sequel 5.60.1 → 5.81.0

data/doc/release_notes/5.63.0.txt

@@ -0,0 +1,33 @@
+= New Features
+
+* On Ruby 3.2, the pool_class: :timed_queue Database option can now
+  be used to use an alternative connection pool that stores
+  connections in a queue, and uses the new Queue#pop :timeout option
+  in Ruby 3.2 to implement the pool timeout.  This new connection
+  pool is simpler than the default connection pool.  It is not yet
+  the default connection pool on Ruby 3.2, but it may become the
+  default in a later version.  Users of Ruby 3.2 are encouraged to
+  try out the pool_class: :timed_queue Database option and provide
+  feedback on how it works in their application.
+
+= Other Improvements
+
+* The tactical_eager_loading plugin now works in combination with the
+  single_table_inheritance and class_table_inheritance plugins, when
+  loading an association only defined in a specific subclass.
+  Previously, eager loading would be skipped in such a case.  Now,
+  an eager load will be attempted for all instances supporting the
+  association.
+
+* The validate_associated plugin now avoids database type errors for
+  non-integer association keys.  In cases where the associated object
+  doesn't have a value for the associated key, and the current object
+  does not have a key value that can be set in the associated object,
+  validation errors in the associated object related to the associated
+  key will be ignored.
+
+* Thread-keyed connection pool hashes now use compare_by_identity for
+  better performance.
+
+* The JRuby workaround in the named_timezones extension is no longer
+  used on JRuby 9.3.9.0+, as JRuby fixed the related bug.

data/doc/release_notes/5.64.0.txt

@@ -0,0 +1,50 @@
+= New Features
+
+* A :graph_use_association_block association option has been added,
+  which makes eager_graph use the association block (as eager does),
+  generally resulting in a JOIN to a subquery:
+
+    Artist.one_to_many :tracks, graph_use_association_block: true do |ds|
+      ds.where(foo: 3)
+    end
+    Artist.eager_graph(:tracks)
+    # SELECT albums.id, tracks.id AS tracks_id, tracks.album_id
+    # FROM albums
+    # LEFT OUTER JOIN (SELECT * FROM tracks WHERE (foo = 3)) AS tracks
+    # ON (tracks.album_id = albums.id)
+
+  Assuming that the database can optimize the query correctly, using
+  the :graph_use_association_block option is probably simpler than
+  than using other :graph_* options to duplicate the conditions added
+  by the association block.
+
+* Numeric/Decimal column schema entries now include :min_value and
+  :max_value entries on most databases, indicating the minimum and
+  maximum values supported for the column. Similar to the support
+  for integer columns added in 5.62.0, this allows the
+  auto_validations plugin to automatically validate the values of
+  the columns are in the allowed range.
+
+= Other Improvements
+
+* many_through_{one,many} associations now support eager_graph
+  callbacks.
+
+* The :db_type column schema entries on SQLAnywhere now include
+  precision/scale information, to work with the numeric/decimal
+  column min_value/max_value support.
+
+* The oracle adapter now includes a :column_size column schema
+  entry containing the precision of the columns, to work with the
+  numeric/decimal column min_value/max_value support.
+
+= Backwards Compatibility
+
+* The private Database#column_schema_integer_min_max_values method
+  added in 5.62.0 now takes a column schema hash instead of a
+  database type string.
+
+* Code that previously looked at the :db_type column schema entry on
+  SQLAnywhere should be updated to look at the :domain_name entry, and
+  code that looked at the :domain_name_with_size entry should be
+  updated to look at the :db_type entry.

data/doc/release_notes/5.65.0.txt

@@ -0,0 +1,21 @@
+= Improvements
+
+* The pg_auto_parameterize extension now uses a modified placeholder
+  literalizer for speeding up the generation of SQL queries in the same
+  cases where a standard dataset would use a placeholder literalizer.
+  This can provide a 4% speedup for simple queries, with greater
+  speedups for more complex queries.
+
+* Database#indexes now returns indexes for partitioned tables on
+  PostgreSQL 11+.
+
+* MySQL versions not supporting CHECK constraints no longer include
+  :min_value/:max_value schema entries for decimal/numeric columns.
+
+= Backwards Compatibility
+
+* The Dataset::PlaceholderLiterlizer::Record.loader API has changed,
+  it now accepts the Dataset::PlaceholderLiterlizer class to use as
+  the first argument.  This makes it easier to create
+  Dataset::PlaceholderLiterlizer subclasses, such as the one now used
+  by the pg_auto_parameterize extension.

