Add setting for enumerating column names in SELECT statements

activerecord/CHANGELOG.md

@@ -1,3 +1,43 @@
+*   Add setting for enumerating column names in SELECT statements.
+
+    Adding a column to a Postgres database while the application is running can
+    change the result of wildcard `SELECT *` queries, which invalidates the result
+    of cached prepared statements and raises a PreparedStatementCacheExpired error.
+
+    When truthy, ActiveRecord will avoid wildcards and always include column names
+    in `SELECT` queries, which will return consistent results and avoid prepared
+    statement errors.
+
+    Before:
+
+    ```ruby
+    Book.limit(5)
+    # SELECT * FROM books LIMIT 5
+    ```
+
+    After:
+
+    ```ruby
+    # config/application.rb
+    module MyApp
+      class Application < Rails::Application
+        config.active_record.enumerate_columns_in_select_statements = true
+      end
+    end
+
+    # or, configure per-model
+    class Book < ApplicationRecord
+      self.enumerate_columns_in_select_statements = true
+    end
+    ```
+
+    ```ruby
+    Book.limit(5)
+    # SELECT id, author_id, name, format, status, language, etc FROM books LIMIT 5
+    ```
+
+    *Matt Duszynski*
+
 *   Only update dirty attributes once for cyclic autosave callbacks.
 
     Instead of calling `changes_applied` everytime `save` is called,

activerecord/lib/active_record/core.rb

@@ -129,6 +129,14 @@ def self.configurations
       # for multiple databases.
       mattr_accessor :suppress_multiple_database_warning, instance_writer: false, default: false
 
+      ##
+      # :singleton-method:
+      # Force enumeration of all columns in SELECT statements
+      # e.g. `SELECT first_name, last_name FROM ...` instead of `SELECT * FROM ...`
+      # This avoids PreparedStatementCacheExpired errors when a column is added
+      # to a Postgres database while the app is running.
+      class_attribute :enumerate_columns_in_select_statements, instance_accessor: false, default: false
+
       mattr_accessor :maintain_test_schema, instance_accessor: false
 
       class_attribute :belongs_to_required_by_default, instance_accessor: false

activerecord/lib/active_record/relation/query_methods.rb

@@ -1390,7 +1390,7 @@ def build_joins(join_sources, aliases = nil)
       def build_select(arel)
         if select_values.any?
           arel.project(*arel_columns(select_values))
-        elsif klass.ignored_columns.any?
+        elsif klass.ignored_columns.any? || klass.enumerate_columns_in_select_statements
           arel.project(*klass.column_names.map { |field| table[field] })
         else
           arel.project(table[Arel.star])

activerecord/test/cases/relation/select_test.rb

@@ -77,5 +77,15 @@ def test_star_select_with_joins_and_includes
         tags_count indestructible_tags_count tags_with_destroy_count tags_with_nullify_count
       ), posts.first.attributes.keys
     end
+
+    def test_enumerate_columns_in_select_statements
+      original_value = Post.enumerate_columns_in_select_statements
+      Post.enumerate_columns_in_select_statements = true
+      sql = Post.all.to_sql
+      Post.column_names.each do |column_name|
+        assert_includes sql, column_name
+      end
+      Post.enumerate_columns_in_select_statements = original_value
+    end
   end
 end

guides/source/configuring.md

@@ -473,6 +473,8 @@ in controllers and views. This defaults to `false`.
 
 * `config.active_record.queues.destroy` allows specifying the Active Job queue to use for destroy jobs. When this option is `nil`, purge jobs are sent to the default Active Job queue (see `config.active_job.default_queue_name`). It defaults to `nil`.
 
+* `config.active_record.enumerate_columns_in_select_statements` when truthy, will always include column names in `SELECT` statements, and avoid wildcard `SELECT * FROM ...` queries. This avoids prepared statement cache errors when adding columns to a Postgres database. Defaults to `false`.
+
 The MySQL adapter adds one additional configuration option:
 
 * `ActiveRecord::ConnectionAdapters::Mysql2Adapter.emulate_booleans` controls whether Active Record will consider all `tinyint(1)` columns as booleans. Defaults to `true`.