sequel 4.36.0 → 5.61.0

data/doc/release_notes/4.10.0.txt

@@ -1,226 +0,0 @@
-= Performance Enhancements
-
-* Dataset literalization for simple datasets is now faster by
-  creating a per-adapter SQL literalization method instead of
-  having all adapters share a generic method with higher overhead.
-  Sequel.split_symbol now caches results globally. Symbol
-  literalization is now cached per Database.
-
-  Combining these three optimizations, here are the performance
-  increases compared to 4.9.0 for a couple example datasets:
-
-    ds1 = DB[:a]
-    ds2 = DB[:a].select(:a, :b).where(:c=>1).order(:d, :e)
-
-            .sql     .all (1 row)
-    ds1     140%     11%
-    ds2     187%     32%
-
-* Regular association loading now uses a placeholder literalizer
-  in most cases, for up to an 85% improvement when loading
-  simple associations.
-
-* Eager loading associations using Dataset#eager now uses a
-  placeholder literalizer in most cases, for up to a
-  20% improvement when eager loading simple associations.
-
-* Eager loading associations with limits using Dataset#eager now
-  uses a UNION-based strategy by default.  After extensive
-  testing, this was found to be the fastest strategy if the
-  key columns are indexed.  Unfortunately, it is a much slower
-  strategy if the key columns are not indexed.  You can override
-  the default UNION strategy by using the :eager_limit_strategy
-  association option.
-
-  On some databases, execution time of UNION queries with n subqueries
-  increases faster than O(n).  Also, there are limits on the number of
-  subqueries supported in a single UNION query.  Sequel chooses a
-  default limit of 40 subqueries per UNION query.  You can increase
-  this via the :subqueries_per_union association option.
-
-* Dataset#import and #multi_insert can now insert multiple rows
-  in a single query on H2, HSQLDB, Derby, SQLAnywhere, CUBRID,
-  SQLite, Oracle, DB2, and Firebird, which should be significantly
-  faster than previous versions that issued a separate INSERT query
-  per row.
-
-* The many_to_many setter method in the association_pks plugin now
-  uses Dataset#import to insert many rows at once, instead of using
-  a seperate query per insert.
-
-* The jdbc adapter's type conversion has been rewritten to be
-  more similar to the other adapters, setting up the type
-  conversion procs before iterating over results.  This increases
-  performance up to 20%.
-
-* The jdbc/oracle adapter now defaults to a fetch_size of 100,
-  similar to the oci8-based oracle adapter, significantly improving
-  performance for large datasets.
-
-= New Features
-
-* Database#transaction now supports an :auto_savepoint option.  This
-  option makes it so that transactions inside the transaction block
-  automatically use savepoints unless they use the :savepoint=>false
-  option.  This should make testing transactional behavior easier.
-
-* Model.prepared_finder has been added.  This has an API similar to
-  Model.finder, but it uses a prepared statement instead of a
-  placeholder literalizer.  It is less flexible than Model.finder
-  as prepared statements have fixed SQL, but it may perform better.
-
-* Common table expressions (WITH clauses) are now supported on SQLite
-  3.8.3+.
-
-* :correlated_subquery has been added as an eager_graph and filter by
-  association limit strategy for one_to_one and one_to_many
-  associations.  In certain cases it was found that this is faster
-  than the :window_function limit strategy.  It is the default
-  filter by associations limit strategy on databases that do not
-  support window functions.
-
-  Filtering by limited associations using a correlated subquery
-  strategy does not work in all cases, but it should handle most
-  cases correctly.
-
-* The prepared_statement_associations plugin now handles
-  one_through_one and one_through_many associations.
-
-* Sequel now emulates support for offsets without limits on MySQL,
-  SQLite, H2, SQLAnywhere, and CUBRID.
-
-* In the jdbc adapter, the Database#fetch_size accessor and
-  :fetch_size option can be used to automatically set the JDBC
-  fetch size for JDBC Statement objects created by the database.
