= Deprecated Features

* The identifier_columns plugin is now deprecated.  There is no reason
  to use it when Sequel.split_symbols = false, which will be the
  default in Sequel 5.

* The filter_having, hash_aliases, and sequel_3_dataset_methods
  extensions are now deprecated.  They only existed for backwards
  compatibility with Sequel 3.

* The query_literals extension is now deprecated.  It changes behavior
  in a way that makes SQL injections more likely.

* The meta_def extension is now deprecated.  It is no longer necessary,
  since on ruby 1.9+ you can use define_singleton_method.

* The empty_array_ignore_nulls extension has been deprecated.  It
  has been a no-op since Sequel 4.25.0.

* The cti_base_model, cti_key, and cti_model_map class methods in
  the class_table_inheritance plugin are now deprecated. Use
  cti_models.first instead of cti_base_model, sti_key
  instead of cti_key, and sti_model_map instead of cti_model_map.

* The :strict option in the nested_attributes plugin is now deprecated.
  Switch to using the :unmatched_pk option.

* Database#reset_conversion_procs on PostgreSQL is now deprecated.
  There will be no need to call it in Sequel 5.

* Using global conversion procs added by the pg_* extensions, without
  loading the pg_* extension into the Database instance, are now
  deprecated.  Additionally, using PGArray.register or PGRange.register
  to register global types is now also deprecated.  Use
  Database#register_array_type or Database#register_range_type instead
  to register the types on a per-Database basis.

* Treating :natural_inner join type as NATURAL LEFT JOIN on MySQL is
  now deprecated.  MySQL doesn't support NATURAL INNER JOIN, but if
  you were going to convert it, NATURAL JOIN would make more sense.

* Unexpected values passed to Dataset#insert_conflict on SQLite are
  now deprecated.  Only values that result in one of the following
  strings will be allowed in Sequel 5: ROLLBACK, ABORT, FAIL, IGNORE,
  and REPLACE.

* The Dataset#and, #exclude_where, #interval, and #range methods
  are now deprecated.  Undeprecated copies are now available in the
  new sequel_4_dataset_methods extension.

* Model.<< is now deprecated.  Intuitivately, you would except this
  to call Model.create and return the Model class, but it calls <<
  on the model's dataset and returns the dataset.

* The Sequel::Postgres::PG_TYPES constant is now deprecated.  All
  conversion procs should now be added on a per-Database basis using
  add_conversion_proc or add_named_conversion_proc.  The following
  private Database methods related to conversion procs are now
  deprecated, though some are still called internally and therefore
  do not have deprecation warnings:

  * add_named_conversion_procs
  * conversion_procs_updated
  * convert_named_procs_to_procs
  * copy_conversion_procs
  * get_conversion_procs

  Related to this, loading the sequel/adapters/utils/pg_types.rb file
  is now deprecated.

* The following adapter or database specific global accessors for setting
  defaults are now deprecated:

  * Sequel::DB2.use_clob_as_blob
  * Sequel::IBMDB.convert_smallint_to_bool
  * Sequel::MySQL.convert_invalid_date_time
  * Sequel::MySQL.convert_tinyint_to_bool
  * Sequel::MySQL.default_charset
  * Sequel::MySQL.default_collate
  * Sequel::MySQL.default_engine
  * Sequel::Postgres.use_iso_date_format
  * Sequel::Postgres.client_min_messages
  * Sequel::Postgres.force_standard_strings 
  * Sequel::SqlAnywhere.convert_smallint_to_bool

  Use the Database instance accessors or Database options instead to
  change behavior.

* The following adapter or database specific dataset mutation methods are
  now deprecated:

  * convert_smallint_to_bool= (ibmdb adapter, SQLAnywhere)
  * convert_types= (jdbc adapter)
  * mssql_unicode_strings= (Microsoft SQL Server)

  Use the with_* methods which return a modified copy of the dataset
  instead of these mutation methods.

* The Dataset#non_sql_options private method is now deprecated.
  External adapters that overrode this method should switch to
  overriding Dataset#non_sql_option?.

* The Database#timestamp_convertor private method in the jdbc adapter
  is now deprecated.  Users should switch to method(:timestamp_convert).

* Modification of the Sequel::JDBC::TypeConvertor class is now
  deprecated.  External jdbc subadapters that were using this to add
  custom conversion procs should be modified.

* Having the pg_row extension respect conversion procs for subtypes
  added after the registeration of the composite type is now deprecated.
  Now, all subtypes should have the appropriate conversion proc added
  before the composite type is registered.

* Array#sql_array in the core_extensions extension is now deprecated.
  Switch to using Array#sql_value_list.

* The SEQUEL_POSTGRES_USES_PG constant added by the postgres adapter
  is now deprecated. Sequel::Postgres::USES_PG should be used instead.

* Many more internal Sequel constants have been deprecated.

= New Features

* The Model#to_json and Dataset#to_json methods in the json_serializer
  plugin now support a block.  This block is called with the
  hash/array that would have be serialized to JSON, and the block
  should return the object to serialize.  This makes it easy to
  customize the JSON output by adding new entries, or wrapping the
  object in other object.

  The Dataset#to_json method supports an :instance_block option, which
  should be a proc that will be passed to Model#to_json.

