Document and test with column collation

.github/workflows/run_test_suite.yml

@@ -83,7 +83,6 @@ jobs:
       - name: run pg tests
         env:
           DB: pg
-          COLLATE_SYMBOLS: false
         run: |
           COUNT=1
           while ! pg_isready ; do

README.md

@@ -48,24 +48,32 @@ $ rails g migration add_ancestry_to_[table] ancestry:string:index
 # rails g migration add_name_to_[people] name:string:index
 ```
 
+You will have best results to set the collation of the field. It works without
+the collation, but without collation, it will not use the ancestry index. Alternatively
+adding oppset will fix this.
+
+Postgres on ubuntu in particular has been troublesome because by default,
+it sorts ignoring slashes.
+
+
+For postgres use `collate: :default`, and for mysql and sqllite use `collate: :binary`.
+
+```ruby
+class AddAncestryToTable < ActiveRecord::Migration[6.1]
+  def change
+    add_column :table, :ancestry, :string, collation: :default # alt: :binary
+    add_index :table, :ancestry
+  end
+end
+```
+
 *   Migrate your database:
 
 ```bash
 $ rake db:migrate
 ```
 
-Depending upon your comfort with databases, you may want to create the column
-with `C` or `POSIX` encoding. This is a more primitive encoding and just compares
-bytes. Since this column will just contain numbers and slashes, it works much
-better. It also works better for the uuid case as well.
-
-Alternatively, if you create a [`text_pattern_ops`](https://www.postgresql.org/docs/current/indexes-opclass.html) index for your postgresql column, subtree selection will use an efficient index for you regardless of whether you created the column with `POSIX` encoding.
-
-If you opt out of this, and are trying to run tests on postgres, you may need to
-set the environment variable `COLLATE_SYMBOLS=false`. Sorry to say that a discussion
-on this topic is out of scope. The important take away is postgres sort order is
-not consistent across operating systems but other databases do not have this same
-issue.
+NOTE: If you decide to add collation to this column at a later time, please remember to drop and recreate this index.
 
 NOTE: A Btree index (as is recommended) has a limitaton of 2704 characters for the ancestry column. This means you can't have an tree with a depth that is too great (~> 900 items at most).
 

test/concerns/scopes_test.rb

@@ -52,12 +52,8 @@ def test_chained_scopes
   def test_order_by
     AncestryTestDatabase.with_model :depth => 3, :width => 3 do |model, _roots|
       # Some pg databases do not use symbols in sorting
-      # if this is failing, try running the test via DB=pg COLLATE_SYMBOLS=false rake test
-      if ENV["COLLATE_SYMBOLS"].to_s =~ /false/i
-        expected = model.all.sort_by { |m| [m.ancestor_ids.map(&:to_s).join, m.id.to_i] }
-      else
-        expected = model.all.sort_by { |m| [m.ancestor_ids.map(&:to_s), m.id.to_i] }
-      end
+      # if this is failing, try tweaking the collation of your ancestry columns
+      expected = model.all.sort_by { |m| [m.ancestor_ids.map(&:to_s), m.id.to_i] }
       actual = model.ordered_by_ancestry_and(:id)
       assert_equal (expected.map { |r| [r.ancestor_ids, r.id.to_s] }), (actual.map { |r| [r.ancestor_ids, r.id.to_s] })
     end

test/environment.rb

@@ -56,7 +56,7 @@ def self.setup
 
       # This only affects postgres
       # the :ruby code path will get tested in mysql and sqlite3
-      Ancestry.default_update_strategy = :sql if db_type == "pg"
+      Ancestry.default_update_strategy = :sql if postgres?
 
     rescue => err
       if ENV["CI"]
@@ -69,6 +69,18 @@ def self.setup
     end
   end
 
+  # pass ANCESTRY_LOCALE=default to not override locale on ancestry
+  def self.ancestry_collation
+    env = ENV["ANCESTRY_LOCALE"].presence
+    if env
+      env
+    elsif postgres?
+      "C"
+    else
+      "binary"
+    end
+  end
+
   def self.with_model options = {}
     depth                = options.delete(:depth) || 0
     width                = options.delete(:width) || 0
@@ -79,8 +91,11 @@ def self.with_model options = {}
     table_options={}
     table_options[:id] = options.delete(:id) if options.key?(:id)
 
+    column_options = {:collation => ancestry_collation}
+    column_options = {} if column_options[:collation] == "default"
+
     ActiveRecord::Base.connection.create_table 'test_nodes', **table_options do |table|
-      table.string options[:ancestry_column] || :ancestry
+      table.string options[:ancestry_column] || :ancestry, **column_options
       table.integer options[:depth_cache_column] || :ancestry_depth if options[:cache_depth]
       if options[:counter_cache]
         counter_cache_column = options[:counter_cache] == true ? :children_count : options[:counter_cache]
@@ -128,6 +143,9 @@ def self.create_test_nodes model, depth, width, parent = nil
     else; []; end
   end
 
+  def self.postgres?
+    db_type == "pg"
+  end
   private
 
   def self.db_type