data/doc/release_notes/5.66.0.txt

@@ -0,0 +1,24 @@
+= Improvements
+
+* Dataset#empty? now correctly handles datasets using custom SQL or
+  Dataset#values where the first value in the first row is NULL.
+
+* Dataset#count without an argument or block now works correctly on
+  Microsoft SQL Server when using custom SQL that uses ORDER BY.
+
+* Dataset#count now works correctly for datasets using Dataset#values.
+
+* Sequel now recognizes an additional SQLite constraint violation
+  error that occurs with recent versions of amalgalite.
+
+* Dataset#values will now raise an exception when called with an empty
+  array. Previously, an exception would not be raised until the query
+  was sent to the database.
+
+= Backwards Compatibility
+
+* The changes to make Dataset#empty? and #count work with custom SQL
+  on Microsoft SQL Server now result in running the custom SQL, which
+  could result in worse performance than in previous versions. You can
+  wrap such datasets with Dataset#from_self manually to restore the
+  previous behavior.

data/doc/release_notes/5.67.0.txt

@@ -0,0 +1,32 @@
+= New Features
+
+* A set_literalizer extension has been added, for treating Set
+  instances in datasets similar to Array instances:
+
+    DB.extension :set_literalizer
+    DB[:table].where(column: Set.new([1, 2, 3]))
+    # SELECT FROM table WHERE (column IN (1, 2, 3))
+
+= Improvements
+
+* Sequel now avoids the use of singleton classes for datasets on Ruby
+  2.4+, instead creating a regular subclass whenever a dataset would
+  be extended via #extension or #with_extend.  This significantly
+  improves performance, up to 20-40% for common dataset usage,
+  because it avoids creating new singleton classes for every dataset
+  clone, and it allows for cached method lookup.
+
+* Database#tables and #views now support a :qualify option on Microsoft
+  SQL Server to returned qualified identifiers.
+
+* The schema_dumper extension can now dump tables in non-default schemas
+  when using Microsoft SQL Server.
+
+* The schema_dumper extension now correctly dumps string column sizes
+  when using Microsoft SQL Server.
+
+= Backwards Compatibility
+
+* Calling Sequel::Dataset.register_extension where the second argument
+  is not a module now issues a deprecation warning.  Support for this
+  will be removed in Sequel 6.

data/doc/release_notes/5.68.0.txt

@@ -0,0 +1,61 @@
+= New Features
+
+* On PostgreSQL 9.6+, date, timestamp, and timestamptz columns now
+  have min_value and max_value column schema entries, allowing the
+  auto_validations plugin to automatically enforce minimum and
+  maximum values for these column types, raising a validation error
+  before saving, instead of database error when the query is sent
+  to the database.
+
+* A validation_helpers_generic_type_messages plugin has been added,
+  which improves the default type validation error messages in
+  validation_helpers.  This change was not made directly to
+  validation_helpers for backwards compatibility reasons, but will
+  probably become the default behavior in Sequel 6.  Some examples
+  of the improvements:
+
+    # :blob type
+    # validation_helpers default: "value is not a valid sequel::sql::blob"
+    # with this plugin: "value is not a blob"
+
+    # :boolean type
+    # validation_helpers default: "value is not a valid trueclass or falseclass"
+    # with this plugin: "value is not true or false"
+
+    # :datetime type
+    # validation_helpers default: "value is not a valid time or datetime"
+    # with this plugin: "value is not a valid timestamp"
+
+    # custom/database-specific types
+    # validation_helpers default: "value is not a valid sequel::class_name"
+    # with this plugin: "value is not the expected type"
+
+= Improvements
+
+* The constraint_validations plugin no longer raises validation
+  errors for missing columns that have a default value. If a column
+  is missing but has a default value, we can assume the default
+  value is valid.  Additionally, the constraint validations are now
+  based on the underlying column value and not any deserialized
+  value, so that the validation matches the constraint.
+
+* On PostgreSQL, normal type name parsing is skipped for enum,
+  array, composite, range, and multirange types, ensuring that
+  such types will not be treated incorrectly based on their
+  type name.
+
+* The pg_extended_date_support extension now works correctly with
+  the jdbc/postgresql adapter if there is already an entry in the
+  oid_convertor_map for the timestamp and timestamptz types.
+
+= Backwards Compatibility
+
+* Database#schema_column_type is no longer called for all columns
+  on PostgreSQL when parsing schema. Enum, array, composite, range,
+  and multirange types each have their own separate type parsing
+  method.  So if you were overriding Database#schema_column_type to
+  implement custom column schema parsing, you may need to adjust
+  your code.
+
+* The Sequel::Postgres::ExtendedDateSupport::DATE_YEAR_1 constant
+  has been removed.