-
-* Dataset#with_fetch_size has been added to jdbc adapter datasets,
-  setting the fetch size to use on ResultSets generated by the
-  dataset.  This generally has the effect of overriding the
-  Database fetch_size setting.
- 
-* On MySQL 5.6.5+, Sequel supports a :fractional_seconds Database
-  option, which will use fractional seconds for timestamp values,
-  and have the schema modification code create timestamp columns
-  that accept fractional timestamps by default.
-
-* Database#call_mssql_sproc on Microsoft SQL Server now handles named
-  parameters:
-
-    DB.call_mssql_sproc(:sproc_name, :args => {
-      'input_arg1_name' => 'input arg1 value',
-      'input_arg2_name' => 'input arg2 value',
-      'output_arg_name' => [:output, 'int', 'result key name']
-    })
-
-* Database#drop_view now supports an :if_exists option on SQLite,
-  MySQL, H2, and HSQLDB.
-
-* Database#drop_table now supports an :if_exists option on HSQLDB.
-
-* A :filter_limit_strategy association option has been added, for
-  choosing the strategy that will be used when filtering/excluding by
-  associations with limits.  For backwards compatibility, Sequel will
-  fallback to looking at the :eager_limit_strategy option.
-
-* A :server_version Database option is now supported on Microsoft SQL
-  Server, which will use the value given instead of querying for it.
-
-= Other Improvements
-
-* Dataset::PlaceholderLiteralizer arguments are how handled
-  correctly when emulating offsets via the row_number window function
-  on DB2, MSSQL <=2012, and Oracle.
-
-* Dataset::PlaceholderLiteralizer now handles DelayedEvaluation
-  objects correctly.
-
-* Offset emulation is skipped if static SQL is used on Access,
-  DB2, and MSSQL <=2008.
-
-* Additional disconnect errors are now recognized in the postgres
-  adapter.
-
-* The :foreign_key_constraint_name option is now respected when
-  adding a foreign key column to an existing table on MySQL.
-
-* Sequel now attempts to work around a bug on MySQL 5.6+ when
-  combining DROP FOREIGN KEY and DROP INDEX in the same ALTER TABLE
-  statement.
-
-* Dataset#for_update is now respected on H2.
-
-* Timestamp with local time zone types are now returned as
-  Time/DateTime objects on jdbc/oracle.
-
-* Model.include now has the same API as Module.include.
-
-* Model#marshallable! now works correctly when using the
-  tactical_eager_loading plugin.
-
-* The pg_array_associations plugin now attempts to automatically
-  determine the correct array type to use, and explicitly casts
-  to that array type in more places.
-
-* The auto_validations plugin now handles models that select from
-  subqueries.
-
-* The association_pks plugin does no longer creates getter and setter
-  methods for one_through_one associations.
-
-* bin/sequel now uses the Sequel code in the related lib directory.
-  This makes it easier to use from a repository checkout.
-
-= Backwards Compatibility
-
-* AssociationReflection#associated_dataset now returns a joined
-  dataset for associations that require joins (e.g. many_to_many).
-  Anyone using this directly for associations that require joins
-  probably needs to update their code.
-
-* Model.associate now adds the association instance methods instead
-  of relying on the def_#{association_type} method doing so.  Anyone
-  using custom association types probably needs to update their code.
-
-* Model.eager_loading_dataset, .apply_association_dataset_opts, and
-  .def_{add_method,association_dataset_methods,remove_methods}
-  are now deprecated.
-
-* Key conditions for associations requiring joins have been moved
-  from the JOIN ON clause to the WHERE clause.  This should be
-  optimized the same by the database, but it can break tests that
-  expect specific SQL.
-
-* Dataset#_insert_sql and #_update_sql are now private instead of
-  protected.
-
-* The install/uninstall rake tasks have been removed.