  In order to implement this, Sequel.object_to_json now passes any
  block given to the to_json call on the object.  If you are
  overriding Sequel.object_to_json, you are responsible for making
  sure the block is passed appropriately.

* The association_pks plugin now supports a
  :association_pks_use_associated_table association option for
  many_to_many associations.  If this option is used, instead of just
  looking at the join table, the association_pks getter will get the
  primary keys from the associated table.  This can be useful if the
  association's right_primary_key does not match the associated
  model's primary key, and you are interested in the primary keys of
  the associated objects.  If this option is used, no association_pks
  setter method is created.

* Dataset#as_hash has been added as a replacement to #to_hash.
  If you want, you can now undef_method :to_hash and use
  as_hash and things will work.  Doing so can work around
  problems when using keyword argument splats in ruby 2.0+.
  For example:

    def foo(*a, **b)
    end

    foo(City.order(:id))
  
  results in foo being called with a being [] and b being
  City.order(:id).to_hash, which is unexpected and undesired
  behavior.  If you want to use keyword argument splats or other
  places where ruby will call to_hash implicitly if it is defined,
  using undef_method :to_hash is recommended.

* A Database#add_conversion_proc method has been added on
  PostgreSQL.  This method takes a type OID and either a block
  or a callable argument to use as the conversion proc for the
  type OID.

* The following adapter or database specific Database accessors
  have been added for changing settings on a per-Database basis:

  * convert_smallint_to_bool (ibmdb adapter)
  * default_charset (MySQL)
  * default_collate (MySQL)
  * default_engine (MySQL)
  * use_clob_as_blob (DB2)

* A Dataset#with_convert_types method has been added to the jdbc
  adapter, for returning a modified dataset with the convert_types
  setting changed.

= Other Improvements

* Using the postgres adapter with pg 0.21.0 no longer results in
  deprecation warnings.

* When using the class_table_inheritance plugin and using a direct
  subclass of the parent class that does not use a separate table,
  as well as using the :alias option, the insert SQL used is now
  correct.  Previously, it attempted to insert into a subquery,
  which is not valid SQL.

  Additionally, the dataset for such a model no longer uses a
  subquery, since there is no reason to do so as there is no join.

* Model.skip_auto_validations(:not_null) in the auto_validations
  plugin now skips not null checks for columns with default
  values, in addition to skipping not null checks for columns
  without default values.

* The static_cache plugin now supports the options hash argument
  to to_hash and to_hash_groups.  Currently, it only supports
  the :hash option, since the :all option doesn't make sense.

* When touching associations in the touch plugin, clear the
  cached association, because otherwise the cached values will
  be stale.

* The validation_class_methods plugin no longer requires the
  blank extension.

* The validation_helpers plugin methods that support the
  :allow_blank option now work correctly if the blank extension
  is not loaded.

* Loading the column_conflicts plugin into a model a second time
  no longer removes existing column conflict settings.

* On SQLite 3.8.8+, indexes automatically created from unique
  constraints are now included in Database#indexes output.

* On SQLite 3.8.8+, partial indexes are now excluded from
  Database#indexes output.

* Database#indexes on PostgreSQL 9.5+ now uses a simpler query
  with the array_position function.

* Database#foreign_key_list on PostgreSQL now uses a single
  query instead of two queries, and also uses the array_position
  function on PostgreSQL 9.5+ to simplify the queries.

* On PostgreSQL and Derby, when calling Database#create_table
  with the :ignore_index_errors option inside a transaction,
  a savepoint is used around each index creation so that an
  index error does not cause the entire transaction to fail.
  A savepoint is also used on Microsoft SQL Server, but it
  appears that Microsoft SQL Server rolls back the entire
  transaction if CREATE INDEX returns an error, instead of
  just rolling back to the savepoint.

* Encoding is now preserved when parsing PostgreSQL arrays in the
  pg_array extension.

* Database#copy_table in the postgres adapter now does not
  hide the underlying exception if an exception is raised
  during processing.

* Database#copy_into in the jdbc/postgresql adapter now does not
  hide the underlying exception if an exception is raised
  during processing.

* Database#copy_into in the jdbc/postgresql adapter now
  respects the :server option for using a specific shard.

* Calling #reset_conversion_procs on Database instance that
  uses the pg_hstore extension now results in the hstore
  type still being parsed.  Previously, the hstore conversion
  proc would be dropped.

* The postgres adapter no longer monkey-patches postgres-pr if it
  uses that as the driver.

* Multiple thread-safety issues in the mock adapter have been
  fixed.

* Thread safety issues when simultaneously loading multiple
  adapters that access PostgreSQL have been fixed.

* Hash allocations have been reduced in the csv_serializer,
  json_serializer, and xml_serializer plugins.

* The deprecated Sequel::Model::ANONYMOUS_MODEL_CLASSES
  constant is now correctly populated with classes created
  by Sequel::Model().  This was broken in starting in
  Sequel 4.45.0.

= Backwards Compatibility

* The pg_array_associations plugin now loads the pg_array
  extension into the Database instance if it is not already
  loaded.  This can break cases where the pg_array_associations
  plugin is used on a non-PostgreSQL database.

* Support for using the old postgres driver has been removed
  from the postgres adapter.  The postgres adapter now only
  supports pg and postgres-pr.

* When the postgres-pr driver is being used by the postgres
  adapter, connecting to a database is only allowed if
  standard strings are being forced (the default).