data/doc/release_notes/5.69.0.txt

@@ -0,0 +1,26 @@
+= New Features
+
+* An adapter has been added for the trilogy MySQL driver. One large
+  advantage over mysql2 is that trilogy does not require any MySQL
+  client libraries installed on the machine. The trilogy adapter
+  has basically the same issues/skipped specs as the mysql2 adapter,
+  but it also does not support an application_timezone different
+  than the database_timezone.
+
+* Model dataset modules now have a model accessor, allowing for
+  code such as:
+
+    class Foo < Sequel::Model
+      dataset_module do
+        where :kept, Sequel[model.table_name][:discarded_at] => nil
+      end
+    end
+
+= Improvements
+
+* The mysql adapter now works with ruby-mysql 4 (the pure-ruby
+  MySQL driver).  Note that multi-results support does not work
+  with ruby-mysql 4 (it doesn't work with mysql2, trilogy, or
+  other Sequel adapters in general).
+
+* Warnings for unsupported flags are now avoided on ruby-mysql 3.

data/doc/release_notes/5.70.0.txt

@@ -0,0 +1,35 @@
+= New Features
+
+* A sharded_timed_queue connection pool has been added. This offers
+  most of the same features as the sharded_threaded connection pool,
+  but uses the new Queue#pop :timeout features added in Ruby 3.2 to
+  allow for a simpler and possibly faster and more robust
+  implementation.
+
+* If a :pool_class option is not specified when creating a Database,
+  Sequel will now look at the SEQUEL_DEFAULT_CONNECTION_POOL
+  environment variable to determine the connection pool class to use.
+  This allows you to set SEQUEL_DEFAULT_CONNECTION_POOL=timed_queue
+  on Ruby 3.2 to test with the timed_queue connection pool without
+  making any code changes.  If the :servers Database option is given,
+  Sequel will automatically use the sharded version of the connection
+  pool specified by SEQUEL_DEFAULT_CONNECTION_POOL.
+
+= Other Improvements
+
+* The connection_validator, connection_expiration, and
+  async_thread_pool extensions now work with the timed_queue and
+  sharded_timed_queue connection pools.
+
+* The sharded_threaded connection pool now disconnects connections
+  for all specified servers instead of just the last specified server
+  when using remove_server.
+
+* The static_cache plugin now recognizes when the forbid_lazy_load
+  plugin is already loaded, and does not return instances that
+  forbid lazy load for methods that return a single object, such as
+  Database.{[],cache_get_pk,first}.
+
+* Sequel now displays an informative error message if attempting to
+  load the connection_validator or connection_expiration extensions
+  when using the single threaded connection pool.

