Closure Tree

Common applications include modeling hierarchical data, like tags, threaded comments, page graphs in CMSes, and tracking user referrals.

Dramatically more performant than ancestry and acts_as_tree, and even more awesome than awesome_nested_set, closure_tree has some great features:

• Best-in-class select performance:
  • Fetch your whole ancestor lineage in 1 SELECT.
  • Grab all your descendants in 1 SELECT.
  • Get all your siblings in 1 SELECT.
  • Fetch all descendants as a nested hash in 1 SELECT.
  • Find a node by ancestry path in 1 SELECT.
• Best-in-class mutation performance:
  • 2 SQL INSERTs on node creation
  • 3 SQL INSERT/UPDATEs on node reparenting
• Support for concurrency (using with_advisory_lock)
• Tested against ActiveRecord 6.0+ with Ruby 2.7+
• Support for reparenting children (and all their descendants)
• Support for single-table inheritance (STI) within the hierarchy
• find_or_create_by_path for building out heterogeneous hierarchies quickly and conveniently
• Support for deterministic ordering
• Support for preordered traversal of descendants
• Support for rendering trees in DOT format, using Graphviz
• Excellent test coverage in a comprehensive variety of environments

See Bill Karwin's excellent Models for hierarchical data presentation for a description of different tree storage algorithms.

• Installation
• Warning
• Usage
• Accessing Data
• Polymorphic hierarchies with STI
• Deterministic ordering
• Concurrency
• FAQ
• Testing
• Change log

Note that closure_tree only supports ActiveRecord 6.0 and later, and has test coverage for MySQL, PostgreSQL, and SQLite.

1. Add gem 'closure_tree' to your Gemfile

2. Run bundle install

3. Add has_closure_tree (or acts_as_tree, which is an alias of the same method) to your hierarchical model:

   class Tag < ActiveRecord::Base
     has_closure_tree
   end
   
   class AnotherTag < ActiveRecord::Base
     acts_as_tree
   end

   Make sure you check out the large number of options that has_closure_tree accepts.

   IMPORTANT: Make sure you add has_closure_tree after attr_accessible and self.table_name = lines in your model.

   If you're already using other hierarchical gems, like ancestry or acts_as_tree, please refer to the warning section!

4. Add a migration to add a parent_id column to the hierarchical model. You may want to also add a column for deterministic ordering of children, but that's optional.

   class AddParentIdToTag < ActiveRecord::Migration
     def change
       add_column :tags, :parent_id, :integer
     end
   end

   The column must be nullable. Root nodes have a NULL parent_id.

5. Run rails g closure_tree:migration tag (and replace tag with your model name) to create the closure tree table for your model.

   By default the table name will be the model's table name, followed by "_hierarchies". Note that by calling has_closure_tree, a "virtual model" (in this case, TagHierarchy) will be created dynamically. You don't need to create it.

6. Run rake db:migrate

7. If you're migrating from another system where your model already has a parent_id column, run Tag.rebuild! and your tag_hierarchies table will be truncated and rebuilt.

   If you're starting from scratch you don't need to call rebuild!.

NOTE: Run rails g closure_tree:config to create an initializer with extra configurations. (Optional)

As stated above, using multiple hierarchy gems (like ancestry or nested set) on the same model will most likely result in pain, suffering, hair loss, tooth decay, heel-related ailments, and gingivitis. Assume things will break.

Create a root node:

grandparent = Tag.create(name: 'Grandparent')

Child nodes are created by appending to the children collection:

parent = grandparent.children.create(name: 'Parent')

Or by appending to the children collection:

child2 = Tag.new(name: 'Second Child')
parent.children << child2

Or by calling the "add_child" method:

child3 = Tag.new(name: 'Third Child')
parent.add_child child3

Or by setting the parent on the child :

Tag.create(name: 'Fourth Child', parent: parent)

Then:

grandparent.self_and_descendants.collect(&:name)
=> ["Grandparent", "Parent", "First Child", "Second Child", "Third Child", "Fourth Child"]

child1.ancestry_path
=> ["Grandparent", "Parent", "First Child"]

You can find as well as find_or_create by "ancestry paths".

