1.0 Changelog¶
1.0.19¶
Released: August 3, 2017oracle¶
Fixed performance regression caused by the fix for #3937 where cx_Oracle as of version 5.3 dropped the
.UNICODE
symbol from its namespace, which was interpreted as cx_Oracle’s “WITH_UNICODE” mode being turned on unconditionally, which invokes functions on the SQLAlchemy side which convert all strings to unicode unconditionally and causing a performance impact. In fact, per cx_Oracle’s author the “WITH_UNICODE” mode has been removed entirely as of 5.1, so the expensive unicode conversion functions are no longer necessary and are disabled if cx_Oracle 5.1 or greater is detected under Python 2. The warning against “WITH_UNICODE” mode that was removed under #3937 is also restored.References: #4035
1.0.18¶
Released: July 24, 2017oracle¶
A fix to cx_Oracle’s WITH_UNICODE mode which was uncovered by the fact that cx_Oracle 5.3 now seems to hardcode this flag on in the build; an internal method that uses this mode wasn’t using the correct signature.
References: #3937
tests¶
Fixed issue in testing fixtures which was incompatible with a change made as of Python 3.6.2 involving context managers.
References: #4034
1.0.17¶
Released: January 17, 2017orm¶
misc¶
Fixed Python 3.6 DeprecationWarnings related to escaped strings without the ‘r’ modifier, and added test coverage for Python 3.6.
References: #3886
1.0.16¶
Released: November 15, 2016orm¶
Fixed bug in
Session.bulk_update_mappings()
where an alternate-named primary key attribute would not track properly into the UPDATE statement.References: #3849
Fixed bug where joined eager loading would fail for a polymorphically- loaded mapper, where the polymorphic_on was set to an un-mapped expression such as a CASE expression.
References: #3800
Fixed bug where the ArgumentError raised for an invalid bind sent to a Session via
Session.bind_mapper()
,Session.bind_table()
, or the constructor would fail to be correctly raised.References: #3798
Fixed bug in
Session.bulk_save()
where an UPDATE would not function correctly in conjunction with a mapping that implements a version id counter.References: #3781
Fixed bug where the
Mapper.attrs
,Mapper.all_orm_descriptors
and other derived attributes would fail to refresh when mapper properties or other ORM constructs were added to the mapper/class after these accessors were first called.References: #3778
mssql¶
Changed the query used to get “default schema name”, from one that queries the database principals table to using the “schema_name()” function, as issues have been reported that the former system was unavailable on the Azure Data Warehouse edition. It is hoped that this will finally work across all SQL Server versions and authentication styles.
References: #3810
Updated the server version info scheme for pyodbc to use SQL Server SERVERPROPERTY(), rather than relying upon pyodbc.SQL_DBMS_VER, which continues to be unreliable particularly with FreeTDS.
References: #3814
Added error code 20017 “unexpected EOF from the server” to the list of disconnect exceptions that result in a connection pool reset. Pull request courtesy Ken Robbins.
References: #3791
Fixed bug in pyodbc dialect (as well as in the mostly non-working adodbapi dialect) whereby a semicolon present in the password or username fields could be interpreted as a separator for another token; the values are now quoted when semicolons are present.
References: #3762
misc¶
Fixed bug where setting up a single-table inh subclass of a joined-table subclass which included an extra column would corrupt the foreign keys collection of the mapped table, thereby interfering with the initialization of relationships.
References: #3797
1.0.15¶
Released: September 1, 2016orm¶
sql¶
Fixed bug in
Table
where the internal method_reset_exported()
would corrupt the state of the object. This method is intended for selectable objects and is called by the ORM in some cases; an erroneous mapper configuration would could lead the ORM to call this on aTable
object.References: #3755
mysql¶
Added support for parsing MySQL/Connector boolean and integer arguments within the URL query string: connection_timeout, connect_timeout, pool_size, get_warnings, raise_on_warnings, raw, consume_results, ssl_verify_cert, force_ipv6, pool_reset_session, compress, allow_local_infile, use_pure.
References: #3787
misc¶
Fixed bug in
sqlalchemy.ext.baked
where the unbaking of a subquery eager loader query would fail due to a variable scoping issue, when multiple subquery loaders were involved. Pull request courtesy Mark Hahnenberg.References: #3743
1.0.14¶
Released: July 6, 2016examples¶
Fixed a regression that occurred in the examples/vertical/dictlike-polymorphic.py example which prevented it from running.
References: #3704
engine¶
Fixed bug in cross-schema foreign key reflection in conjunction with the
MetaData.schema
argument, where a referenced table that is present in the “default” schema would fail since there would be no way to indicate aTable
that has “blank” for a schema. The special symbolsqlalchemy.schema.BLANK_SCHEMA
has been added as an available value forTable.schema
andSequence.schema
, indicating that the schema name should be forced to beNone
even ifMetaData.schema
is specified.References: #3716
sql¶
Fixed issue in SQL math negation operator where the type of the expression would no longer be the numeric type of the original. This would cause issues where the type determined result set behaviors.
References: #3735
Fixed bug whereby the
__getstate__
/__setstate__
methods for sqlalchemy.util.Properties were non-working due to the transition in the 1.0 series to__slots__
. The issue potentially impacted some third-party applications. Pull request courtesy Pieter Mulder.References: #3728
FromClause.count()
is pending deprecation for 1.1. This function makes use of an arbitrary column in the table and is not reliable; for Core use,func.count()
should be preferred.References: #3724
Fixed bug in
CTE
structure which would cause it to not clone properly when a union was used, as is common in a recursive CTE. The improper cloning would cause errors when the CTE is used in various ORM contexts such as that of acolumn_property()
.References: #3722
Fixed bug whereby
Table.tometadata()
would make a duplicateUniqueConstraint
for eachColumn
object that featured theunique=True
parameter.References: #3721
postgresql¶
Fixed bug whereby
TypeDecorator
andVariant
types were not deeply inspected enough by the PostgreSQL dialect to determine if SMALLSERIAL or BIGSERIAL needed to be rendered rather than SERIAL.References: #3739
oracle¶
Fixed bug in
Select.with_for_update.of
, where the Oracle “rownum” approach to LIMIT/OFFSET would fail to accommodate for the expressions inside the “OF” clause, which must be stated at the topmost level referring to expression within the subquery. The expressions are now added to the subquery if needed.References: #3741
1.0.13¶
Released: May 16, 2016orm¶
Fixed bug in “evaluate” strategy of
Query.update()
andQuery.delete()
which would fail to accommodate a bound parameter with a “callable” value, as which occurs when filtering by a many-to-one equality expression along a relationship.References: #3700
Fixed bug whereby the event listeners used for backrefs could be inadvertently applied multiple times, when using a deep class inheritance hierarchy in conjunction with multiple mapper configuration steps.
References: #3710
Fixed bug whereby passing a
text()
construct to theQuery.group_by()
method would raise an error, instead of interpreting the object as a SQL fragment.References: #3706
Anonymous labeling is applied to a
func
construct that is passed tocolumn_property()
, so that if the same attribute is referred to as a column expression twice the names are de-duped, thus avoiding “ambiguous column” errors. Previously, the.label(None)
would need to be applied in order for the name to be de-anonymized.References: #3663
Fixed regression appearing in the 1.0 series in ORM loading where the exception raised for an expected column missing would incorrectly be a
NoneType
error, rather than the expectedNoSuchColumnError
.References: #3658
examples¶
Changed the “directed graph” example to no longer consider integer identifiers of nodes as significant; the “higher” / “lower” references now allow mutual edges in both directions.
References: #3698
sql¶
Fixed bug where when using
case_sensitive=False
with anEngine
, the result set would fail to correctly accommodate for duplicate column names in the result set, causing an error when the statement is executed in 1.0, and preventing the “ambiguous column” exception from functioning in 1.1.References: #3690
Fixed bug where the negation of an EXISTS expression would not be properly typed as boolean in the result, and also would fail to be anonymously aliased in a SELECT list as is the case with a non-negated EXISTS construct.
References: #3682
Fixed bug where “unconsumed column names” exception would fail to be raised in the case where
Insert.values()
were called with a list of parameter mappings, instead of a single mapping of parameters. Pull request courtesy Athena Yao.References: #3666
postgresql¶
Added disconnect detection support for the error string “SSL error: decryption failed or bad record mac”. Pull request courtesy Iuri de Silvio.
References: #3715
mssql¶
Fixed bug where by ROW_NUMBER OVER clause applied for OFFSET selects in SQL Server would inappropriately substitute a plain column from the local statement that overlaps with a label name used by the ORDER BY criteria of the statement.
References: #3711
Fixed regression appearing in the 1.0 series which would cause the Oracle and SQL Server dialects to incorrectly account for result set columns when these dialects would wrap a SELECT in a subquery in order to provide LIMIT/OFFSET behavior, and the original SELECT statement referred to the same column multiple times, such as a column and a label of that same column. This issue is related to #3658 in that when the error occurred, it would also cause a
NoneType
error, rather than reporting that it couldn’t locate a column.References: #3657
oracle¶
Fixed a bug in the cx_Oracle connect process that caused a TypeError when the either the user, password or dsn was empty. This prevented external authentication to Oracle databases, and prevented connecting to the default dsn. The connect string oracle:// now logs into the default dsn using the Operating System username, equivalent to connecting using ‘/’ with sqlplus.