data/doc/release_notes/5.71.0.txt

@@ -0,0 +1,21 @@
+= New Features
+
+* A pg_xmin_optimistic_locking plugin has been added.  This plugin
+  uses PostgreSQL's xmin system column to implement optimistic
+  locking.  The xmin system column is automatically updated whenever
+  the database row is updated.  You can load this plugin into a
+  base model and have all models that subclass from it use optimistic
+  locking, without needing any user-defined lock columns.
+
+= Other Improvements
+
+* set_column_allow_null is now a reversible migration method inside
+  alter_table blocks.
+
+* The use of ILIKE no longer forces the ESCAPE clause on PostgreSQL,
+  which allows the use of ILIKE ANY and other constructions.  There
+  is no need to use the ESCAPE clause with ILIKE, because the value
+  Sequel uses is PostgreSQL's default.
+
+* The xid PostgreSQL type is now recognized as an integer type in the
+  jdbc/postgresql adapter.

data/doc/release_notes/5.72.0.txt

@@ -0,0 +1,33 @@
+= New Features
+
+* A pg_auto_parameterize_in_array extension has been added, which
+  handles conversion of IN/NOT IN to = ANY or != ALL for more types.
+  The pg_auto_parameterize extension only handles integer types by
+  default, because other types require the pg_array extension. This
+  new extension adds handling for Float, BigDecimal, Date, Time,
+  DateTime, Sequel::SQLTime, and Sequel::SQL::Blob types.  It can
+  also handle String types if the :treat_string_list_as_text_array
+  Database option is present, using the text type for that. Handling
+  String values as text is not the default because that may cause
+  issues for some queries.
+
+= Other Improvements
+
+* The defaults_setter plugin now does a deep copy of database
+  default values that are hash/array or delegates to hash/array.
+  This fixes cases where the database default values are mutated.
+
+* Sequel now correctly handles infinite and NaN float values used
+  inside PostgreSQL array bound variables.
+
+* The data in the cache files used by the schema_caching and
+  index_caching extensions and static_cache_cache and
+  pg_auto_constraint_validations plugins are now sorted before the
+  cache file is saved, increasing consistency between runs.
+
+* bigdecimal has been added as a dependency.  bigdecimal is currently
+  a default gem in Ruby from 1.9 to 3.2, but it will move to a
+  bundled gem in Ruby 3.4, and there will be warnings in Ruby 3.3
+  for cases that will break in Ruby 3.4.  Adding bigdecimal as a
+  dependency should avoid warnings when using bundler in Ruby 3.3,
+  and should avoid errors in Ruby 3.4.