-
-* Model association and association reflection internals have
-  changed significantly, if you were relying on them, you'll
-  probably need to update your code.
-
-* Database transaction internals have changed significantly, if you
-  were relying on them, you'll probably need to update your code.
-
-* Dataset literalization internals have changed significantly, with
-  the Dataset#*_clause_methods private methods being removed.
-  Custom adapters that used these methods should switch to using the
-  new Dataset.def_sql_method method.
-
-* Common table expressions are no longer enabled by default in
-  Sequel.  External adapters for databases that support common
-  table expressions should define Dataset#supports_cte?(type) to
-  return true.
-
-* Support for RETURNING is no longer determined via introspection.
-  External adapters for databases that support RETURNING should
-  define Dataset#supports_returning?(type) to return true.
-
-* The new jdbc adapter type conversion code may not be completely
-  compatible with the previous code.  The currently known case
-  where it is different is on jdbc/postgresql, when using an
-  array type where no conversion proc exists, the returned object
-  will be a ruby array containing java objects, instead of a ruby
-  array containing ruby objects.  It is recommended that
-  jdbc/postgresql users using array types use the pg_array extension
-  to avoid this issue.

data/doc/release_notes/4.11.0.txt

@@ -1,147 +0,0 @@
-= 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.
-
-= Other New Features
-
-* 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.
-
-= Other Improvements
-
-* 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.
-
-= Backwards Compatibility
-
-* 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.

data/doc/release_notes/4.12.0.txt

@@ -1,105 +0,0 @@
-= New Features
-
-* Database#schema now includes :max_length entries for string
-  columns, specifying the size of the string field.  The
-  auto_validations plugin now uses this information to
-  automatically set up max_length validations on those fields.
-
-* The Dataset join methods now support a :reset_implicit_qualifier
-  option.  If set to false, this makes the join not reset the
-  implicit qualifier, so that the next join will not consider this
-  table as the last table joined.  Example:
-
-    DB[:a].join(:b, :c=>:d).
-      join(:e, :f=>:g)
-    # SELECT * FROM a
-    # INNER JOIN b ON (b.c = a.d)
-    # INNER JOIN e ON (e.f = b.g)
-
-    DB[:a].join(:b, {:c=>:d}, :reset_implicit_qualifier=>false).
-      join(:e, :f=>:g)
-    # SELECT * FROM a
-    # INNER JOIN b ON (b.c = a.d)
-    # INNER JOIN e ON (e.f = a.g)
-
-* The Dataset cross and natural join methods now accept an options
-  hash. Example:
-
-    DB[:a].cross_join(:b, :table_alias=>:c)
-    # SELECT * FROM a CROSS JOIN b AS c
-
-* Model#set_nested_attributes has been added to the nested_attributes
-  plugin, which allows you to to set the nested_attributes options to
-  use per-call.  This is very helpful if you have multiple forms that
-  handle associated objects, but with different input fields used
-  for the associated objects depending on the form.  Example:
-
-    album.set_nested_attributes(:tracks,
-      params[:track_attributes],
-      :fields=>[:a, :b, :c])
-
-* Database#values has been added on PostgreSQL, which creates a
-  dataset that uses VALUES instead of SELECT.  Just as PostgreSQL
-  allows, you can also use orders, limits, and offsets with this
-  dataset.
-
-* A :notice_receiver option is now supported in the postgres adapter
-  if the pg driver is used.  This should be a proc, which will be
-  passed to the pg connection's set_notice_receiver method.
-
-* A Database :readonly option is now supported in the sqlite adapter,
-  which opens the database in a read-only mode, causing an error
-  if a query is issued that would modify the database.
-
-* A :before_thread_exit option has been added to
-  Database#listen_for_static_cache_updates in the
-  pg_static_cache_updater extension, allowing you to run code before
-  the created thread exits.
-
-= Other Improvements
-
-* Eager loading limited associations using a UNION now works
-  correctly when an association block is used.  This fixes a
-  regression that first occurred in 4.10.0, when the union
-  eager loader became the default eager loader.