References: #3705
Fixed a bug in the result proxy used mainly by Oracle when binary and other LOB types are in play, such that when query / statement caching were used, the type-level result processors, notably that required by the binary type itself but also any other processor, would become lost after the first run of the statement due to it being removed from the cached result metadata.
References: #3699
misc¶
Fixed bug in “to_list” conversion where a single bytes object would be turned into a list of individual characters. This would impact among other things using the
Query.get()
method on a primary key that’s a bytes object.References: #3660
1.0.12¶
Released: February 15, 2016orm¶
Fixed bug in
Session.merge()
where an object with a composite primary key that has values for some but not all of the PK fields would emit a SELECT statement leaking the internal NEVER_SET symbol into the query, rather than detecting that this object does not have a searchable primary key and no SELECT should be emitted.References: #3647
Fixed regression since 0.9 where the 0.9 style loader options system failed to accommodate for multiple
undefer_group()
loader options in a single query. Multipleundefer_group()
options will now be taken into account even against the same entity.References: #3623
engine¶
Revisiting #2696, first released in 1.0.10, which attempts to work around Python 2’s lack of exception context reporting by emitting a warning for an exception that was interrupted by a second exception when attempting to roll back the already-failed transaction; this issue continues to occur for MySQL backends in conjunction with a savepoint that gets unexpectedly lost, which then causes a “no such savepoint” error when the rollback is attempted, obscuring what the original condition was.
The approach has been generalized to the Core “safe reraise” function which takes place across the ORM and Core in any place that a transaction is being rolled back in response to an error which occurred trying to commit, including the context managers provided by
Session
andConnection
, and taking place for operations such as a failure on “RELEASE SAVEPOINT”. Previously, the fix was only in place for a specific path within the ORM flush/commit process; it now takes place for all transactional context managers as well.References: #2696
sql¶
Fixed issue where the “literal_binds” flag was not propagated for
insert()
,update()
ordelete()
constructs when compiled to string SQL. Pull request courtesy Tim Tate.References: #3643
Fixed issue where inadvertent use of the Python
__contains__
override with a column expression (e.g. by using'x' in col
) would cause an endless loop in the case of an ARRAY type, as Python defers this to__getitem__
access which never raises for this type. Overall, all use of__contains__
now raises NotImplementedError.References: #3642
Fixed bug in
Table
metadata construct which appeared around the 0.9 series where adding columns to aTable
that was unpickled would fail to correctly establish theColumn
within the ‘c’ collection, leading to issues in areas such as ORM configuration. This could impact use cases such asextend_existing
and others.References: #3632
postgresql¶
Fixed bug in
text()
construct where a double-colon expression would not escape properly, e.g.some\:\:expr
, as is most commonly required when rendering PostgreSQL-style CAST expressions.References: #3644
mssql¶
Fixed the syntax of the
extract()
function when used on MSSQL against a datetime value; the quotes around the keyword are removed. Pull request courtesy Guillaume Doumenc.References: #3624
Fixed 1.0 regression where the eager fetch of cursor.rowcount was no longer called for an UPDATE or DELETE statement emitted via plain text or via the
text()
construct, affecting those drivers that erase cursor.rowcount once the cursor is closed such as SQL Server ODBC and Firebird drivers.References: #3622
oracle¶
Fixed a small issue in the Jython Oracle compiler involving the rendering of “RETURNING” which allows this currently unsupported/untested dialect to work rudimentarily with the 1.0 series. Pull request courtesy Carlos Rivas.
References: #3621
misc¶
Fixed bug where some exception re-raise scenarios would attach the exception to itself as the “cause”; while the Python 3 interpreter is OK with this, it could cause endless loops in iPython.
References: #3625
1.0.11¶
Released: December 22, 2015orm¶
Fixed regression caused in 1.0.10 by the fix for #3593 where the check added for a polymorphic joinedload from a poly_subclass->class->poly_baseclass connection would fail for the scenario of class->poly_subclass->class.
References: #3611
Fixed bug where
Session.bulk_update_mappings()
and related would not bump a version id counter when in use. The experience here is still a little rough as the original version id is required in the given dictionaries and there’s not clean error reporting on that yet.References: #3610
Major fixes to the
Mapper.eager_defaults
flag, this flag would not be honored correctly in the case that multiple UPDATE statements were to be emitted, either as part of a flush or a bulk update operation. Additionally, RETURNING would be emitted unnecessarily within update statements.References: #3609
Fixed bug where use of the
Query.select_from()
method would cause a subsequent call to theQuery.with_parent()
method to fail.References: #3606
sql¶
Fixed bug in
Update.return_defaults()
which would cause all insert-default holding columns not otherwise included in the SET clause (such as primary key cols) to get rendered into the RETURNING even though this is an UPDATE.References: #3609
mysql¶
An adjustment to the regular expression used to parse MySQL views, such that we no longer assume the “ALGORITHM” keyword is present in the reflected view source, as some users have reported this not being present in some Amazon RDS environments.
References: #3613
Added new reserved words for MySQL 5.7 to the MySQL dialect, including ‘generated’, ‘optimizer_costs’, ‘stored’, ‘virtual’. Pull request courtesy Hanno Schlichting.
misc¶
Further fixes to #3605, pop method on
MutableDict
, where the “default” argument was not included.References: #3605
Fixed bug in baked loader system where the systemwide monkeypatch for setting up baked lazy loaders would interfere with other loader strategies that rely on lazy loading as a fallback, e.g. joined and subquery eager loaders, leading to
IndexError
exceptions at mapper configuration time.References: #3612
1.0.10¶
Released: December 11, 2015orm¶
Fixed issue where post_update on a many-to-one relationship would fail to emit an UPDATE in the case where the attribute were set to None and not previously loaded.
References: #3599
Fixed bug which is actually a regression that occurred between versions 0.8.0 and 0.8.1, due #2714. The case where joined eager loading needs to join out over a subclass-bound relationship when “with_polymorphic” were also used would fail to join from the correct entity.
References: #3593
Fixed joinedload bug which would occur when a. the query includes limit/offset criteria that forces a subquery b. the relationship uses “secondary” c. the primaryjoin of the relationship refers to a column that is either not part of the primary key, or is a PK col in a joined-inheritance subclass table that is under a different attribute name than the parent table’s primary key column d. the query defers the columns that are present in the primaryjoin, typically via not being included in load_only(); the necessary column(s) would not be present in the subquery and produce invalid SQL.
References: #3592
A rare case which occurs when a
Session.rollback()
fails in the scope of aSession.flush()
operation that’s raising an exception, as has been observed in some MySQL SAVEPOINT cases, prevents the original database exception from being observed when it was emitted during flush, but only on Py2K because Py2K does not support exception chaining; on Py3K the originating exception is chained. As a workaround, a warning is emitted in this specific case showing at least the string message of the original database error before we proceed to raise the rollback-originating exception.References: #2696
orm declarative¶
Fixed bug where in Py2K a unicode literal would not be accepted as the string name of a class or other argument within declarative using
backref()
onrelationship()
. Pull request courtesy Nils Philippsen.
sql¶
Added support for parameter-ordered SET clauses in an UPDATE statement. This feature is available by passing the
update.preserve_parameter_order
flag either to the coreUpdate
construct or alternatively adding it to theQuery.update.update_args
dictionary at the ORM-level, also passing the parameters themselves as a list of 2-tuples. Thanks to Gorka Eguileor for implementation and tests.See also
Fixed issue within the
Insert.from_select()
construct whereby theSelect
construct would have its._raw_columns
collection mutated in-place when compiling theInsert
construct, when the targetTable
has Python-side defaults. TheSelect
construct would compile standalone with the erroneous column present subsequent to compilation of theInsert
, and theInsert
statement itself would fail on a second compile attempt due to duplicate bound parameters.References: #3603
Fixed bug where CREATE TABLE with a no-column table, but a constraint such as a CHECK constraint would render an erroneous comma in the definition; this scenario can occur such as with a PostgreSQL INHERITS table that has no columns of its own.
References: #3598
postgresql¶
Fixed issue where the “FOR UPDATE OF” PostgreSQL-specific SELECT modifier would fail if the referred table had a schema qualifier; PG needs the schema name to be omitted. Pull request courtesy Diana Clarke.
References: #3573
Fixed bug where some varieties of SQL expression passed to the “where” clause of
ExcludeConstraint
would fail to be accepted correctly. Pull request courtesy aisch.Fixed the
.python_type
attribute ofINTERVAL
to returndatetime.timedelta
in the same way as that ofpython_type
, rather than raisingNotImplementedError
.References: #3571
mysql¶
Fixed bug in MySQL reflection where the “fractional sections portion” of the
DATETIME
,TIMESTAMP
andTIME
types would be incorrectly placed into thetimezone
attribute, which is unused by MySQL, instead of thefsp
attribute.References: #3602
mssql¶
Added the error “20006: Write to the server failed” to the list of disconnect errors for the pymssql driver, as this has been observed to render a connection unusable.
References: #3585
A descriptive ValueError is now raised in the event that SQL server returns an invalid date or time format from a DATE or TIME column, rather than failing with a NoneType error. Pull request courtesy Ed Avis.