data/doc/release_notes/5.73.0.txt

@@ -0,0 +1,66 @@
+= New Features
+
+* A paged_operations plugin has been added, which adds support for
+  paged_datasets, paged_update, and paged_delete dataset methods.
+  This methods are designed to be used on large datasets, to split 
+  a large query into separate smaller queries, to avoid locking the
+  related database table for a long period of time.
+  paged_update and paged_delete operate the same as update and delete,
+  returning the number of rows updated or deleted. paged_datasets yields
+  one or more datasets representing subsets of the receiver, with the
+  union of all of those datasets comprising all records in the receiver:
+
+    Album.plugin :paged_operations
+
+    Album.where{name > 'M'}.paged_datasets{|ds| puts ds.sql}
+    # Runs: SELECT id FROM albums WHERE (name <= 'M') ORDER BY id LIMIT 1 OFFSET 1000
+    # Prints: SELECT * FROM albums WHERE ((name <= 'M') AND ("id" < 1002))
+    # Runs: SELECT id FROM albums WHERE ((name <= 'M') AND (id >= 1002)) ORDER BY id LIMIT 1 OFFSET 1000
+    # Prints: SELECT * FROM albums WHERE ((name <= 'M') AND ("id" < 2002) AND (id >= 1002))
+    # ...
+    # Runs: SELECT id FROM albums WHERE ((name <= 'M') AND (id >= 10002)) ORDER BY id LIMIT 1 OFFSET 1000
+    # Prints: SELECT * FROM albums WHERE ((name <= 'M') AND (id >= 10002))
+
+    Album.where{name <= 'M'}.paged_update(:updated_at=>Sequel::CURRENT_TIMESTAMP)
+    # SELECT id FROM albums WHERE (name <= 'M') ORDER BY id LIMIT 1 OFFSET 1000
+    # UPDATE albums SET updated_at = CURRENT_TIMESTAMP WHERE ((name <= 'M') AND ("id" < 1002))
+    # SELECT id FROM albums WHERE ((name <= 'M') AND (id >= 1002)) ORDER BY id LIMIT 1 OFFSET 1000
+    # UPDATE albums SET updated_at = CURRENT_TIMESTAMP WHERE ((name <= 'M') AND ("id" < 2002) AND (id >= 1002))
+    # ...
+    # SELECT id FROM albums WHERE ((name <= 'M') AND (id >= 10002)) ORDER BY id LIMIT 1 OFFSET 1000
+    # UPDATE albums SET updated_at = CURRENT_TIMESTAMP WHERE ((name <= 'M') AND (id >= 10002))
+
+    Album.where{name > 'M'}.paged_delete
+    # SELECT id FROM albums WHERE (name > 'M') ORDER BY id LIMIT 1 OFFSET 1000
+    # DELETE FROM albums WHERE ((name > 'M') AND (id < 1002))
+    # SELECT id FROM albums WHERE (name > 'M') ORDER BY id LIMIT 1 OFFSET 1000
+    # DELETE FROM albums WHERE ((name > 'M') AND (id < 2002))
+    # ...
+    # SELECT id FROM albums WHERE (name > 'M') ORDER BY id LIMIT 1 OFFSET 1000
+    # DELETE FROM albums WHERE (name > 'M')
+
+* A Dataset#transaction :skip_transaction option is now support to
+  checkout a connection from the pool without opening a transaction.  This
+  makes it easier to handle cases where a transaction may or not be used
+  based on configuration/options.  Dataset#import and Dataset#paged_each
+  now both support the :skip_transaction option to skip transactions.
+
+* Dataset#full_text_search now supports the to_tsquery: :websearch option
+  on PostgreSQL 11+, to use the websearch_to_tsquery database function.
+
+* The Sequel::MassAssignmentRestriction exception now supports model
+  and column methods to get provide additional information about the
+  exception.  Additionally, the exception message now includes information
+  about the model class.
+
+= Other Improvements
+
+* The ibmdb and jdbc/db2 adapter now both handle disconnect errors
+  correctly, removing the related connection from the pool.
+
+* Dataset#import no longer uses an explicit transaction if given a dataset
+  value, as in that case, only a single query is used.
+
+* The column_encryption plugin no longer uses the base64 library.  The
+  base64 library is moving from the standard library to a bundled gem
+  in Ruby 3.4, and this avoids having a dependency on it.

data/doc/release_notes/5.74.0.txt

@@ -0,0 +1,45 @@
+= New Features
+
+* Sequel.migration blocks now support a revert method, which reverts
+  the changes in the block on up, and applies them on down.  So if
+  you have a migration such as:
+
+    Sequel.migration do
+      change do
+        create_table :table do
+          # ...
+        end
+      end
+    end
+
+  and you later want to add a migration that drops the table, you
+  can use:
+
+    Sequel.migration do
+      revert do
+        create_table :table do
+          # ...
+        end
+      end
+    end
+
+  This will drop the table when migrating up, and create a table 
+  with the given schema when migrating down.
+
+* is_json and is_not_json methods have been added to the pg_json_ops
+  extension, for the IS [NOT] JSON operator supported in PostgreSQL
+  16+.  These were previously added in Sequel 5.59.0, and removed
+  in Sequel 5.61.0 as support was removed in PostgreSQL 15 beta 4.
+  PostgreSQL 16 shipped with support for them, so support has been
+  recommitted to Sequel.
+
+= Other Improvements
+
+* SQLite generated columns now show up in Database#schema when using
+  SQLite 3.37+.
+
+* Sequel now attempts to avoid an infinite loop in pathlogical cases
+  in the jdbc adapter, where the exception cause chain has a loop.
+  Additionally, if an exception is already recognized as a disconnect,
+  or an exception already responds to a getSQLState method, Sequel no
+  longer looks at the causes of the exception.