If you provide an array of strings to these methods, they reference the name column in your model, which can be overridden with the :name_column option provided to has_closure_tree.

child = Tag.find_or_create_by_path(%w[grandparent parent child])

As of v5.0.0, find_or_create_by_path can also take an array of attribute hashes:

child = Tag.find_or_create_by_path([
  {name: 'Grandparent', title: 'Sr.'},
  {name: 'Parent', title: 'Mrs.'},
  {name: 'Child', title: 'Jr.'}
])

If you're using STI, The attribute hashes can contain the sti_name and things work as expected:

child = Label.find_or_create_by_path([
  {type: 'DateLabel', name: '2014'},
  {type: 'DateLabel', name: 'August'},
  {type: 'DateLabel', name: '5'},
  {type: 'EventLabel', name: 'Visit the Getty Center'}
])

Nodes can be moved around to other parents, and closure_tree moves the node's descendancy to the new parent for you:

d = Tag.find_or_create_by_path %w[a b c d]
h = Tag.find_or_create_by_path %w[e f g h]
e = h.root
d.add_child(e) # "d.children << e" would work too, of course
h.ancestry_path
=> ["a", "b", "c", "d", "e", "f", "g", "h"]

When it is more convenient to simply change the parent_id of a node directly (for example, when dealing with a form <select>), closure_tree will handle the necessary changes automatically when the record is saved:

j = Tag.find 102
j.self_and_ancestor_ids
=> [102, 87, 77]
j.update parent_id: 96
j.self_and_ancestor_ids
=> [102, 96, 95, 78]

hash_tree provides a method for rendering a subtree as an ordered nested hash:

b = Tag.find_or_create_by_path %w(a b)
a = b.parent
b2 = Tag.find_or_create_by_path %w(a b2)
d1 = b.find_or_create_by_path %w(c1 d1)
c1 = d1.parent
d2 = b.find_or_create_by_path %w(c2 d2)
c2 = d2.parent

Tag.hash_tree
=> {a => {b => {c1 => {d1 => {}}, c2 => {d2 => {}}}, b2 => {}}}

Tag.hash_tree(:limit_depth => 2)
=> {a => {b => {}, b2 => {}}}

b.hash_tree
=> {b => {c1 => {d1 => {}}, c2 => {d2 => {}}}}

b.hash_tree(:limit_depth => 2)
=> {b => {c1 => {}, c2 => {}}}

If your tree is large (or might become so), use :limit_depth.

Without this option, hash_tree will load the entire contents of that table into RAM. Your server may not be happy trying to do this.

HT: ancestry and elhoyos

Since most of closure_tree's methods (e.g. children) return regular ActiveRecord scopes, you can use the includes method for eager loading, e.g.

comment.children.includes(:author)

However, note that the above approach only eager loads the requested associations for the immediate children of comment. If you want to walk through the entire tree, you may still end up making many queries and loading duplicate copies of objects.

In some cases, a viable alternative is the following:

comment.self_and_descendants.includes(:author)

This would load authors for comment and all its descendants in a constant number of queries. However, the return value is an array of Comments, and the tree structure is thus lost, which makes it difficult to walk the tree using elegant recursive algorithms.

A third option is to use has_closure_tree_root on the model that is composed by the closure_tree model (e.g. a Post may be composed by a tree of Comments). So in post.rb, you would do:

# app/models/post.rb
has_closure_tree_root :root_comment

This gives you a plain has_one association (root_comment) to the root Comment (i.e. that with null parent_id).

It also gives you a method called root_comment_including_tree, which you can invoke as follows:

a_post.root_comment_including_tree(:author)

The result of this call will be the root Comment with all descendants and associations loaded in a constant number of queries. Inverse associations are also setup on all nodes, so as you walk the tree, calling children or parent on any node will not trigger any further queries and no duplicate copies of objects are loaded into memory.

The class and foreign key of root_comment are assumed to be Comment and post_id, respectively. These can be overridden in the usual way.

The same caveat stated above with hash_tree also applies here: this method will load the entire tree into memory. If the tree is very large, this may be a bad idea, in which case using the eager loading methods above may be preferred.

to_dot_digraph is suitable for passing into Graphviz.

For example, for the above tree, write out the DOT file with ruby:

File.open("example.dot", "w") { |f| f.write(Tag.root.to_dot_digraph) }

Then, in a shell, dot -Tpng example.dot > example.png, which produces:

If you want to customize the label value, override the #to_digraph_label instance method in your model.