-
-* When creating a new associated object in the nested_attributes
-  plugin, where the reciprocal association is a many_to_one
-  association, set the cached reciprocal object in the new
-  associated object before saving it.
-
-  This fixes issues when validations in the associated object
-  require access to the current object, which may not yet be
-  saved in the database.
-
-* The prepared_statements and prepared_statements_associations
-  plugins now automatically use explicit column references when
-  preparing statements.  This fixes issues on PostgreSQL when a
-  column is added to a table while a prepared statement exists
-  that selects * from the table.  Previously, all further attempts
-  to use the prepared statement will fail.
-
-  This allows you to run migrations that add columns to tables
-  while concurrently running an application that uses the
-  prepared statements plugins.  Note that many other schema
-  modifications can cause issues when running migrations
-  while concurrently running an application, but most of those
-  are not specific to usage of prepared statements.
-
-* Dataset#insert_select on PostgreSQL now respects an existing
-  RETURNING clause, and won't override it to use RETURNING *.
-
-  A similar fix was applied to the generalized prepared statements
-  support as well.
-
-* The interval parser in the pg_interval extension now supports
-  intervals with 2-10 digits for hours.  Previously, it only
-  supported using 2 digits for hours.
-
-= Backwards Compatibility
-
-* The methods and classes deprecated in 4.11.0 have been removed.
-
-* The nested_attributes internal API has changed significantly. If
-  you were calling any private nested_attributes methods, you'll
-  probably need to update your code.

data/doc/release_notes/4.13.0.txt

@@ -1,169 +0,0 @@
-= New Features
-
-* A modification_detection plugin has been added, for automatic
-  detection of in-place column value modifications.  This makes
-  it so you don't have to call Model#modified! manually when
-  changing a value in place.
-
-* A column_select plugin has been added, for automatically
-  selecting explicitly qualified columns in model datasets.
-  Example:
-
-    Sequel::Model.plugin :column_select
-    class Album < Sequel::Model
-    end
-    Album.dataset.sql
-    # SELECT albums.id, albums.name, albums.artist_id
-    # FROM albums
-
-* An insert_returning_select plugin has been added, for automatically
-  setting up RETURNING clauses for models that select explicit
-  columns.  This is useful when using the column_select or
-  lazy_attributes plugins.
-
-* A pg_enum extension has been added, for easier dealing with
-  PostgreSQL enum types.  The possible values for the type
-  are then returned in the schema hashes under the :enum_values
-  key.  It also adds create_enum, drop_enum, and add_enum_value
-  Database methods for migration support.
-
-* A round_timestamps extension has been added, for automatically
-  rounding timestamps to database supported precision when
-  literalizing.
-
-* A dataset_source_alias extension has been added, for automatically
-  aliasing datasets to their first source, instead of using t1, t2.
-  Example:
-
-    DB.from(:a, DB[:b]).sql
-    # SELECT * FROM a, (SELECT * FROM b) AS t1
-
-    DB.extension(:dataset_source_alias)
-    DB.from(:a, DB[:b]).sql
-    # SELECT * FROM a, (SELECT * FROM b) AS b
-
-* On Microsoft SQL Server, Sequel now emulates RETURNING support
-  using the OUTPUT clause, as long as only simple column references
-  are used.
-
-= Other Improvements
-
-* A regression has been fixed in the timestamps and table
-  inheritance plugins, where column values would not be
-  saved when skipping validations.  This was first broken in
-  4.11.0.
-
-* A regression has been fixed on JRuby and Rubinius when using
-  Sequel::Model(dataset) if the dataset needs to literalize a
-  symbol (and most do).  This was first broken in 4.10.0.
-
-* Primary keys are now automatically setup for models even if
-  the models select specific columns.
-
-* The lazy_attributes plugin now uses qualified columns in its
-  selection, instead of unqualified columns.
