New Features¶ ↑

• Ruby 2.6+ endless ranges are now supported as condition specifier values, using a >= operator for them:

  DB[:t].where(c: 1...)
  # SELECT * FROM t WHERE (c >= 1)
  

• Ruby 2.6+ endless ranges are now supported in the pg_range extension:

  DB[:t].where(id: 1).update(r: 1...)
  # UPDATE t SET r = '[1,)' WHERE (id = 1)
  

• The :include option when creating indexes is now supported on PostgreSQL 11, specifying additional columns to include in the index without indexing them. This is useful to allow index only scans in additional cases.

• The :tablespace option is now supported when creating tables, indexes, and materialized views on PostgreSQL.

• The list plugin now supports a :top option, which can be used to specify the top of the list. The default value for the top of the list is 1, but using this option you can make the top of the list be 0.

Other Improvements¶ ↑

• In the pg_array_associations plugin, filtering by associations for many_to_pg_array associations now works correctly on PostgreSQL 11. Previously it did not work on PostgreSQL 11 due to new restrictions on using set returning functions in the the SELECT list.

• When setting the value of a column to the same value the column already has, for a new model object that has not yet been persisted, where the column is used as the foreign key for at least one many_to_one association, do not clear any related associations from the associations cache.

• In the pg_array extension, if there are separate conversion procs for timetz and time types, the conversion proc for the timetz[] type now correctly uses the conversion proc for the timetz type to convert scalar values, instead of the conversion proc for the time type.

• Empty arrays and hashes are now correctly handled in Dataset#{first,where_all,where_each,where_single_value} when a cached placeholder literalizer is used.

• In the tree plugin, Model#{ancestors,descendants,self_and_siblings} now work correctly when custom parent/children association names are used.

• The inner loop of the postgres adapter row fetching code is now 2-3% faster.

• When using the postgres adapter with pg-0.18+, set a type_map_for_queries for the connection to allow it to handle input type casts for Integer, Float, TrueClass, and FalseClass values without allocating strings.

• SQLTime.parse (and therefore Sequel.string_to_time) now respects the SQLTime.date and Sequel.application_timezone settings.

• The jdbc/postgresql adapter now correctly parses timetz types.

• On JRuby 9.2.0.0, when handling BC timestamps without timezones in the pg_extended_date_support extension, assume local time and not UTC time if the database timezone is not specified and Sequel.datetime_class is Time.

• Errors indicating that a MySQL database is in read-only mode are now treated as disconnect errors in the mysql and mysql2 adapters, for better behavior in failover scenarios.

• Sequel::Model datasets now support the use of IN/NOT IN operators where the second argument for the operator (the right hand side) is a set returning function. Previously, the Sequel::Model code assumed the right hand side of an IN/NOT IN operator was a datasets or array, since those are the only values where Sequel will automatically create such an operator.

• Sequel no longer loads the strscan library in the pg_array extension if it is not necessary because the parser from sequel_pg is used.