Just for kicks, this is the test tree I used for proving that preordered tree traversal was correct:

When you include has_closure_tree in your model, you can provide a hash to override the following defaults:

• :parent_column_name to override the column name of the parent foreign key in the model's table. This defaults to "parent_id".
• :hierarchy_class_name to override the hierarchy class name. This defaults to the singular name of the model + "Hierarchy", like TagHierarchy.
• :hierarchy_table_name to override the hierarchy table name. This defaults to the singular name of the model + "_hierarchies", like tag_hierarchies.
• :dependent determines what happens when a node is destroyed. Defaults to nullify.
  • :nullify will simply set the parent column to null. Each child node will be considered a "root" node. This is the default.
  • :delete_all will delete all descendant nodes (which circumvents the destroy hooks)
  • :destroy will destroy all descendant nodes (which runs the destroy hooks on each child node)
  • nil does nothing with descendant nodes
• :name_column used by #find_or_create_by_path, #find_by_path, and ancestry_path instance methods. This is primarily useful if the model only has one required field (like a "tag").
• :order used to set up deterministic ordering
• :touch delegates to the belongs_to annotation for the parent, so touching cascades to all children (the performance of this for deep trees isn't currently optimal).

• Tag.root returns an arbitrary root node
• Tag.roots returns all root nodes
• Tag.leaves returns all leaf nodes
• Tag.hash_tree returns an ordered, nested hash that can be depth-limited.
• Tag.find_by_path(path, attributes) returns the node whose name path is path. See (#find_or_create_by_path).
• Tag.find_or_create_by_path(path, attributes) returns the node whose name path is path, and will create the node if it doesn't exist already.See (#find_or_create_by_path).
• Tag.find_all_by_generation(generation_level) returns the descendant nodes who are generation_level away from a root. Tag.find_all_by_generation(0) is equivalent to Tag.roots.
• Tag.with_ancestor(ancestors) scopes to all descendants whose ancestors(s) is/are in the given list.
• Tag.with_descendant(ancestors) scopes to all ancestors whose descendant(s) is/are in the given list.
• Tag.lowest_common_ancestor(descendants) finds the lowest common ancestor of the descendants.

• tag.root returns the root for this node
• tag.root? returns true if this is a root node
• tag.root_of?(node) returns true if current node is root of another one
• tag.child? returns true if this is a child node. It has a parent.
• tag.leaf? returns true if this is a leaf node. It has no children.
• tag.leaves is scoped to all leaf nodes in self_and_descendants.
• tag.depth returns the depth, or "generation", for this node in the tree. A root node will have a value of 0.
• tag.parent returns the node's immediate parent. Root nodes will return nil.
• tag.parent_of?(node) returns true if current node is parent of another one
• tag.children is a has_many of immediate children (just those nodes whose parent is the current node).
• tag.child_ids is an array of the IDs of the children.
• tag.child_of?(node) returns true if current node is child of another one
• tag.ancestors is a ordered scope of [ parent, grandparent, great grandparent, … ]. Note that the size of this array will always equal tag.depth.
• tag.ancestor_ids is an array of the IDs of the ancestors.
• tag.ancestor_of?(node) returns true if current node is ancestor of another one
• tag.self_and_ancestors returns a scope containing self, parent, grandparent, great grandparent, etc.
• tag.self_and_ancestors_ids returns IDs containing self, parent, grandparent, great grandparent, etc.
• tag.siblings returns a scope containing all nodes with the same parent as tag, excluding self.
• tag.sibling_ids returns an array of the IDs of the siblings.
• tag.self_and_siblings returns a scope containing all nodes with the same parent as tag, including self.
• tag.descendants returns a scope of all children, childrens' children, etc., excluding self ordered by depth.
• tag.descendant_ids returns an array of the IDs of the descendants.
• tag.descendant_of?(node) returns true if current node is descendant of another one
• tag.self_and_descendants returns a scope of self, all children, childrens' children, etc., ordered by depth.
• tag.self_and_descendant_ids returns IDs of self, all children, childrens' children, etc., ordered by depth.
• tag.family_of? returns true if current node and another one have a same root.
• tag.hash_tree returns an ordered, nested hash that can be depth-limited.
• tag.find_by_path(path) returns the node whose name path from tag is path. See (#find_or_create_by_path).
• tag.find_or_create_by_path(path) returns the node whose name path from tag is path, and will create the node if it doesn't exist already.See (#find_or_create_by_path).
• tag.find_all_by_generation(generation_level) returns the descendant nodes who are generation_level away from tag.
  • tag.find_all_by_generation(0).to_a == [tag]
  • tag.find_all_by_generation(1) == tag.children
  • tag.find_all_by_generation(2) will return the tag's grandchildren, and so on.
• tag.destroy will destroy a node and do something to its children, which is determined by the :dependent option passed to has_closure_tree.

