg)
end

caller(lambda {|obj,arg| obj.meth(arg) })

1
2
3
4
5
6

# 2. send
def caller sym
  obj.send(sym, arg)
end

caller(:meth)

1
2
3
4
5
6

# 3. bind/call
def caller meth
  meth.bind(obj).call(arg)
end

caller(MyClass.instance_method(:meth))

A simple benchmark of the three approaches with results follows. The benchmark uses an elide(n) method that has been added to String:

1
2
3
4
5
6
7
8
9

lam = lambda {|s,n| s.elide(n) }
sen = :elide
met = String.instance_method(:elide)

bm(5) do |x|
  x.report('lambda') { 100_000.times { lam.call('abcdef',4) } }
  x.report('send') { 100_000.times { 'abcdef'.send(sen,4) } }
  x.report('meth') { 100_000.times { met.bind('abc').call(4) } }
end

           user     system      total        real
lambda 1.460000   0.510000   1.970000 (  1.982085)
send   0.810000   0.260000   1.070000 (  1.075201)
meth   0.660000   0.250000   0.910000 (  0.913455)

# results with the 3 preparation lines w/in their respective loops

           user     system      total        real
lambda 1.900000   0.590000   2.490000 (  2.498358)
send   0.800000   0.250000   1.050000 (  1.068035)
meth   0.760000   0.260000   1.020000 (  1.021275)

I personally like the lambda approach, but Ruby does impose a penalty for using it.