Fixed issue where DDL generated for the MSSQL types DATETIME2, TIME and DATETIMEOFFSET with a precision of “zero” would not generate the precision field. Pull request courtesy Jacobo de Vera.
tests¶
The ORM and Core tutorials, which have always been in doctest format, are now exercised within the normal unit test suite in both Python 2 and Python 3.
misc¶
Added support for the
dict.pop()
anddict.popitem()
methods to theMutableDict
class.References: #3605
Updates to internal getargspec() calls, some py36-related fixture updates, and alterations to two iterators to “return” instead of raising StopIteration, to allow tests to pass without errors or warnings on Py3.5, Py3.6, pull requests courtesy Jacob MacDonald, Luri de Silvio, and Phil Jones.
Fixed an issue in baked queries where the .get() method, used either directly or within lazy loads, didn’t consider the mapper’s “get clause” as part of the cache key, causing bound parameter mismatches if the clause got re-generated. This clause is cached by mappers on the fly but in highly concurrent scenarios may be generated more than once when first accessed.
References: #3597
1.0.9¶
Released: October 20, 2015orm¶
Added new method
Query.one_or_none()
; same asQuery.one()
but returns None if no row found. Pull request courtesy esiegerman.Fixed regression in 1.0 where new feature of using “executemany” for UPDATE statements in the ORM (e.g. UPDATE statements are now batched with executemany() in a flush) would break on PostgreSQL and other RETURNING backends when using server-side version generation schemes, as the server side value is retrieved via RETURNING which is not supported with executemany.
References: #3556
Fixed rare TypeError which could occur when stringifying certain kinds of internal column loader options within internal logging.
References: #3539
Fixed bug in
Session.bulk_save_objects()
where a mapped column that had some kind of “fetch on update” value and was not locally present in the given object would cause an AttributeError within the operation.References: #3525
Fixed 1.0 regression where the “noload” loader strategy would fail to function for a many-to-one relationship. The loader used an API to place “None” into the dictionary which no longer actually writes a value; this is a side effect of #3061.
References: #3510
examples¶
Fixed two issues in the “history_meta” example where history tracking could encounter empty history, and where a column keyed to an alternate attribute name would fail to track properly. Fixes courtesy Alex Fraser.
sql¶
Fixed regression in 1.0-released default-processor for multi-VALUES insert statement, #3288, where the column type for the default-holding column would not be propagated to the compiled statement in the case where the default was being used, leading to bind-level type handlers not being invoked.
References: #3520
postgresql¶
An adjustment to the new PostgreSQL feature of reflecting storage options and USING of #3455 released in 1.0.6, to disable the feature for PostgreSQL versions < 8.2 where the
reloptions
column is not provided; this allows Amazon Redshift to again work as it is based on an 8.0.x version of PostgreSQL. Fix courtesy Pete Hollobon.
oracle¶
Fixed support for cx_Oracle version 5.2, which was tripping up SQLAlchemy’s version detection under Python 3 and inadvertently not using the correct unicode mode for Python 3. This would cause issues such as bound variables mis-interpreted as NULL and rows silently not being returned.
This change is also backported to: 0.7.0b1
References: #3491
Fixed bug in Oracle dialect where reflection of tables and other symbols with names quoted to force all-lower-case would not be identified properly in reflection queries. The
quoted_name
construct is now applied to incoming symbol names that detect as forced into all-lower-case within the “name normalize” process.References: #3548
misc¶
Added the
AssociationProxy.info
parameter to theAssociationProxy
constructor, to suit theAssociationProxy.info
accessor that was added in #2971. This is possible becauseAssociationProxy
is constructed explicitly, unlike a hybrid which is constructed implicitly via the decorator syntax.References: #3551
Fixed two issues regarding Sybase reflection, allowing tables without primary keys to be reflected as well as ensured that a SQL statement involved in foreign key detection is pre-fetched up front to avoid driver issues upon nested queries. Fixes here courtesy Eugene Zapolsky; note that we cannot currently test Sybase to locally verify these changes.
1.0.8¶
Released: July 22, 2015engine¶
Fixed critical issue whereby the pool “checkout” event handler may be called against a stale connection without the “connect” event handler having been called, in the case where the pool attempted to reconnect after being invalidated and failed; the stale connection would remain present and would be used on a subsequent attempt. This issue has a greater impact in the 1.0 series subsequent to 1.0.2, as it also delivers a blanked-out
.info
dictionary to the event handler; prior to 1.0.2 the.info
dictionary is still the previous one.This change is also backported to: 0.7.0b1
References: #3497
sqlite¶
Fixed bug in SQLite dialect where reflection of UNIQUE constraints that included non-alphabetic characters in the names, like dots or spaces, would not be reflected with their name.
This change is also backported to: 0.9.10
References: #3495
misc¶
Fixed an issue where a particular base class within utils didn’t implement
__slots__
, and therefore meant all subclasses of that class didn’t either, negating the rationale for__slots__
to be in use. Didn’t cause any issue except on IronPython which apparently does not implement__slots__
behavior compatibly with cPython.References: #3494
1.0.7¶
Released: July 20, 2015orm¶
Fixed 1.0 regression where value objects that override
__eq__()
to return a non-boolean-capable object, such as some geoalchemy types as well as numpy types, were being tested forbool()
during a unit of work update operation, where in 0.9 the return value of__eq__()
was tested against “is True” to guard against this.References: #3469
Fixed 1.0 regression where a “deferred” attribute would not populate correctly if it were loaded within the “optimized inheritance load”, which is a special SELECT emitted in the case of joined table inheritance used to populate expired or unloaded attributes against a joined table without loading the base table. This is related to the fact that SQLA 1.0 no longer guesses about loading deferred columns and must be directed explicitly.
References: #3468
Fixed 1.0 regression where the “parent entity” of a synonym- mapped attribute on top of an
aliased()
object would resolve to the original mapper, not thealiased()
version of it, thereby causing problems for aQuery
that relies on this attribute (e.g. it’s the only representative attribute given in the constructor) to figure out the correct FROM clause for the query.References: #3466
orm declarative¶
Fixed bug in
AbstractConcreteBase
extension where a column setup on the ABC base which had a different attribute name vs. column name would not be correctly mapped on the final base class. The failure on 0.9 would be silent whereas on 1.0 it raised an ArgumentError, so may not have been noticed prior to 1.0.References: #3480
engine¶
Fixed regression where new methods on
ResultProxy
used by the ORMQuery
object (part of the performance enhancements of #3175) would not raise the “this result does not return rows” exception in the case where the driver (typically MySQL) fails to generate cursor.description correctly; an AttributeError against NoneType would be raised instead.References: #3481
Fixed regression where
ResultProxy.keys()
would return un-adjusted internal symbol names for “anonymous” labels, which are the “foo_1” types of labels we see generated for SQL functions without labels and similar. This was a side effect of the performance enhancements implemented as part of #918.References: #3483
sql¶
Added a
ColumnElement.cast()
method which performs the same purpose as the standalonecast()
function. Pull request courtesy Sebastian Bank.References: #3459
Fixed bug where coercion of literal
True
orFalse
constant in conjunction withand_()
oror_()
would fail with an AttributeError.References: #3490
Fixed potential issue where a custom subclass of
FunctionElement
or other column element that incorrectly states ‘None’ or any other invalid object as the.type
attribute will report this exception instead of recursion overflow.References: #3485
Fixed bug where the modulus SQL operator wouldn’t work in reverse due to a missing
__rmod__
method. Pull request courtesy dan-gittik.
schema¶
Added support for the MINVALUE, MAXVALUE, NO MINVALUE, NO MAXVALUE, and CYCLE arguments for CREATE SEQUENCE as supported by PostgreSQL and Oracle. Pull request courtesy jakeogh.
1.0.6¶
Released: June 25, 2015orm¶
Fixed a major regression in the 1.0 series where the version_id_counter feature would cause an object’s version counter to be incremented when there was no net change to the object’s row, but instead an object related to it via relationship (e.g. typically many-to-one) were associated or de-associated with it, resulting in an UPDATE statement that updates the object’s version counter and nothing else. In the use case where the relatively recent “server side” and/or “programmatic/conditional” version counter feature were used (e.g. setting version_id_generator to False), the bug could cause an UPDATE without a valid SET clause to be emitted.
References: #3465
Fixed 1.0 regression where the enhanced behavior of single-inheritance joins of #3222 takes place inappropriately for a JOIN along explicit join criteria with a single-inheritance subclass that does not make use of any discriminator, resulting in an additional “AND NULL” clause.