Polymorphic models using single table inheritance (STI) are supported:

1. Create a db migration that adds a String type column to your model
2. Subclass the model class. You only need to add has_closure_tree to your base class:

class Tag < ActiveRecord::Base
  has_closure_tree
end
class WhenTag < Tag ; end
class WhereTag < Tag ; end
class WhatTag < Tag ; end

Note that if you call rebuild! on any of the subclasses, the complete Tag hierarchy will be emptied, thus taking the hiearchies of all other subclasses with it (issue #275). However, only the hierarchies for the class rebuild! was called on will be rebuilt, leaving the other subclasses without hierarchy entries.

You can work around that by overloading the rebuild! class method in all your STI subclasses and call the super classes rebuild! method:

class WhatTag < Tag
  def self.rebuild!
    Tag.rebuild!
  end
end

This way, the complete hierarchy including all subclasses will be rebuilt.

By default, children will be ordered by your database engine, which may not be what you want.

If you want to order children alphabetically, and your model has a name column, you'd do this:

class Tag < ActiveRecord::Base
  has_closure_tree order: 'name'
end

If you want a specific order, add a new integer column to your model in a migration:

t.integer :sort_order

and in your model:

class OrderedTag < ActiveRecord::Base
  has_closure_tree order: 'sort_order', numeric_order: true
end

When you enable order, you'll also have the following new methods injected into your model:

• tag.siblings_before is a scope containing all nodes with the same parent as tag, whose sort order column is less than self. These will be ordered properly, so the last element in scope will be the sibling immediately before self
• tag.siblings_after is a scope containing all nodes with the same parent as tag, whose sort order column is more than self. These will be ordered properly, so the first element in scope will be the sibling immediately "after" self

If your order column is an integer attribute, you'll also have these:

• The class method #roots_and_descendants_preordered, which returns all nodes in your tree, pre-ordered.

• node1.self_and_descendants_preordered which will return descendants, pre-ordered.

• node1.append_child(node2) (which is an alias to add_child), which will

  1. set node2's parent to node1
  2. set node2's sort order to place node2 last in the children array

• node1.prepend_child(node2) which will

  1. set node2's parent to node1
  2. set node2's sort order to place node2 first in the children array Note that all of node1's children's sort_orders will be incremented

• node1.prepend_sibling(node2) which will

  1. set node2 to the same parent as node1,
  2. set node2's order column to 1 less than node1's value, and
  3. increment the order_column of all children of node1's parents whose order_column is > node2's new value by 1.

• node1.append_sibling(node2) which will

  1. set node2 to the same parent as node1,
  2. set node2's order column to 1 more than node1's value, and
  3. increment the order_column of all children of node1's parents whose order_column is > node2's new value by 1.

root = OrderedTag.create(name: 'root')
a = root.append_child(Label.new(name: 'a'))
b = OrderedTag.create(name: 'b')
c = OrderedTag.create(name: 'c')

# We have to call 'root.reload.children' because root won't be in sync with the database otherwise:

a.append_sibling(b)
root.reload.children.pluck(:name)
=> ["a", "b"]

a.prepend_sibling(b)
root.reload.children.pluck(:name)
=> ["b", "a"]

a.append_sibling(c)
root.reload.children.pluck(:name)
=> ["b", "a", "c"]

b.append_sibling(c)
root.reload.children.pluck(:name)
=> ["b", "c", "a"]

With numeric ordering, root nodes are, by default, assigned order values globally across the whole database table. So for instance if you have 5 nodes with no parent, they will be ordered 0 through 4 by default. If your model represents many separate trees and you have a lot of records, this can cause performance problems, and doesn't really make much sense.

You can disable this default behavior by passing dont_order_roots: true as an option to your delcaration:

has_closure_tree order: 'sort_order', numeric_order: true, dont_order_roots: true

In this case, calling prepend_sibling and append_sibling on a root node or calling roots_and_descendants_preordered on the model will raise a RootOrderingDisabledError.

The dont_order_roots option will be ignored unless numeric_order is set to true.

