Ruby version: 1.9.3-p327

Open Struct

Here is another read of a Ruby core class: OpenStruct.

Initializing an OpenStruct

You can create an OpenStuct using hash - or without any attributes.

def initialize(hash=nil)  
  @table = {}
  if hash
    hash.each_pair do |k, v|
    k = k.to_sym
    @table[k] = v

If you use a hash openstruct will go ahead and store all the key/value pairs in an internal hash called table.

def new_ostruct_member(name)  
  name = name.to_sym
  unless respond_to?(name)
    define_singleton_method(name) { @table[name] }
    define_singleton_method("#{name}=") { |x| modifiable[name] = x }
protected :new_ostruct_member  

It will also create getters / setters using definesingletonmethod.

require 'ostruct'
coordinate = => 1, :y => 2)
#<OpenStruct x=1, y=2>


But wait.. What does this modifiable thing in the dynamic setter do? Lets have a look at the code.

def modifiable  
    @modifiable = true
     raise TypeError, "can't modify frozen #{self.class}", caller(3)
protected :modifiable  

Huh? It sets a attribute modifiable and if it does not succeed it throws an exception? If it is modifiable it returns the @table?

Well to be honest this is not clear at all. The text of the exception states that this is about frozen objects. Hm lets validate this:

foo =
# {}
# nil
# TypeError: can't modify frozen OpenStruct

Ah! - It acts as kind of a gatekeeper to the table, by preventing you from changing the OpenStruct if it's frozen. I'm not sure if I like this - it not to clear to me - on the other hand it is tell don't ask. I guess I have to think about it.

Method missing

There is another feature of OpenStruct. It has setters for attributes that do not yet exist.

foo =
foo.x = 1
foo.y = 2
#<OpenStruct x=1, y=2>

In Ruby you can achieve this using method_missing.

def method_missing(mid, *args) # :nodoc:  
  mname = mid.id2name
  len = args.length
  if mname.chomp!('=') && mid != :[]=
    if len != 1
      raise ArgumentError, "wrong number of arguments (#{len} for 1)", caller(1)
    modifiable[new_ostruct_member(mname)] = args[0]
  elsif len == 0 && mid != :[]
    raise NoMethodError, "undefined method `#{mid}' for #{self}", caller(1)

Hm… Wow there is some stuff I have never seen before. Lets try to read it line by line.

mid.id2name converts the method's id (which is a symbol). Into a string.

If the method name has a = and only argument new_ostruct_member creates a new entry in the table. Afterwards the value if assigned.

The elseif block handles reads on attributes that were not defined. Getter method have no arguments of course.

foo =
# nil

It looks the mid up in the @table. I have no idea why it does not return nil in the first place here. The effect is the same, because ach time we add a new attributes new_ostruct_member creates dynamic methods. So if the attribute is already defined, we do not end up in method missing.

For all other functions except:
* These that have an "=" and one argument
* And these that have no argument
* And are not :[], :[]=

It raises an exception. If you wondering what this caller(1) thing is:

Returns the current execution stack—an array containing strings in the form “file:line” or “file:line: in method’”. The optional start parameter determines the number of initial stack entries to omit from the result.
Link to the Documentation

Duplicating an OpenStruct

I had quite a hard time of figuring out what the following function was doing. I tried really hard to call it from the console.. :/

 def initialize_copy(orig)
    @table = @table.dup
    @table.each_key{|key| new_ostruct_member(key)}

Ruby uses initialize_copy whenever you .clone or .dup an object. To get the whole picture read this great article about initializedup, initializeclone and initialize_copy!


InspectKey = :__inspect_key__ # :nodoc:

def inspect  
  str = "#<#{self.class}"

  ids = (Thread.current[InspectKey] ||= [])
  if ids.include?(object_id)
    return str << ' ...>'

  ids << object_id
    first = true
    for k,v in @table
      str << "," unless first
      first = false
      str << " #{k}=#{v.inspect}"
    return str << '>'
alias :to_s :inspect  

The last part of this is straightforward. The method prints all attributes like lenght=21. It separates them by a space and a comma and appends the attribute info to the class name.

#<OpenStruct name="Brussel Sprouts">

What this inspect_key stuff is all about? Have a look at this experiment.

  puts :__inspect_key__
  # nil { puts :__inspect_key__ }
#<Thread:0x007fe2b487b650 sleep>

You see? Every thread has its own key. do
    a = => "foo", :world => "bar")
    puts a.inspect
#<OpenStruct hello="foo", world="bar"> => #<Thread:0x007fe2b4817d80 run>

Marshalling support

The marshaling library converts collections of Ruby objects into a byte stream, allowing them to be stored outside the currently active script. This data may subsequently be read and the original objects reconstituted. Link to the Documentation

def marshal_dump  

def marshal_load(x)  
  @table = x
  @table.each_key{|key| new_ostruct_member(key)}

I would have called it serialization / deserialization 😉 There are 2 function that OpenStruct implements to achieve this.

*marshal_dump just returns the table - it's already a hash, so Marshal can handle it.

marshal_load just assign this same hash and creates all the required method to access the attributes.

Lets have a look at this in action.

point = => 1, :y => 2)
#<OpenStruct x=1, y=2>

serialized_point = Marshal.dump(a)

#<OpenStruct x=1, y=2>

Some changes in trunk

The version in trunk is slightly different from the version disscussed here. It contains the functions [] and []=.

 def [](name)

def []=(name, value)  
  modifiable[new_ostruct_member(name)] = value

Last words

OpenStruct is quite nice. It comes in handy when you want to replace a hash with a value object. The addition in trunk will make this even easier in the future because it's a drop in replacement for simple cases and the Tests will stay green 😉

Many people find that the attr_reader part for non-specified attributes is a big problem, when you use OpenStruct. I agree. You have to be aware of this or it will bite you.

It was fun reading this class. I did't like the modifiable method. It's not really as clear as it could be.

  • There's a method called .id2name
  • Some random stuff about Threads
  • Rubyists call serialization marshalling
  • A weired way to detect a frozen object
  • Ways to hook into the dup and clone methods

Go read some code as well! Its a highly educating activity.

Picture of cup licensed under: (CC BY-NC-SA 2.0) Licensor: rebecca∞mahoney