-
-* When looking up model instances by primary key, Sequel now uses a
-  qualified primary key if the model uses a joined dataset.
-
-* For associations that require joins, Sequel will now use the
-  associated model's selection directly (instead of
-  associated_table.*) if the associated model's selection consists
-  solely of qualified columns.
-
-  Among other things, this means that a many_to_many association to 
-  a model that uses lazy attributes will not eagerly load the lazy
-  attributes by default.
-
-* Model#save now uses insert_select if there is an existing
-  RETURNING clause used by the underlying dataset, even if the model
-  selects specific columns.
-
-* In Dataset#insert, aliased tables are now automatically unaliased.
-  This allows you to use a dataset with an aliased table and have
-  full SELECT/INSERT/UPDATE/DELETE support, assuming the database
-  supports aliased tables in UPDATE and DELETE.
-
-* Dataset#graph now qualifies columns correctly if the current
-  dataset is a joined dataset and it moves the current dataset to
-  a subselect.
-
-* Dataset#joined_dataset? is now a public method, and can be used to
-  determine whether the dataset uses a join, either explicitly via
-  JOIN or implicitly via multiple FROM tables.
-
-* The Dataset#unqualified_column_for helper method has been added,
-  returning the unqualified version of a possibly qualified column.
-
-* The composition and serialization plugins now support validations
-  on the underlying columns.  Previously, they didn't update the
-  underlying columns until after validations were performed.  This
-  works better when using the auto_validations plugin.
-
-* The class_table_inheritance plugin now uses JOIN ON instead of
-  JOIN USING, which makes it work on all databases that Sequel
-  supports.  Additionally, the plugin now explicitly selects
-  qualified columns from all of the tables.
-
-* The list plugin now adds an after_destroy hook that will renumber
-  rows after the current row, similar to how moving existing values
-  in the list works.
-
-* The pg_json extension is now faster when json column value is a
-  plain string, number, true, false, or nil, if the underlying json
-  library handles such values natively.
-
-* External jdbc, odbc, and do subadapters can now be loaded
-  automatically without requiring them first, assuming proper
-  support in the external subadapter.
-
-* When using create_table on MySQL, correctly handle the :key
-  option to when calling foreign_key with a column reference.
-
-* On Oracle, use all_tab_cols instead of user_tab_cols for getting
-  default values when parsing the schema.  This makes it work if the
-  user does not own the table.
-
-* On Oracle, use all_tables and all_views for Database#tables and
-  Database#views.  This works better for users with limited rights.
-
-* Additional disconnect errors are now recognized in the postgres and
-  jdbc/mysql adapters.
-
-* Sequel::Model now uses copy constructors (e.g. initialize_copy)
-  instead of overriding #dup and #clone.
-
-* The rake default task now runs plugin specs in addition to
-  core and model specs.
-
-= bin/sequel Improvements
-
-* Add the sequel lib directory to the front of the load path
-  instead of the end, fixing cases where you end up requiring an
-  old version of the sequel gem (e.g. by using sequel_pg).
-
-* Add the sequel lib directory as an absolute path, fixing cases
-  where you later change the current directory.
-
-* Require sequel later in the code, so that bin/sequel -h doesn't
-  need to require sequel, and full backtrace is not printed if
-  requiring sequel raises an error (unless -t is used).
-
-* If an exception is raised, put a newline between the exception
-  message and backtrace.
-
-* Don't allow usage of -C with any of -cdDmS.
-
-* If sequel -v is given along with a database or code string to
-  execute, print the Sequel version but also continue, similar
-  to how ruby -v works.
-
-= Backwards Compatibility
-
-* The switch from JOIN ON to JOIN USING in the
-  class_table_inheritance can break certain usage, such as querying
-  using unqualified primary key.  Users should switch to using a
-  qualified primary key instead.
-
-* Calling Dataset#returning when the underlying database does not
-  support it now raises an Error.