data/doc/release_notes/5.75.0.txt

@@ -0,0 +1,35 @@
+= New Features
+
+* Database#{defer,immediate}_constraints methods have been added on
+  PostgreSQL for changing handling of deferrable constraints inside
+  a transaction. defer_constraints sets deferrable constraints to
+  be deferred (not checked until transaction commit), and
+  immediate_constraints sets deferrable constraints to be checked
+  as part of the related query, and any already deferred constraint
+  checks to be applied immediately. You can pass the :constraints
+  option to only apply the changes to specific constraints.
+
+* TimestampMigrator.run_single has been added, to migrate a single
+  migration up or down.
+
+= Other Improvements
+
+* INSERT RETURNING is now supported on MariaDB 10.5+, and used
+  automatically when saving new model objects. Note that this
+  is not supported when using the jdbc adapter, because the
+  jdbc-mysql driver doesn't support it.  A jdbc/mariadb adapter
+  could be added, as it's likely recent versions of the
+  jdbc-mariadb driver would support it, but the jdbc-mariadb gem 
+  hasn't been updated in over 4 years.  Talk to the jdbc-mariadb
+  gem maintainers if you want to use this feature with the jdbc
+  adapter.
+
+* The Dataset#paged_each optimization in the postgres adapter
+  now respects the :skip_transaction option, making it the
+  same as the :hold option.  Note that this has effects beyond
+  just skipping the transaction, but non-HOLD cursors are only
+  supported inside transactions.
+
+* The any_not_empty? extension's Dataset#any? method now supports
+  an argument, passing it to Enumerable#any? (which has supported
+  an argument since Ruby 2.5).

data/doc/release_notes/5.76.0.txt

@@ -0,0 +1,86 @@
+= New Features
+
+* An auto_cast_date_and_time extension has been added, which will
+  automatically cast date and time values using SQL standard functions.
+  This makes sure the database will treat the value as a date, time,
+  or timestamp, instead of treating it as a string or unknown type:
+
+    DB.get(Date.today).class
+    # SELECT '2024-01-01' AS v LIMIT 1
+    String
+
+    DB.extension(:auto_cast_date_and_time)
+    DB.get(Date.today).class
+    # SELECT DATE '2024-01-01' AS v LIMIT 1
+    Date
+
+  This was already Sequel's default behavior on adapters that required
+  it.  This extension is usable on PostgreSQL and MySQL.  It is not
+  usable on SQLite (no date/time types) or Microsoft SQL Server (no
+  support for the SQL standard conversion syntax).
+
+  This extension can break code that currently works. If using it on
+  PostgreSQL, it will cast the values to TIMESTAMP, not TIMESTAMP
+  WITH TIME ZONE, which can break code that depended on an implicit
+  conversion to TIMESTAMP WITH TIME ZONE.  The pg_timestamptz
+  extension integrates with the the auto_cast_date_and_time extension
+  and will implicitly cast Time/DateTime to TIMESTAMP WITH TIME ZONE.
+
+* The sqlite adapter now supports a :cached value for the
+  :setup_regexp_function Database option, which will cache regexp
+  values instead of creating a new regexp per value to compare.  This
+  is much faster when using a regexp comparison on a large dataset,
+  but can result in a memory leak if using dynamic regexps. You can
+  also provide a Proc value for the :setup_regexp_function option,
+  which will be passed both the regexp source string and the database
+  string to compare, and should return whether the database string
+  matches the regexp string.
+
+* The rcte_tree plugin now supports a :union_all option, which can
+  be set to false to use UNION instead of UNION ALL in the recursive
+  common table expression.
+
+= Other Improvements
+
+* Time/DateTime/SQLTime literalization speed has more than doubled
+  compared to the previous version.  The internal code is also much
+  simpler, as the speedup resulted from removing multiple abstraction
+  layers that mostly existed for Ruby 1.8 support.
+
+* Database#table_exists? on PostgreSQL now handles lock or statement
+  timeout errors as evidence the table exists.
+
+* The round_timestamps extension now correctly rounds SQLTime values
+  on Microsoft SQL Server (the only database Sequel supports where
+  time precision is different than timestamp precision).
+
+* Fractional times and timestamps are now supported on SQLAnywhere,
+  except for time values when using the jdbc adapter due to a
+  limitation in the JDBC sqlanywhere driver.
+
+* Database#tables and #views on PostgreSQL now supports
+  SQL::Identifier values for the :schema option.
+
+* The named_timezones extension now works around a bug in DateTime.jd
+  on JRuby.
+
+= Backwards Compatibility
+
+* Time/DateTime/SQLTime literalization internals have changed.
+  If you are using an external adapter and the external adapter
+  overrides or calls any of the following methods:
+
+  * requires_sql_standard_datetimes?
+  * supports_timestamp_usecs?
+  * supports_timestamp_timezones?
+  * timestamp_precision
+  * sqltime_precision
+
+  then the adapter may need to be updated to support Sequel 5.76.0.
+  Additionally, if the adapter uses %N or %z in
+  default_timestamp_format, it may need to be updated. Adapters
+  should now just override default_timestamp_format and/or
+  default_time_format methods as appropriate for the database.
+
+* The Dataset#format_timestamp_offset private method has been
+  removed.