Several methods, especially #rebuild and #find_or_create_by_path, cannot run concurrently correctly. #find_or_create_by_path, for example, may create duplicate nodes.

Database row-level locks work correctly with PostgreSQL, but MySQL's row-level locking is broken, and erroneously reports deadlocks where there are none. To work around this, and have a consistent implementation for both MySQL and PostgreSQL, with_advisory_lock is used automatically to ensure correctness.

If you are already managing concurrency elsewhere in your application, and want to disable the use of with_advisory_lock, pass with_advisory_lock: false in the options hash:

class Tag
  has_closure_tree with_advisory_lock: false
end

Note that you will eventually have data corruption if you disable advisory locks, write to your database with multiple threads, and don't provide an alternative mutex.

You can customize error messages using I18n:

en-US:
  closure_tree:
    loop_error: Your descendant cannot be your parent!

Yup! Ilya Bodrov wrote Nested Comments with Rails.

No. Please see issue 86 for details.

No. update_attribute skips the validation hook that is required for maintaining the hierarchy table.

No. Please see issue 197 for details.

No. This gem's API is based on the assumption that each node has either 0 or 1 parent.

The underlying closure tree structure will support multiple parents, but there would be many breaking-API changes to support it. I'm open to suggestions and pull requests.

Test fixtures aren't going to be running your after_save hooks after inserting all your fixture data, so you need to call .rebuild! before your test runs. There's an example in the spec tag_spec.rb:

  describe "Tag with fixtures" do
    fixtures :tags
    before :each do
      Tag.rebuild! # <- required if you use fixtures
    end

However, if you're just starting with Rails, may I humbly suggest you adopt a factory library, rather than using fixtures? Lots of people have written about this already.

This is expected if you aren't using MySQL or Postgresql for your tests.

SQLite doesn't have advisory locks, so we resort to file locking, which will only work if the FLOCK_DIR is set consistently for all ruby processes.

In your spec_helper.rb or minitest_helper.rb, add a before and after block:

before do
  ENV['FLOCK_DIR'] = Dir.mktmpdir
end

after do
  FileUtils.remove_entry_secure ENV['FLOCK_DIR']
end

When building from source, the mysql2, pg, and sqlite gems need their native client libraries installed on your system. Note that this error isn't specific to ClosureTree.

On Ubuntu/Debian systems, run:

sudo apt-get install libpq-dev libsqlite3-dev libmysqlclient-dev
bundle install

A bug was introduced in MySQL's query optimizer. See the workaround here.

Upgrade to MySQL 5.7.12 or later if you see this issue:

Mysql2::Error: You can't specify target table '*_hierarchies' for update in FROM clause

Closure tree comes with some RSpec2/3 matchers which you may use for your tests:

require 'spec_helper'
require 'closure_tree/test/matcher'

describe Category do
 # Should syntax
 it { should be_a_closure_tree }
 # Expect syntax
 it { is_expected.to be_a_closure_tree }
end

describe Label do
 # Should syntax
 it { should be_a_closure_tree.ordered }
 # Expect syntax
 it { is_expected.to be_a_closure_tree.ordered }
end

describe TodoList::Item do
 # Should syntax
 it { should be_a_closure_tree.ordered(:priority_order) }
 # Expect syntax
 it { is_expected.to be_a_closure_tree.ordered(:priority_order) }
end

Closure tree is tested under every valid combination of

• Ruby 2.7+
• ActiveRecord 6.0+
• PostgreSQL, MySQL, and SQLite. Concurrency tests are only run with MySQL and PostgreSQL.

$ bundle
$ appraisal bundle # this will install the matrix of dependencies
$ appraisal rake # this will run the tests in all combinations
$ appraisal activerecord-7.0 rake # this will run the tests in AR 7.0 only
$ appraisal activerecord-7.0 rake spec # this will run rspec in AR 7.0 only
$ appraisal activerecord-7.0 rake test # this will run minitest in AR 7.0 only

By default the test are run with sqlite3 only. You run test with other databases by passing the database url as environment variable:

$ DATABASE_URL=postgres://localhost/my_database appraisal activerecord-7.0 rake test 

See the change log.

• The 45+ engineers around the world that have contributed their time and code to this gem (see the changelog!)
• https://github.com/collectiveidea/awesome_nested_set
• https://github.com/patshaughnessy/class_factory
• JetBrains, which provides an open-source license to RubyMine for the development of this project.