I am implementing a parent-child relationship, represented by hashes that look like this:

parent = { 'parent' => nil, 'id' => 3 }
child = { 'parent' => parent, 'id' => 7 }

These hashes are data given to me, for example, through a JSON API.

(As per a question in the comments, the JSON for the child would look like:)

  "parent": { "parent": null, "id": 3, ...etc.},
  "id": 7,

Parent and child have more keys like name, etc. (they are a lot so I am only showing the id as an example). They both have the same keys. There are only parents and children, i.e., no grandchildren or grandparents, etc. There is only one "level". Each child has only one parent, but each parent may have several children.

I wrapped the data in a class to represent it, so that my application can use an object. Both parent and child are represented by the same class. So I did this:

class Item
  def initialize(data)
    @data = data

  def parent
    @parent ||= Item.new(data['parent']) if data['parent']

  def id


  attr_reader :data

Every key in the hash has its method to retrieve it, and Item is the class that represents parents and children. I use it like this:

describe 'children' do
  let(:subject1) { Item.new(child) }

  it 'is an item' do

  it 'has an id' do

  it 'has a parent' do

describe 'parent' do
  let(:subject2) { subject1.parent }

  it 'is also an item' do

  it 'does not have a parent' do

With this, children know who is their parent, and for a parent to know its children we can iterate through all the children and compare their parent's id with the id we are searching.

However, I have a bad feeling about calling Item.new inside of Item. Is like a lot of things could go wrong. I thought of using a Item factory through dependency injection, but really all the knowledge about how to build a Item is in Item, so why introduce that dependency. Is this like a weird kind of recursion?

It doesn't help that I haven't seen this implementation anywhere (I'm not a computer scientist though). Is this bad code? I have a weird feeling, but I can't quite word why.

  • 1
    \$\begingroup\$ This is not bad code sometimes this is exactly what you need. The best example of this is Trees where every node is a tree \$\endgroup\$
    – Sam
    Mar 2, 2017 at 23:51
  • \$\begingroup\$ I don't think it is necessarily bad programming style. My concern would be that I think it would be better to have the parent create the children than the other way around. In your case, if two children have the same parents, they would actually create two distinct parent objects. It would also be difficult to ask a parent for all its children. \$\endgroup\$ Mar 3, 2017 at 18:15
  • \$\begingroup\$ Ah, that's a great point, @MarcRohloff, +1 \$\endgroup\$ Mar 3, 2017 at 18:56
  • \$\begingroup\$ Can you post an example of the JSON the API returns? \$\endgroup\$ Mar 24, 2017 at 11:50
  • \$\begingroup\$ @GregBurghardt I did already :) Sorry I may not have been all the clear I should when I wrote this. The parent and child JSON in the first code example. As I say further down the post, there may be more keys like name, type, etc. There is only one level though, i.e. no nesting. The only exception being the parent key, which saves another JSON object with the same set of keys as the root JSON object returned by the API. \$\endgroup\$ Mar 24, 2017 at 13:17

1 Answer 1


I don't think you should worry about calling new like that, but you can avoid it by embracing objects and Ruby, it's not perl, you're not stuck with hashes! Make your data into objects and if you need a hash representation of it then write a method that converts the object into a hash.

If you want to be able to keep track of objects that's when I'd use a hash, by putting it into a class variable, or even better, a class instance variable.


Since you've mentioned JSON in the comments, or perhaps you'd be passed the hash, then I'd still use objects, just add a way to coerce a hash into an object. A library I've forked called Grackle communicates with Twitter and can receive either JSON or XML. It does this by loading handlers, which means the same data object class can receive many differing types of input and still produce an object with a standard interface.

module Handlers
  module JSONHandler
    require 'json'

    def from_json json
      hash = JSON.parse(json)
      recursive hash

    # You might want to add an extra check to stop an infinite loop
    # Beware the data!
    def recursive hash
      return nil if hash.nil?
      return hash unless hash.respond_to? :fetch
      Item.new id: hash["id"], parent: recursive(hash["parent"])

class Item
  extend Handlers::JSONHandler

  def self.items  
    @items ||= {}    

  def initialize(id:, parent:nil)  
    @parent = parent    
    @id = id    
    self.class.items[@id] = self  

  attr_reader :parent, :id  

  def to_h  
    h = {id: @id, parent: (@parent && parent.id) }    
    h.reject{|k,v| v.nil? }    

  alias_method :to_hash,:to_h  
# => Item
parent = Item.new id: 3
# => #<Item:0x007fa94a4010c8 @parent=nil, @id=3>
child = Item.new id: 7, parent: parent
# => #<Item:0x007fa94a9abe38 @parent=#<Item:0x007fa94a4010c8 @parent=nil, @id=3>, @id=7>
# => {3=>#<Item:0x007fa94a4010c8 @parent=nil, @id=3>, 7=>#<Item:0x007fa94a9abe38 @parent=#<Item:0x007fa94a4010c8 @parent=nil, @id=3>, @id=7>}
# => #<Item:0x007fa94a4010c8 @parent=nil, @id=3>
# => #<Item:0x007fa94a9abe38 @parent=#<Item:0x007fa94a4010c8 @parent=nil, @id=3>, @id=7>
# => #<Item:0x007fa94a4010c8 @parent=nil, @id=3>
# => {:id=>3}
# => {:id=>7, :parent=>3}
json_parent = parent.to_h.to_json
# => "{\"id\":3}"
json_child = child.to_h.to_json
# => "{\"id\":7,\"parent\":3}"
parent2 = Item.from_json json_parent
# => #<Item:0x007fb543e26940 @parent=nil, @id=3>
child2 = Item.from_json json_child
# => #<Item:0x007fb5431d7270 @parent=3, @id=7>

and to make calling it easy:

def self.Item(*args, **keywords)
  case args.first
  when String # it's JSON
    Item.from_json args.first
  when Hash
    # you might want to nick the recursive bit from the handler
    # or write an Hash handler etc
    Item.new id: args.first["id"], parent: args.first["parent"]
  when Item
    Item.new id: keywords[:id], parent: keywords[:parent]

# => #<Item:0x007fb9b2765160 @parent=nil, @id=3>
# => #<Item:0x007fb9b272f330 @parent=3, @id=7>
# => #<Item:0x007fb9b26fdd30 @parent=nil, @id=3>
# => #<Item:0x007fb9b26c7dc0 @parent=3, @id=7>
# => #<Item:0x007fb9b26965b8 @parent=nil, @id=3>
# => #<Item:0x007fb9b2664928 @parent=3, @id=7>
Item(id: 3)
# => #<Item:0x007fb9b262ed50 @parent=nil, @id=3>
Item(id: 7, parent: parent)
# => #<Item:0x007fb9b25fc990 @parent=#<Item:0x007fb9b23bd7b0 @parent=nil, @id=3>, @id=7>
  • 1
    \$\begingroup\$ Thank you for your answer, it made me realize that I hadn't explained my question with enough clarity. The hashes are given to me. For example, through a JSON API. That's why I am creating this class to turn those hashes into an object that my application can use. Your example works perfectly fine for the opposite case, when you have an object and want to generate a hash from it. \$\endgroup\$ Mar 17, 2017 at 21:38
  • \$\begingroup\$ I see. Regardless of the type of input I'd alway coerce it into objects. If you're working with a tool then use its advantages, else you may as well pipe the JSON through awk and sed (not necessarily a bad idea if you're skilled with them:) I've updated the code to give an example. \$\endgroup\$
    – ian
    Mar 18, 2017 at 2:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.