References: #3462
Fixed bug in new
Session.bulk_update_mappings()
feature where the primary key columns used in the WHERE clause to locate the row would also be included in the SET clause, setting their value to themselves unnecessarily. Pull request courtesy Patrick Hayes.References: #3451
Fixed an unexpected-use regression whereby custom
Comparator
objects that made use of the__clause_element__()
method and returned an object that was an ORM-mappedInstrumentedAttribute
and not explicitly aColumnElement
would fail to be correctly handled when passed as an expression toSession.query()
. The logic in 0.9 happened to succeed on this, so this use case is now supported.References: #3448
sql¶
Fixed a bug where clause adaption as applied to a
Label
object would fail to accommodate the labeled SQL expression in all cases, such that any SQL operation that made use ofLabel.self_group()
would use the original unadapted expression. One effect of this would be that an ORMaliased()
construct would not fully accommodate attributes mapped bycolumn_property
, such that the un-aliased table could leak out when the property were used in some kinds of SQL comparisons.References: #3445
postgresql¶
Added support for storage parameters under CREATE INDEX, using a new keyword argument
postgresql_with
. Also added support for reflection to support both thepostgresql_with
flag as well as thepostgresql_using
flag, which will now be set onIndex
objects that are reflected, as well present in a new “dialect_options” dictionary in the result ofInspector.get_indexes()
. Pull request courtesy Pete Hollobon.See also
postgresql_index_storage
References: #3455
Added new execution option
max_row_buffer
which is interpreted by the psycopg2 dialect when thestream_results
option is used, which sets a limit on the size of the row buffer that may be allocated. This value is also provided based on the integer value sent toQuery.yield_per()
. Pull request courtesy mcclurem.Re-fixed this issue first released in 1.0.5 to fix psycopg2cffi JSONB support once again, as they suddenly switched on unconditional decoding of JSONB types in version 2.7.1. Version detection now specifies 2.7.1 as where we should expect the DBAPI to do json encoding for us.
References: #3439
Repaired the
ExcludeConstraint
construct to support common features that other objects likeIndex
now do, that the column expression may be specified as an arbitrary SQL expression such ascast
ortext
.References: #3454
mssql¶
Fixed issue when using
VARBINARY
type in conjunction with an INSERT of NULL + pyodbc; pyodbc requires a special object be passed in order to persist NULL. As theVARBINARY
type is now usually the default forLargeBinary
due to #3039, this issue is partially a regression in 1.0. The pymssql driver appears to be unaffected.References: #3464
misc¶
Fixed an internal “memoization” routine for method types such that a Python descriptor is no longer used; repairs inspectability of these methods including support for Sphinx documentation.
References: #2077
1.0.5¶
Released: June 7, 2015orm¶
Added new event
InstanceEvents.refresh_flush()
, invoked when an INSERT or UPDATE level default value fetched via RETURNING or Python-side default is invoked within the flush process. This is to provide a hook that is no longer present as a result of #3167, where attribute and validation events are no longer called within the flush process.References: #3427
The “lightweight named tuple” used when a
Query
returns rows failed to implement__slots__
correctly such that it still had a__dict__
. This is resolved, but in the extremely unlikely case someone was assigning values to the returned tuples, that will no longer work.References: #3420
engine¶
Added new engine event
ConnectionEvents.engine_disposed()
. Called after theEngine.dispose()
method is called.Adjustments to the engine plugin hook, such that the
URL.get_dialect()
method will continue to return the ultimateDialect
object when a dialect plugin is used, without the need for the caller to be aware of theDialect.get_dialect_cls()
method.References: #3379
Fixed bug where known boolean values used by
engine_from_config()
were not being parsed correctly; these includedpool_threadlocal
and the psycopg2 argumentuse_native_unicode
.References: #3435
Added support for the case of the misbehaving DBAPI that has pep-249 exception names linked to exception classes of an entirely different name, preventing SQLAlchemy’s own exception wrapping from wrapping the error appropriately. The SQLAlchemy dialect in use needs to implement a new accessor
DefaultDialect.dbapi_exception_translation_map
to support this feature; this is implemented now for the py-postgresql dialect.References: #3421
Fixed bug involving the case when pool checkout event handlers are used and connection attempts are made in the handler itself which fail, the owning connection record would not be freed until the stack trace of the connect error itself were freed. For the case where a test pool of only a single connection were used, this means the pool would be fully checked out until that stack trace were freed. This mostly impacts very specific debugging scenarios and is unlikely to have been noticeable in any production application. The fix applies an explicit checkin of the record before re-raising the caught exception.
References: #3419
sql¶
Added official support for a CTE used by the SELECT present inside of
Insert.from_select()
. This behavior worked accidentally up until 0.9.9, when it no longer worked due to unrelated changes as part of #3248. Note that this is the rendering of the WITH clause after the INSERT, before the SELECT; the full functionality of CTEs rendered at the top level of INSERT, UPDATE, DELETE is a new feature targeted for a later release.This change is also backported to: 0.9.10
References: #3418
postgresql¶
Repaired some typing and test issues related to the pypy psycopg2cffi dialect, in particular that the current 2.7.0 version does not have native support for the JSONB type. The version detection for psycopg2 features has been tuned into a specific sub-version for psycopg2cffi. Additionally, test coverage has been enabled for the full series of psycopg2 features under psycopg2cffi.
References: #3439
mssql¶
Added a new dialect flag to the MSSQL dialect
legacy_schema_aliasing
which when set to False will disable a very old and obsolete behavior, that of the compiler’s attempt to turn all schema-qualified table names into alias names, to work around old and no longer locatable issues where SQL server could not parse a multi-part identifier name in all circumstances. The behavior prevented more sophisticated statements from working correctly, including those which use hints, as well as CRUD statements that embed correlated SELECT statements. Rather than continue to repair the feature to work with more complex statements, it’s better to just disable it as it should no longer be needed for any modern SQL server version. The flag defaults to True for the 1.0.x series, leaving current behavior unchanged for this version series. In the 1.1 series, it will default to False. For the 1.0 series, when not set to either value explicitly, a warning is emitted when a schema-qualified table is first used in a statement, which suggests that the flag be set to False for all modern SQL Server versions.See also
legacy_schema_rendering
misc¶
Added support for
*args
to be passed to the baked query initial callable, in the same way that*args
are supported for theBakedQuery.add_criteria()
andBakedQuery.with_criteria()
methods. Initial PR courtesy Naoki INADA.Added a new semi-public method to
MutableBase
MutableBase._get_listen_keys()
. Overriding this method is needed in the case where aMutableBase
subclass needs events to propagate for attribute keys other than the key to which the mutable type is associated with, when intercepting theInstanceEvents.refresh()
orInstanceEvents.refresh_flush()
events. The current example of this is composites usingMutableComposite
.References: #3427
Fixed regression in the
sqlalchemy.ext.mutable
extension as a result of the bugfix for #3167, where attribute and validation events are no longer called within the flush process. The mutable extension was relying upon this behavior in the case where a column level Python-side default were responsible for generating the new value on INSERT or UPDATE, or when a value were fetched from the RETURNING clause for “eager defaults” mode. The new value would not be subject to any event when populated and the mutable extension could not establish proper coercion or history listening. A new eventInstanceEvents.refresh_flush()
is added which the mutable extension now makes use of for this use case.References: #3427
1.0.4¶
Released: May 7, 2015orm¶
Fixed unexpected-use regression where in the odd case that the primaryjoin of a relationship involved comparison to an unhashable type such as an HSTORE, lazy loads would fail due to a hash-oriented check on the statement parameters, modified in 1.0 as a result of #3061 to use hashing and modified in #3368 to occur in cases more common than “load on pending”. The values are now checked for the
__hash__
attribute beforehand.References: #3416
Liberalized an assertion that was added as part of #3347 to protect against unknown conditions when splicing inner joins together within joined eager loads with
innerjoin=True
; if some of the joins use a “secondary” table, the assertion needs to unwrap further joins in order to pass.Repaired / added to tests yet more expressions that were reported as failing with the new ‘entity’ key value added to
Query.column_descriptions
, the logic to discover the “from” clause is again reworked to accommodate columns from aliased classes, as well as to report the correct value for the “aliased” flag in these cases.
schema¶
Fixed bug in enhanced constraint-attachment logic introduced in #3341 where in the unusual case of a constraint that refers to a mixture of
Column
objects and string column names at the same time, the auto-attach-on-column-attach logic will be skipped; for the constraint to be auto-attached in this case, all columns must be assembled on the target table up front. Added a new section to the migration document regarding the original feature as well as this change.References: #3411
tests¶
Fixed an import that prevented “pypy setup.py test” from working correctly.
