New SQL Function Features¶ ↑

SQL::Function now supports an options hash for functions. Unfortunately, since SQL::Function#initialize does not support an options hash, you need to use SQL::Function.new! to create a function with an options hash. You can also call methods on the SQL::Function instance, which will return a new SQL::Function with the appropriate option set.
SQL::Function#quoted has been added, which will return a new SQL::Function instance that will quote the function name (if the database supports quoting function names).
SQL::Function#unquoted has been added, which will return a new SQL::Function instance that will not quote the function name.
SQL::Function#lateral has been added, which will return a new SQL::Function instance that will be preceded by LATERAL when literalized, useful for set-returning functions.
SQL::Function#within_group has been added, for creating ordered-set and hypothetical-set functions that use WITHIN GROUP.
SQL::Function#filter has been added, for creating filtered aggregate function calls using FILTER.
SQL::Function#with_ordinality has been added, for creating set returning functions that also include a row number for every row in the set, using WITH ORDINALITY.

New PostgreSQL Features¶ ↑

The jsonb type added in 9.4 is now supported in the pg_json extension. To create a jsonb type manually, you need to call Sequel.pg_jsonb.

The new json and jsonb functions and operators added in 9.4 are now supported in the pg_json_ops extension. You can use the jsonb functions and operators by creating a Postgres::JSONBOp using Sequel.pg_jsonb_op.
Database#full_text_search now takes a :rank option to order by the ranking.
Database#refresh_view now supports a :concurrently option, to refresh a materialized view concurrently, supported on 9.4+.
Postgres::ArrayOp#cardinality has been added to the pg_array_ops extension, for easy use of the cardinality method added in 9.4.
Postgres::ArrayOp#unnest in the pg_array_ops extension now accepts arguments. PostgreSQL 9.4+ supports this if unnest is used in the FROM clause.

Sequel now supports derived column lists (table aliases that include column aliases) via Sequel.as and SQL::AliasedMethods#as:
```
Sequel.as(:table, :alias, [:c1, :c2])
# table AS alias(c1, c2)
```
Not all databases support this, but it is in SQL92 and Sequel now supports it by default. Derived column lists make it easier to alias columns when using set-returning functions.

Dataset#from_self now supports derived column lists via the new :column_aliases option (which requires the :alias option to take effect).
Database#create_view now supports a :check option, to use WITH CHECK OPTION. You can also use :check=>:local for WITH LOCAL CHECK OPTION. These clauses make it so when you are inserting into/updating the view, you can only modify rows in the underlying table if the result would be returned by the view.
The :after_connect Database option proc now can accept two arguments. If the arity of the proc is 2, Sequel will pass both the connection object and the shard symbol.
The class_table_inheritance plugin now supports a :model_map option similar to the single_table_inheritance plugin, allowing use of the plugin without storing ruby class names in the database. Note that if you use this option, you must set the correct value for the kind column manually when creating the row.
Support for CUBRID/SQLAnywhere emulation has been added to the mock adapter.

Dataset#import now supports a default slice size, which Sequel sets to 500 on SQLite as that is the limit that SQLite supports in a single statement.
The serialization plugin now only modifies changed_columns in the setter method if the deserialized value has changed, similar to how Sequel's standard column setters work. Note that if you are mutating the deserialized value (i.e. not calling the setter method), you still need to use the serialization_modification_detection plugin.
Plugins that set column values for new objects before creation now use before_validation instead of before_create, which works better when the auto_validations plugin is used.
The :read_only transaction option is now applied per-savepoint on PostgreSQL. Note that this allows you to have a READ ONLY savepoint in a READ WRITE transaction, it does not allow you to have a READ WRITE savepoint in a READ ONLY transaction.
In the ibm_db adapter, fix warnings when using certain column names.
Support connecting to a DB2 catalog name in the ibm_db adapter, by providing a :database option without a :host or :port option.
The mock adapter now sets an emulated version when using MySQL and SQLite. Additionally, the emulated version for PostgreSQL and Microsoft SQL Server has been updated.

External adapters that override Dataset#as_sql_append now need to have the method accept two arguments.
Model.eager_loading_dataset, .apply_association_dataset_opts, and .def_{add_method,association_dataset_methods,remove_methods} have been removed (they were deprecated in 4.10.0).
SQL::WindowFunction and SQL::EmulatedFunction classes are now deprecated, as well as Dataset methods that literalize instances of these classes. These classes are replaced by using options on SQL::Function instances.
Passing a table_alias argument when creating an SQL::JoinClause manually is no longer supported. You now need to pass the table as an SQL::AliasedExpression if the table needs to be aliased.
ASTTransformer no longer transforms the table alias for SQL::JoinClause. This is for consistency with SQL::AliasedExpression.
SQL standard casts are now used in Database#full_text_search, which can break tests that expect specific SQL.
The to_dot extension now uses slightly different output for SQL::Function and SQL::JoinClause instances.