data/doc/release_notes/5.77.0.txt

@@ -0,0 +1,63 @@
+= New Features
+
+* A transaction_connection_validator extension has been added.  This
+  extension allows for transparently switching to a new connection if
+  a disconnect error is raised while trying to start a transaction, as
+  long as a connection was not already checked out from the pool
+  when the transaction method was called.  Transparent reconnection
+  is safe in this case, since no user code is retried.
+
+  This extension can have lower overhead than the
+  connection_validator extension if that is configured to check for
+  validity more often than the default of one hour.  However, it
+  only handles cases where transactions are used.  It can detect
+  disconnects that would not be detected by default with the
+  connection_validator extension, since that extension defaults to
+  only checking validity if the connection has not been used in the
+  last hour.
+
+* Sequel now supports a create_table :without_rowid option on SQLite,
+  to create a table WITHOUT ROWID, for better performance in some
+  cases. Users are encouraged to read the SQLite documentation on
+  WITHOUT ROWID before using this option.
+
+* The sqlite adapter now supports a :regexp_function_cache option, if
+  the :setup_regexp_function option is set to :cached.  The
+  :regexp_function_cache option should be a Proc (returning a cache
+  object to use), or a class.  It's possible to use
+  ObjectSpace::WeakKeyMap as the value of the option on Ruby 3.3+
+  to avoid the memory leaks that are possible when using
+  :setup_regexp_function option :cached value with dynamic regexps.
+
+* The duplicate_columns_handler extension now supports specifying
+  the on_duplicate_columns option as a connection string parameter.
+
+= Other Improvements
+
+* The list plugin now honors the :top option for the position when
+  adding the first item to the list, instead of always using 1.
+
+* Regexp matches on SQLite are now faster on Ruby 2.4+, using
+  Regexp#match?.
+
+* The uniqueness validation in the validation_helpers plugin now
+  uses empty? instead of count == 0, for better performance.
+
+* On Ruby 3.4+, Sequel uses the timed_queue connection pool instead
+  of the threaded connection pool by default.  This should make it
+  so no existing applications are affected by the default switch.
+  This should hopefully allow ample testing of the timed_queue
+  connection pool.  At some point in the future, if no problems
+  are repoted, Sequel will likely switch to using the timed_queue
+  connection pool by default on Ruby 3.2+.
+
+= Backwards Compatibility
+
+* Sequel now warns by default if using eager_graph/association_join
+  with an association that uses a block, in the cases where the
+  block would be ignored and there are no appropriate graph options
+  set. In Sequel 6, this warning will be turned into an exception.
+  It is recommended that users use the auto_restrict_eager_graph
+  plugin to turn this into an exception now, or use the
+  :graph_use_association_block option so that the block is not
+  ignored when graphing.