hayeah-Rant

h1. Imperative Random Data Generator and Quickcheck

You can use Rant to generate random test data, and use its Test::Unit extension for property-based testing.

Rant is basically a recursive descent interpreter, each of its method returns a random value of some type (string, integer, float, etc.).

Its implementation has no alien mathematics inside. Completely side-effect-free-free.

h1. Install

$ gem install hayeah-rant --source http://gems.github.com

$ irb
> gem 'rant'
> require 'rant'
> Rant.gen.value { [integer,float] }
=> [20991307, 0.025756845811823]
> Rant.gen.value { [integer,float]}
=> [-376856492, 0.452245765751706]

h1. Data Generation

You can create random generators from the Rant class. Rant.gen is just returns a class instance of Rant.

> gen = Rant.new
> gen.value { [integer,float] }
=> [-163260081, 0.356075765934108]

h2. Getting Random Data Values

Rant#map(n,limit=10)
  call the generator n times, and collect values
Rant#each(n,limit=10)
  call a random block n times
Rant#value(limit=10)
  call a random block once, and get its value.

To collect an array of random data,

# we want 5
> gen.map(5) { integer }
=> [-380638946, -29645239, 344840868, 308052180, -154360970]

To iterate over random data,

> gen.each(5) { puts integer }
296971291
504994512
-402790444
113152364
502842783
=> nil

To get one value of random data,

> gen.value { integer }
=> 278101042

The optional argument @limit@ is used with generator guard. By default, if you want to generate n items, the generator tries at most n * 10 times.

This almost always succeeds,

> gen.map(5) { i = integer; guard i > 0; i }
=> [511765059, 250554234, 305947804, 127809156, 285960387]

This always fails,

> gen.map(10) { guard integer.is_a?(Float) }
Rant::TooManyTries: Exceed gen limit 100: 101 failed guards)

h2. Random Generating Methods

The API is similiar to QuickCheck, but not exactly the same. In particular @choose@ picks a random element from an array, and @range@ picks a integer from an interval.

h3. Simple Randomness

Rant#integer(n=nil)
  random positive or negative integer. Fixnum only.
Rant#range(lo,hi)
  random integer between lo and hi.
Rant#float
  random float
Rant#bool
  true or false
Rant#literal(value)
  No-op. returns value.
Rant#choose(*vals)
  Pick one value from among vals.

h3. Meta Randomness

A rant generator is just a mini interpreter. It's often useful to go meta,

Rant#call(gen)
  If gen is a Symbol, just do a method call with send.
  If gen is an Array, the first element of the array is the method name, the rest are args.
  If gen is a Proc, instance_eval it with the generator.

> gen.value { call(:integer) }
=> -240998958

> gen.value { call([:range,0,10]) }
=> 2

> gen.value { call(Proc.new { [integer] })}
=> [522807620]

The @call@ method is useful to implement other abstractions (See next subsection).

Rant#branch(*args)
  Pick a random arg among args, and Rant#call it.

50-50 chance getting an integer or float,

> gen.value { branch :integer, :float }
=> 0.0489446702931332
> gen.value { branch :integer, :float }
=> 494934533

h3. Frequencies

Rant#freq(*pairs)
  Takes a list of 2-tuples, the first of which is the weight, and the second a Rant#callable value, and returns a random value picked from the pairs. Follows the distribution pattern specified by the weights.

Twice as likely to get a float than integer. Never gets a ranged integer.

> gen.value { freq [1,:integer], [2,:float], [0,:range,0,10] }

If the "pair" is not an array, but just a symbol, @freq@ assumes that the weight is 1.

# 50-50 between integer and float
> gen.value { freq :integer, :float }

If a "pair" is an Array, but the first element is not an Integer, @freq@ assumes that it's a Rant method-call with arguments, and the weight is one.

# 50-50 chance generating integer limited by 10, or by 20.
> gen.value { freq [:integer,10], [:integer 20] }

h3. Sized Structure

A Rant generator keeps track of how large a datastructure it should generate with its @size@ attribute.

Rant#size
 returns the current size
Rant#sized(n,&block)
 sets the size for the duration of recursive call of block. Block is instance_eval with the generator.

Rant provides two methods that depends on the size

Rant#array(*branches)
  returns a sized array consisted of elements by Rant#calling random branches.
Rant#string(char_class=:print)
  returns a sized random string, consisted of only chars from a char_class.

The avaiable char classes for strings are:

:alnum
:alpha
:blank
:cntrl
:digit
:graph
:lower
:print
:punct
:space
:upper
:xdigit
:ascii

# sized 10 array of integer or float
> gen.value { sized(10) { array(:integer,:float)}}
=> [417733046, -375385433, 0.967812380000118, 26478621, 0.888588160450082, 250944144, 305584916, -151858342, 0.308123867823313, 0.316824642414253]

# fails if you forget to set the size.
> gen.value { array(:integer,:float)}
RuntimeError: size not set

If you set the size once, it applies to all subsequent recursive structures. Here's a sized 10 array of sized 10 strings,

> gen.value { sized(10) { array(:string)} }
=> ["1c}C/,9I#}", "hpA/UWPJ\\j", "H'~ERtI`|]", "%OUaW\\%uQZ", "Z2QdY=G~G!", "HojnxGDT3", "]a:L[B>bhb", "_Kl=&{tH^<", "ly]Yfb?`6c"]

Or a sized 10 array of sized 5 strings,

> gen.value { sized(10) { array Proc.new {sized(5) {string}}}}
=> ["S\"jf ", "d\\F-$", "-_8pa", "IN0iF", "SxRV$", ".{kQ7", "6>;fo", "}.D8)", "P(tS'", "y0v/v"]

Rant#array actually just delegate to Rant#freq, so you can use freq pairs:

> gen.value { sized(10) {array [1,:integer],[2,:float] }}
=> [0.983334733158678, -418176338, 0.976947175363592, 0.703390570421286, -478680395, 5483631, 0.966944106783513, 110469205, 0.540859146793544, 0.521813810037025]

h1. Property Testing

Rant extends Test::Unit for property testing. The extension is in its own module. So you need to require it.

require 'rant/check'

It defines,

Test::Unit::Assertions#property_of(&block)
  The block is used to generate random data with a generator. The method returns a Rant::Property instance, that has the method 'check'.

It's like this, using the gem 'shoulda'

# checks that integer only generates fixnum.
should "generate Fixnum only" do
   property_of  { integer }.check { |i| assert i.is_a?(Integer) }
end

The check block takes the generated data as its argument. One idiom I find useful is to include a parameter of the random data for the check argument. For example, if I want to check that Rant#array generates the right sized array, I could say,

should "generate right sized array" do
  property_of {
    len = integer
    [len,sized(len) { array :integer }]
  }.check { |(len,arr)|
    assert_equal len, arr.length
  }
end

That's about it. Enjoy :)

h1. Copyright

Copyright (c) 2009 Howard Yeh. See LICENSE for details.