This change is also backported to: 0.9.10
References: #3406
misc¶
Fixed bug where when using extended attribute instrumentation system, the correct exception would not be raised when
class_mapper()
were called with an invalid input that also happened to not be weak referencable, such as an integer.This change is also backported to: 0.9.10
References: #3408
1.0.3¶
Released: April 30, 2015orm¶
Fixed regression from 0.9.10 prior to release due to #3349 where the check for query state on
Query.update()
orQuery.delete()
compared the empty tuple to itself usingis
, which fails on PyPy to produceTrue
in this case; this would erroneously emit a warning in 0.9 and raise an exception in 1.0.References: #3405
Fixed regression from 0.9.10 prior to release where the new addition of
entity
to theQuery.column_descriptions
accessor would fail if the target entity was produced from a core selectable such as aTable
orCTE
object.Fixed regression within the flush process when an attribute were set to a SQL expression for an UPDATE, and the SQL expression when compared to the previous value of the attribute would produce a SQL comparison other than
==
or!=
, the exception “Boolean value of this clause is not defined” would raise. The fix ensures that the unit of work will not interpret the SQL expression in this way.References: #3402
Fixed unexpected use regression due to #2992 where textual elements placed into the
Query.order_by()
clause in conjunction with joined eager loading would be added to the columns clause of the inner query in such a way that they were assumed to be table-bound column names, in the case where the joined eager load needs to wrap the query in a subquery to accommodate for a limit/offset.Originally, the behavior here was intentional, in that a query such as
query(User).order_by('name').limit(1)
would order byuser.name
even if the query was modified by joined eager loading to be within a subquery, as'name'
would be interpreted as a symbol to be located within the FROM clauses, in this caseUser.name
, which would then be copied into the columns clause to ensure it were present for ORDER BY. However, the feature fails to anticipate the case whereorder_by("name")
refers to a specific label name present in the local columns clause already and not a name bound to a selectable in the FROM clause.Beyond that, the feature also fails for deprecated cases such as
order_by("name desc")
, which, while it emits a warning thattext()
should be used here (note that the issue does not impact cases wheretext()
is used explicitly), still produces a different query than previously where the “name desc” expression is copied into the columns clause inappropriately. The resolution is such that the “joined eager loading” aspect of the feature will skip over these so-called “label reference” expressions when augmenting the inner columns clause, as though they weretext()
constructs already.References: #3392
Fixed a regression regarding the
MapperEvents.instrument_class()
event where its invocation was moved to be after the class manager’s instrumentation of the class, which is the opposite of what the documentation for the event explicitly states. The rationale for the switch was due to Declarative taking the step of setting up the full “instrumentation manager” for a class before it was mapped for the purpose of the new@declared_attr
features described in Improvements to declarative mixins, @declared_attr and related features, but the change was also made against the classical use ofmapper()
for consistency. However, SQLSoup relies upon the instrumentation event happening before any instrumentation under classical mapping. The behavior is reverted in the case of classical and declarative mapping, the latter implemented by using a simple memoization without using class manager.References: #3388
Fixed issue in new
QueryEvents.before_compile()
event where changes made to theQuery
object’s collection of entities to load within the event would render in the SQL, but would not be reflected during the loading process.References: #3387
engine¶
New features added to support engine/pool plugins with advanced functionality. Added a new “soft invalidate” feature to the connection pool at the level of the checked out connection wrapper as well as the
_ConnectionRecord
. This works similarly to a modern pool invalidation in that connections aren’t actively closed, but are recycled only on next checkout; this is essentially a per-connection version of that feature. A new eventPoolEvents.soft_invalidate()
is added to complement it.Also added new flag
ExceptionContext.invalidate_pool_on_disconnect
. Allows an error handler withinConnectionEvents.handle_error()
to maintain a “disconnect” condition, but to handle calling invalidate on individual connections in a specific manner within the event.References: #3379
Added new event
do_connect
, which allows interception / replacement of when theDialect.connect()
hook is called to create a DBAPI connection. Also added dialect plugin hooksDialect.get_dialect_cls()
andDialect.engine_created()
which allow external plugins to add events to existing dialects using entry points.References: #3355
sql¶
Added a placeholder method
TypeEngine.compare_against_backend()
which is now consumed by Alembic migrations as of 0.7.6. User-defined types can implement this method to assist in the comparison of a type against one reflected from the database.Fixed bug where the truncation of long labels in SQL could produce a label that overlapped another label that is not truncated; this because the length threshold for truncation was greater than the portion of the label that remains after truncation. These two values have now been made the same; label_length - 6. The effect here is that shorter column labels will be “truncated” where they would not have been truncated before.
References: #3396
Fixed regression due to #3282 where the
tables
collection passed as a keyword argument to theDDLEvents.before_create()
,DDLEvents.after_create()
,DDLEvents.before_drop()
, andDDLEvents.after_drop()
events would no longer be a list of tables, but instead a list of tuples which contained a second entry with foreign keys to be added or dropped. As thetables
collection, while documented as not necessarily stable, has come to be relied upon, this change is considered a regression. Additionally, in some cases for “drop”, this collection would be an iterator that would cause the operation to fail if prematurely iterated. The collection is now a list of table objects in all cases and test coverage for the format of this collection is now added.References: #3391
misc¶
Fixed bug in association proxy where an any()/has() on an relationship->scalar non-object attribute comparison would fail, e.g.
filter(Parent.some_collection_to_attribute.any(Child.attr == 'foo'))
References: #3397
1.0.2¶
Released: April 24, 2015orm declarative¶
Fixed unexpected use regression regarding the declarative
__declare_first__
and__declare_last__
accessors where these would no longer be called on the superclass of the declarative base.References: #3383
sql¶
Fixed a regression that was incorrectly fixed in 1.0.0b4 (hence becoming two regressions); reports that SELECT statements would GROUP BY a label name and fail was misconstrued that certain backends such as SQL Server should not be emitting ORDER BY or GROUP BY on a simple label name at all; when in fact, we had forgotten that 0.9 was already emitting ORDER BY on a simple label name for all backends, as described in Label constructs can now render as their name alone in an ORDER BY, even though 1.0 includes a rewrite of this logic as part of #2992. As far as emitting GROUP BY against a simple label, even PostgreSQL has cases where it will raise an error even though the label to group on should be apparent, so it is clear that GROUP BY should never be rendered in this way automatically.
In 1.0.2, SQL Server, Firebird and others will again emit ORDER BY on a simple label name when passed a
Label
construct that is also present in the columns clause. Additionally, no backend will emit GROUP BY against the simple label name only when passed aLabel
construct.
1.0.1¶
Released: April 23, 2015orm¶
Fixed issue where a query of the form
query(B).filter(B.a != A(id=7))
would render theNEVER_SET
symbol, when given a transient object. For a persistent object, it would always use the persisted database value and not the currently set value. Assuming autoflush is turned on, this usually would not be apparent for persistent values, as any pending changes would be flushed first in any case. However, this is inconsistent vs. the logic used for the non-negated comparison,query(B).filter(B.a == A(id=7))
, which does use the current value and additionally allows comparisons to transient objects. The comparison now uses the current value and not the database-persisted value.Unlike the other
NEVER_SET
issues that are repaired as regressions caused by #3061 in this release, this particular issue is present at least as far back as 0.8 and possibly earlier, however it was discovered as a result of repairing the relatedNEVER_SET
issues.References: #3374
Fixed unexpected use regression cause by #3061 where the NEVER_SET symbol could leak into relationship-oriented queries, including
filter()
andwith_parent()
queries. TheNone
symbol is returned in all cases, however many of these queries have never been correctly supported in any case, and produce comparisons to NULL without using the IS operator. For this reason, a warning is also added to that subset of relationship queries that don’t currently provide forIS NULL
.References: #3371
Fixed a regression caused by #3061 where the NEVER_SET symbol could leak into a lazyload query, subsequent to the flush of a pending object. This would occur typically for a many-to-one relationship that does not use a simple “get” strategy. The good news is that the fix improves efficiency vs. 0.9, because we can now skip the SELECT statement entirely when we detect NEVER_SET symbols present in the parameters; prior to #3061, we couldn’t discern if the None here were set or not.
References: #3368
engine¶
Added the string value
"none"
to those accepted by thePool.reset_on_return
parameter as a synonym forNone
, so that string values can be used for all settings, allowing utilities likeengine_from_config()
to be usable without issue.This change is also backported to: 0.9.10
References: #3375
sql¶
Fixed issue where a straight SELECT EXISTS query would fail to assign the proper result type of Boolean to the result mapping, and instead would leak column types from within the query into the result map. This issue exists in 0.9 and earlier as well, however has less of an impact in those versions. In 1.0, due to #918 this becomes a regression in that we now rely upon the result mapping to be very accurate, else we can assign result-type processors to the wrong column. In all versions, this issue also has the effect that a simple EXISTS will not apply the Boolean type handler, leading to simple 1/0 values for backends without native boolean instead of True/False. The fix includes that an EXISTS columns argument will be anon-labeled like other column expressions; a similar fix is implemented for pure-boolean expressions like
not_(True())
.References: #3372
sqlite¶
Fixed a regression due to #3282, where due to the fact that we attempt to assume the availability of ALTER when creating/dropping schemas, in the case of SQLite we simply said to not worry about foreign keys at all, since ALTER is not available, when creating and dropping tables. This meant that the sorting of tables was basically skipped in the case of SQLite, and for the vast majority of SQLite use cases, this is not an issue.
However, users who were doing DROPs on SQLite with tables that contained data and with referential integrity turned on would then experience errors, as the dependency sorting does matter in the case of DROP with enforced constraints, when those tables have data (SQLite will still happily let you create foreign keys to nonexistent tables and drop tables referring to existing ones with constraints enabled, as long as there’s no data being referenced).
In order to maintain the new feature of #3282 while still allowing a SQLite DROP operation to maintain ordering, we now do the sort with full FKs taken under consideration, and if we encounter an unresolvable cycle, only then do we forego attempting to sort the tables; we instead emit a warning and go with the unsorted list. If an environment needs both ordered DROPs and has foreign key cycles, then the warning notes they will need to restore the
use_alter
flag to theirForeignKey
andForeignKeyConstraint
objects so that just those objects will be omitted from the dependency sort.See also
The use_alter flag on ForeignKeyConstraint is (usually) no longer needed - contains an updated note about SQLite.
