Monkey-patch Oracle adapter to fix illegal zero-length identifier

ORA-01741: illegal zero-length identifier

This error is caused by composite key feature:
rails/rails@c18a95e

PR to fix it in oracle-enhanced:
rsim/oracle-enhanced#2471

Author: mayorova

config/initializers/oracle.rb

@@ -52,6 +52,26 @@ def _quote(value)
             super
           end
         end
+
+        # Fixing OCIError: ORA-01741: illegal zero-length identifier
+        # because of https://github.com/rails/rails/commit/c18a95e38e9860953236aed94c1bfb877fa3be84
+        # the value of `columns` is  [ "\"ACCOUNTS\".\"ID\"" ] which forms an incorrect query
+        # ... OVER (PARTITION BY ["\"ACCOUNTS\".\"ID\""] ORDER BY "ACCOUNTS"."ID") ...
+        # Will not be needed after https://github.com/rsim/oracle-enhanced/pull/2471 is merged and Rails upgraded
+        def columns_for_distinct(columns, orders) # :nodoc:
+          # construct a valid columns name for DISTINCT clause,
+          # ie. one that includes the ORDER BY columns, using FIRST_VALUE such that
+          # the inclusion of these columns doesn't invalidate the DISTINCT
+          #
+          # It does not construct DISTINCT clause. Just return column names for distinct.
+          orders.reject(&:blank?).map { |s|
+            s = visitor.compile(s) unless s.is_a?(String)
+            # remove any ASC/DESC modifiers
+            s.gsub(/\s+(ASC|DESC)\s*?/i, "")
+          }.reject(&:blank?).map.with_index { |column, i|
+            "FIRST_VALUE(#{column}) OVER (PARTITION BY #{columns.join(', ')} ORDER BY #{column}) AS alias_#{i}__"
+          }
+        end
       end)
     end
 
@@ -177,7 +197,7 @@ def set_database_settings(settings)
       end
     end)
 
-    ActiveRecord::ConnectionAdapters::OracleEnhanced::SchemaStatements.module_eval do
+      ActiveRecord::ConnectionAdapters::OracleEnhanced::SchemaStatements.module_eval do
       def add_index(table_name, column_name, **options) #:nodoc:
         # All this code is exactly the same as the original except the line of the ALTER TABLE, which adds an additional USING INDEX #{quote_column_name(index_name)}
         # The reason of this is otherwise it picks the first index that finds that contains that column name, even if it is shared with other columns and it is not unique.