References: #3378
firebird¶
Fixed a regression due to #3034 where limit/offset clauses were not properly interpreted by the Firebird dialect. Pull request courtesy effem-git.
References: #3380
Fixed support for “literal_binds” mode when using limit/offset with Firebird, so that the values are again rendered inline when this is selected. Related to #3034.
References: #3381
1.0.0¶
Released: April 16, 2015orm¶
Added new argument
Query.update.update_args
which allows kw arguments such asmysql_limit
to be passed to the underlyingUpdate
construct. Pull request courtesy Amir Sadoughi.Identified an inconsistency when handling
Query.join()
to the same target more than once; it implicitly dedupes only in the case of a relationship join, and due to #3233, in 1.0 a join to the same table twice behaves differently than 0.9 in that it no longer erroneously aliases. To help document this change, the verbiage regarding #3233 in the migration notes has been generalized, and a warning has been added whenQuery.join()
is called against the same target relationship more than once.References: #3367
Made a small improvement to the heuristics of relationship when determining remote side with semi-self-referential (e.g. two joined inh subclasses referring to each other), non-simple join conditions such that the parententity is taken into account and can reduce the need for using the
remote()
annotation; this can restore some cases that might have worked without the annotation prior to 0.9.4 via #2948.References: #3364
sql¶
The topological sorting used to sort
Table
objects and available via theMetaData.sorted_tables
collection will now produce a deterministic ordering; that is, the same ordering each time given a set of tables with particular names and dependencies. This is to help with comparison of DDL scripts and other use cases. The tables are sent to the topological sort sorted by name, and the topological sort itself will process the incoming data in an ordered fashion. Pull request courtesy Sebastian Bank.References: #3084
Fixed issue where a
MetaData
object that used a naming convention would not properly work with pickle. The attribute was skipped leading to inconsistencies and failures if the unpickledMetaData
object were used to base additional tables from.This change is also backported to: 0.9.10
References: #3362
postgresql¶
Fixed a long-standing bug where the
Enum
type as used with the psycopg2 dialect in conjunction with non-ascii values andnative_enum=False
would fail to decode return results properly. This stemmed from when the PGENUM
type used to be a standalone type without a “non native” option.This change is also backported to: 0.9.10
References: #3354
mssql¶
Fixed a regression where the “last inserted id” mechanics would fail to store the correct value for MSSQL on an INSERT where the primary key value was present in the insert params before execution, as well as in the case where an INSERT from SELECT would state the target columns as column objects, instead of string keys.
References: #3360
Using the
Binary
constructor now present in pymssql rather than patching one in. Pull request courtesy Ramiro Morales.
tests¶
Fixed the pathing used when tests run; for sqla_nose.py and py.test, the “./lib” prefix is again inserted at the head of sys.path but only if sys.flags.no_user_site isn’t set; this makes it act just like the way Python puts “.” in the current path by default. For tox, we are setting the PYTHONNOUSERSITE flag now.
References: #3356
1.0.0b5¶
Released: April 3, 2015orm¶
Fixed bug where the state tracking within multiple, nested
Session.begin_nested()
operations would fail to propagate the “dirty” flag for an object that had been updated within the inner savepoint, such that if the enclosing savepoint were rolled back, the object would not be part of the state that was expired and therefore reverted to its database state.This change is also backported to: 0.9.10
References: #3352
Query
doesn’t support joins, subselects, or special FROM clauses when using theQuery.update()
orQuery.delete()
methods; instead of silently ignoring these fields if methods likeQuery.join()
orQuery.select_from()
has been called, an error is raised. In 0.9.10 this only emits a warning.References: #3349
Added a list() call around a weak dictionary used within the commit phase of the session, which without it could cause a “dictionary changed size during iter” error if garbage collection interacted within the process. Change was introduced by #3139.
Fixed a bug related to “nested” inner join eager loading, which exists in 0.9 as well but is more of a regression in 1.0 due to #3008 which turns on “nested” by default, such that a joined eager load that travels across sibling paths from a common ancestor using innerjoin=True will correctly splice each “innerjoin” sibling into the appropriate part of the join, when a series of inner/outer joins are mixed together.
References: #3347
sql¶
The warning emitted by the unicode type for a non-unicode type has been liberalized to warn for values that aren’t even string values, such as integers; previously, the updated warning system of 1.0 made use of string formatting operations which would raise an internal TypeError. While these cases should ideally raise totally, some backends like SQLite and MySQL do accept them and are potentially in use by legacy code, not to mention that they will always pass through if unicode conversion is turned off for the target backend.
References: #3346
postgresql¶
1.0.0b4¶
Released: March 29, 2015sql¶
Fixed bug in new “label resolution” feature of #2992 where a label that was anonymous, then labeled again with a name, would fail to be locatable via a textual label. This situation occurs naturally when a mapped
column_property()
is given an explicit label in a query.References: #3340
Fixed bug in new “label resolution” feature of #2992 where the string label placed in the order_by() or group_by() of a statement would place higher priority on the name as found inside the FROM clause instead of a more locally available name inside the columns clause.
References: #3335
schema¶
The “auto-attach” feature of constraints such as
UniqueConstraint
andCheckConstraint
has been further enhanced such that when the constraint is associated with non-table-boundColumn
objects, the constraint will set up event listeners with the columns themselves such that the constraint auto attaches at the same time the columns are associated with the table. This in particular helps in some edge cases in declarative but is also of general use.References: #3341
mysql¶
Fixed unicode support for PyMySQL when using an “executemany” operation with unicode parameters. SQLAlchemy now passes both the statement as well as the bound parameters as unicode objects, as PyMySQL generally uses string interpolation internally to produce the final statement, and in the case of executemany does the “encode” step only on the final statement.
This change is also backported to: 0.9.10
References: #3337
mssql¶
Turned off the “simple order by” flag on the MSSQL, Oracle dialects; this is the flag that per #2992 causes an order by or group by an expression that’s also in the columns clause to be copied by label, even if referenced as the expression object. The behavior for MSSQL is now the old behavior that copies the whole expression in by default, as MSSQL can be picky on these particularly in GROUP BY expressions. The flag is also turned off defensively for the Firebird and Sybase dialects.
Note
this resolution was incorrect, please see version 1.0.2 for a rework of this resolution.
References: #3338
1.0.0b3¶
Released: March 20, 2015mysql¶
Repaired the commit for issue #2771 which was inadvertently commented out.
References: #2771
1.0.0b2¶
Released: March 20, 2015orm¶
Fixed unexpected use regression from pullreq github:137 where Py2K unicode literals (e.g.
u""
) would not be accepted by therelationship.cascade
option. Pull request courtesy Julien Castets.References: #3327
orm declarative¶
Loosened some restrictions that were added to
@declared_attr
objects, such that they were prevented from being called outside of the declarative process; this is related to the enhancements of #3150 which allow@declared_attr
to return a value that is cached based on the current class as it’s being configured. The exception raise has been removed, and the behavior changed so that outside of the declarative process, the function decorated by@declared_attr
is called every time just like a regular@property
, without using any caching, as none is available at this stage.References: #3331
engine¶
The “auto close” for
ResultProxy
is now a “soft” close. That is, after exhausting all rows using the fetch methods, the DBAPI cursor is released as before and the object may be safely discarded, but the fetch methods may continue to be called for which they will return an end-of-result object (None for fetchone, empty list for fetchmany and fetchall). Only ifResultProxy.close()
is called explicitly will these methods raise the “result is closed” error.
mysql¶
Fixed the
BIT
type on Py3K which was not using theord()
function correctly. Pull request courtesy David Marin.This change is also backported to: 0.9.10
References: #3333
Fixes to fully support using the
'utf8mb4'
MySQL-specific charset with MySQL dialects, in particular MySQL-Python and PyMySQL. In addition, MySQL databases that report more unusual charsets such as ‘koi8u’ or ‘eucjpms’ will also work correctly. Pull request courtesy Thomas Grainger.References: #2771
1.0.0b1¶
Released: March 13, 2015Version 1.0.0b1 is the first release of the 1.0 series. Many changes described here are also present in the 0.9 and sometimes the 0.8 series as well. For changes that are specific to 1.0 with an emphasis on compatibility concerns, see What’s New in SQLAlchemy 1.0?.
general¶
Structural memory use has been improved via much more significant use of
__slots__
for many internal objects. This optimization is particularly geared towards the base memory size of large applications that have lots of tables and columns, and greatly reduces memory size for a variety of high-volume objects including event listening internals, comparator objects and parts of the ORM attribute and loader strategy system.The
__module__
attribute is now set for all those SQL and ORM functions that are derived as “public factory” symbols, which should assist with documentation tools being able to report on the target module.References: #3218
orm¶
Added a new entry
"entity"
to the dictionaries returned byQuery.column_descriptions
. This refers to the primary ORM mapped class or aliased class that is referred to by the expression. Compared to the existing entry for"type"
, it will always be a mapped entity, even if extracted from a column expression, or None if the given expression is a pure core expression. See also #3403 which repaired a regression in this feature which was unreleased in 0.9.10 but was released in the 1.0 version.This change is also backported to: 0.9.10
References: #3320
Added new parameter
Session.connection.execution_options
which may be used to set up execution options on aConnection
when it is first checked out, before the transaction has begun. This is used to set up options such as isolation level on the connection before the transaction starts.See also
Setting Transaction Isolation Levels / DBAPI AUTOCOMMIT - new documentation section detailing best practices for setting transaction isolation with sessions.
This change is also backported to: 0.9.9
References: #3296
Added new method
Session.invalidate()
, functions similarly toSession.close()
, except also callsConnection.invalidate()
on all connections, guaranteeing that they will not be returned to the connection pool. This is useful in situations e.g. dealing with gevent timeouts when it is not safe to use the connection further, even for rollbacks.This change is also backported to: 0.9.9
The “primaryjoin” model has been stretched a bit further to allow a join condition that is strictly from a single column to itself, translated through some kind of SQL function or expression. This is kind of experimental, but the first proof of concept is a “materialized path” join condition where a path string is compared to itself using “like”. The
ColumnOperators.like()
operator has also been added to the list of valid operators to use in a primaryjoin condition.This change is also backported to: 0.9.5
References: #3029
Added new utility function
make_transient_to_detached()
which can be used to manufacture objects that behave as though they were loaded from a session, then detached. Attributes that aren’t present are marked as expired, and the object can be added to a Session where it will act like a persistent one.This change is also backported to: 0.9.5
References: #3017
Added a new event suite
QueryEvents
. TheQueryEvents.before_compile()
event allows the creation of functions which may place additional modifications toQuery
objects before the construction of the SELECT statement. It is hoped that this event be made much more useful via the advent of a new inspection system that will allow for detailed modifications to be made againstQuery
objects in an automated fashion.See also
QueryEvents
References: #3317
The subquery wrapping which occurs when joined eager loading is used with a one-to-many query that also features LIMIT, OFFSET, or DISTINCT has been disabled in the case of a one-to-one relationship, that is a one-to-many with
relationship.uselist
set to False. This will produce more efficient queries in these cases.References: #3249
Mapped state internals have been reworked to allow for a 50% reduction in callcounts specific to the “expiration” of objects, as in the “auto expire” feature of
Session.commit()
and forSession.expire_all()
, as well as in the “cleanup” step which occurs when object states are garbage collected.References: #3307
A warning is emitted when the same polymorphic identity is assigned to two different mappers in the same hierarchy. This is typically a user error and means that the two different mapping types cannot be correctly distinguished at load time. Pull request courtesy Sebastian Bank.
References: #3262
A new series of
Session
methods which provide hooks directly into the unit of work’s facility for emitting INSERT and UPDATE statements has been created. When used correctly, this expert-oriented system can allow ORM-mappings to be used to generate bulk insert and update statements batched into executemany groups, allowing the statements to proceed at speeds that rival direct use of the Core.See also
References: #3100
Added a parameter
Query.join.isouter
which is synonymous with callingQuery.outerjoin()
; this flag is to provide a more consistent interface compared to CoreFromClause.join()
. Pull request courtesy Jonathan Vanasco.References: #3217
Added new event handlers
AttributeEvents.init_collection()
andAttributeEvents.dispose_collection()
, which track when a collection is first associated with an instance and when it is replaced. These handlers supersede thecollection.linker()
annotation. The old hook remains supported through an event adapter.The
Query
will raise an exception whenQuery.yield_per()
is used with mappings or options where either subquery eager loading, or joined eager loading with collections, would take place. These loading strategies are not currently compatible with yield_per, so by raising this error, the method is safer to use. Eager loads can be disabled with thelazyload('*')
option orQuery.enable_eagerloads()
.A new implementation for
KeyedTuple
used by theQuery
object offers dramatic speed improvements when fetching large numbers of column-oriented rows.References: #3176
The behavior of
joinedload.innerjoin
as well asrelationship.innerjoin
is now to use “nested” inner joins, that is, right-nested, as the default behavior when an inner join joined eager load is chained to an outer join eager load.References: #3008
UPDATE statements can now be batched within an ORM flush into more performant executemany() call, similarly to how INSERT statements can be batched; this will be invoked within flush to the degree that subsequent UPDATE statements for the same mapping and table involve the identical columns within the VALUES clause, that no SET-level SQL expressions are embedded, and that the versioning requirements for the mapping are compatible with the backend dialect’s ability to return a correct rowcount for an executemany operation.
The
info
parameter has been added to the constructor forSynonymProperty
andComparableProperty
.References: #2963
The
InspectionAttr.info
collection is now moved down toInspectionAttr
, where in addition to being available on allMapperProperty
objects, it is also now available on hybrid properties, association proxies, when accessed viaMapper.all_orm_descriptors
.References: #2971
Mapped attributes marked as deferred without explicit undeferral will now remain “deferred” even if their column is otherwise present in the result set in some way. This is a performance enhancement in that an ORM load no longer spends time searching for each deferred column when the result set is obtained. However, for an application that has been relying upon this, an explicit
undefer()
or similar option should now be used.The
proc()
callable passed to thecreate_row_processor()
method of customBundle
classes now accepts only a single “row” argument.Deprecated event hooks removed:
populate_instance
,create_instance
,translate_row
,append_result
See also
Fixed bug in subquery eager loading where a long chain of eager loads across a polymorphic-subclass boundary in conjunction with polymorphic loading would fail to locate the subclass-link in the chain, erroring out with a missing property name on an
AliasedClass
.This change is also backported to: 0.9.5, 0.8.7
References: #3055
Fixed ORM bug where the
class_mapper()
function would mask AttributeErrors or KeyErrors that should raise during mapper configuration due to user errors. The catch for attribute/keyerror has been made more specific to not include the configuration step.This change is also backported to: 0.9.5, 0.8.7
References: #3047
Fixed bugs in ORM object comparisons where comparison of many-to-one
!= None
would fail if the source were an aliased class, or if the query needed to apply special aliasing to the expression due to aliased joins or polymorphic querying; also fixed bug in the case where comparing a many-to-one to an object state would fail if the query needed to apply special aliasing due to aliased joins or polymorphic querying.This change is also backported to: 0.9.9
References: #3310
Fixed bug where internal assertion would fail in the case where an
after_rollback()
handler for aSession
incorrectly adds state to thatSession
within the handler, and the task to warn and remove this state (established by #2389) attempts to proceed.This change is also backported to: 0.9.9
References: #3309
Fixed bug where TypeError raised when
Query.join()
called with unknown kw arguments would raise its own TypeError due to broken formatting. Pull request courtesy Malthe Borch.This change is also backported to: 0.9.9
Fixed bug in lazy loading SQL construction whereby a complex primaryjoin that referred to the same “local” column multiple times in the “column that points to itself” style of self-referential join would not be substituted in all cases. The logic to determine substitutions here has been reworked to be more open-ended.
This change is also backported to: 0.9.9
References: #3300
The “wildcard” loader options, in particular the one set up by the
load_only()
option to cover all attributes not explicitly mentioned, now takes into account the superclasses of a given entity, if that entity is mapped with inheritance mapping, so that attribute names within the superclasses are also omitted from the load. Additionally, the polymorphic discriminator column is unconditionally included in the list, just in the same way that primary key columns are, so that even with load_only() set up, polymorphic loading of subtypes continues to function correctly.This change is also backported to: 0.9.9
References: #3287
Fixed bug where if an exception were thrown at the start of a
Query
before it fetched results, particularly when row processors can’t be formed, the cursor would stay open with results pending and not actually be closed. This is typically only an issue on an interpreter like PyPy where the cursor isn’t immediately GC’ed, and can in some circumstances lead to transactions/ locks being open longer than is desirable.This change is also backported to: 0.9.9
References: #3285
Fixed a leak which would occur in the unsupported and highly non-recommended use case of replacing a relationship on a fixed mapped class many times, referring to an arbitrarily growing number of target mappers. A warning is emitted when the old relationship is replaced, however if the mapping were already used for querying, the old relationship would still be referenced within some registries.
This change is also backported to: 0.9.9
References: #3251
Fixed bug regarding expression mutations which could express itself as a “Could not locate column” error when using
Query
to select from multiple, anonymous column entities when querying against SQLite, as a side effect of the “join rewriting” feature used by the SQLite dialect.This change is also backported to: 0.9.9
References: #3241
Fixed bug where the ON clause for
Query.join()
, andQuery.outerjoin()
to a single-inheritance subclass usingof_type()
would not render the “single table criteria” in the ON clause if thefrom_joinpoint=True
flag were set.This change is also backported to: 0.9.9
References: #3232
Fixed bug that affected generally the same classes of event as that of #3199, when the
named=True
parameter would be used. Some events would fail to register, and others would not invoke the event arguments correctly, generally in the case of when an event was “wrapped” for adaption in some other way. The “named” mechanics have been rearranged to not interfere with the argument signature expected by internal wrapper functions.This change is also backported to: 0.9.8
References: #3197
Fixed bug that affected many classes of event, particularly ORM events but also engine events, where the usual logic of “de duplicating” a redundant call to
listen()
with the same arguments would fail, for those events where the listener function is wrapped. An assertion would be hit within registry.py. This assertion has now been integrated into the deduplication check, with the added bonus of a simpler means of checking deduplication across the board.This change is also backported to: 0.9.8
References: #3199
Fixed warning that would emit when a complex self-referential primaryjoin contained functions, while at the same time remote_side was specified; the warning would suggest setting “remote side”. It now only emits if remote_side isn’t present.
This change is also backported to: 0.9.8
References: #3194
Fixed a regression caused by #2976 released in 0.9.4 where the “outer join” propagation along a chain of joined eager loads would incorrectly convert an “inner join” along a sibling join path into an outer join as well, when only descendant paths should be receiving the “outer join” propagation; additionally, fixed related issue where “nested” join propagation would take place inappropriately between two sibling join paths.
This change is also backported to: 0.9.7
References: #3131
Fixed a regression from 0.9.0 due to #2736 where the
Query.select_from()
method no longer set up the “from entity” of theQuery
object correctly, so that subsequentQuery.filter_by()
orQuery.join()
calls would fail to check the appropriate “from” entity when searching for attributes by string name.This change is also backported to: 0.9.7
Fixed bug where items that were persisted, deleted, or had a primary key change within a savepoint block would not participate in being restored to their former state (not in session, in session, previous PK) after the outer transaction were rolled back.
This change is also backported to: 0.9.7
References: #3108
Fixed bug in subquery eager loading in conjunction with
with_polymorphic()
, the targeting of entities and columns in the subquery load has been made more accurate with respect to this type of entity and others.This change is also backported to: 0.9.7
References: #3106
Additional checks have been added for the case where an inheriting mapper is implicitly combining one of its column-based attributes with that of the parent, where those columns normally don’t necessarily share the same value. This is an extension of an existing check that was added via #1892; however this new check emits only a warning, instead of an exception, to allow for applications that may be relying upon the existing behavior.
This change is also backported to: 0.9.5
References: #3042
Modified the behavior of
load_only()
such that primary key columns are always added to the list of columns to be “undeferred”; otherwise, the ORM can’t load the row’s identity. Apparently, one can defer the mapped primary keys and the ORM will fail, that hasn’t been changed. But as load_only is essentially saying “defer all but X”, it’s more critical that PK cols not be part of this deferral.This change is also backported to: 0.9.5
References: #3080
Fixed a few edge cases which arise in the so-called “row switch” scenario, where an INSERT/DELETE can be turned into an UPDATE. In this situation, a many-to-one relationship set to None, or in some cases a scalar attribute set to None, may not be detected as a net change in value, and therefore the UPDATE would not reset what was on the previous row. This is due to some as-yet unresolved side effects of the way attribute history works in terms of implicitly assuming None isn’t really a “change” for a previously un-set attribute. See also #3061.
Note
This change has been REVERTED in 0.9.6. The full fix will be in version 1.0 of SQLAlchemy.
This change is also backported to: 0.9.5
References: #3060
Related to #3060, an adjustment has been made to the unit of work such that loading for related many-to-one objects is slightly more aggressive, in the case of a graph of self-referential objects that are to be deleted; the load of related objects is to help determine the correct order for deletion if passive_deletes is not set.
This change is also backported to: 0.9.5
Fixed bug in SQLite join rewriting where anonymized column names due to repeats would not correctly be rewritten in subqueries. This would affect SELECT queries with any kind of subquery + join.
This change is also backported to: 0.9.5
References: #3057
Fixes to the newly enhanced boolean coercion in #2804 where the new rules for “where” and “having” wouldn’t take effect for the “whereclause” and “having” kw arguments of the
select()
construct, which is also whatQuery
uses so wasn’t working in the ORM either.This change is also backported to: 0.9.5
References: #3013
Fixed bug where the session attachment error “object is already attached to session X” would fail to prevent the object from also being attached to the new session, in the case that execution continued after the error raise occurred.
References: #3301
The primary
Mapper
of aQuery
is now passed to theSession.get_bind()
method when calling uponQuery.count()
,Query.update()
,Query.delete()
, as well as queries against mapped columns,column_property
objects, and SQL functions and expressions derived from mapped columns. This allows sessions that rely upon either customizedSession.get_bind()
schemes or “bound” metadata to work in all relevant cases.The
PropComparator.of_type()
modifier has been improved in conjunction with loader directives such asjoinedload()
andcontains_eager()
such that if twoPropComparator.of_type()
modifiers of the same base type/path are encountered, they will be joined together into a single “polymorphic” entity, rather than replacing the entity of type A with the one of type B. E.g. a joinedload ofA.b.of_type(BSub1)->BSub1.c
combined with joinedload ofA.b.of_type(BSub2)->BSub2.c
will create a single joinedload ofA.b.of_type((BSub1, BSub2)) -> BSub1.c, BSub2.c
, without the need for thewith_polymorphic
to be explicit in the query.See also
Eager Loading of Specific or Polymorphic Subtypes - contains an updated example illustrating the new format.
References: #3256
Repaired support of the
copy.deepcopy()
call when used by theCascadeOptions
argument, which occurs ifcopy.deepcopy()
is being used withrelationship()
(not an officially supported use case). Pull request courtesy duesenfranz.Fixed bug where
Session.expunge()
would not fully detach the given object if the object had been subject to a delete operation that was flushed, but not committed. This would also affect related operations likemake_transient()
.References: #3139
A warning is emitted in the case of multiple relationships that ultimately will populate a foreign key column in conflict with another, where the relationships are attempting to copy values from different source columns. This occurs in the case where composite foreign keys with overlapping columns are mapped to relationships that each refer to a different referenced column. A new documentation section illustrates the example as well as how to overcome the issue by specifying “foreign” columns specifically on a per-relationship basis.
See also
References: #3230
The
Query.update()
method will now convert string key names in the given dictionary of values into mapped attribute names against the mapped class being updated. Previously, string names were taken in directly and passed to the core update statement without any means to resolve against the mapped entity. Support for synonyms and hybrid attributes as the subject attributes ofQuery.update()
are also supported.References: #3228
Improvements to the mechanism used by
Session
to locate “binds” (e.g. engines to use), such engines can be associated with mixin classes, concrete subclasses, as well as a wider variety of table metadata such as joined inheritance tables.References: #3035
Fixed bug in single table inheritance where a chain of joins that included the same single inh entity more than once (normally this should raise an error) could, in some cases depending on what was being joined “from”, implicitly alias the second case of the single inh entity, producing a query that “worked”. But as this implicit aliasing is not intended in the case of single table inheritance, it didn’t really “work” fully and was very misleading, since it wouldn’t always appear.
References: #3233
The ON clause rendered when using
Query.join()
,Query.outerjoin()
, or the standalonejoin()
/outerjoin()
functions to a single-inheritance subclass will now include the “single table criteria” in the ON clause even if the ON clause is otherwise hand-rolled; it is now added to the criteria using AND, the same way as if joining to a single-table target using relationship or similar.This is sort of in-between feature and bug.
References: #3222
A major rework to the behavior of expression labels, most specifically when used with ColumnProperty constructs with custom SQL expressions and in conjunction with the “order by labels” logic first introduced in 0.9. Fixes include that an
order_by(Entity.some_col_prop)
will now make use of “order by label” rules even if Entity has been subject to aliasing, either via inheritance rendering or via the use of thealiased()
construct; rendering of the same column property multiple times with aliasing (e.g.query(Entity.some_prop, entity_alias.some_prop)
) will label each occurrence of the entity with a distinct label, and additionally “order by label” rules will work for both (e.g.order_by(Entity.some_prop, entity_alias.some_prop)
). Additional issues that could prevent the “order by label” logic from working in 0.9, most notably that the state of a Label could change such that “order by label” would stop working depending on how things were called, has been fixed.Changed the approach by which the “single inheritance criterion” is applied, when using
Query.from_self()
, or its common userQuery.count()
. The criteria to limit rows to those with a certain type is now indicated on the inside subquery, not the outside one, so that even if the “type” column is not available in the columns clause, we can filter on it on the “inner” query.References: #3177
Made a small adjustment to the mechanics of lazy loading, such that it has less chance of interfering with a joinload() in the very rare circumstance that an object points to itself; in this scenario, the object refers to itself while loading its attributes which can cause a mixup between loaders. The use case of “object points to itself” is not fully supported, but the fix also removes some overhead so for now is part of testing.
References: #3145
The “resurrect” ORM event has been removed. This event hook had no purpose since the old “mutable attribute” system was removed in 0.8.
References: #3171
Fixed bug where attribute “set” events or columns with
@validates
would have events triggered within the flush process, when those columns were the targets of a “fetch and populate” operation, such as an autoincremented primary key, a Python side default, or a server-side default “eagerly” fetched via RETURNING.References: #3167
The
IdentityMap
exposed fromSession.identity_map
now returns lists foritems()
andvalues()
in Py3K. Early porting to Py3K here had these returning iterators, when they technically should be “iterable views”..for now, lists are OK.The “evaluator” for query.update()/delete() won’t work with multi-table updates, and needs to be set to synchronize_session=False or synchronize_session=’fetch’; this now raises an exception, with a message to change the synchronize setting. This is upgraded from a warning emitted as of 0.9.7.
References: #3117
Adjustment to attribute mechanics concerning when a value is implicitly initialized to None via first access; this action, which has always resulted in a population of the attribute, no longer does so; the None value is returned but the underlying attribute receives no set event. This is consistent with how collections work and allows attribute mechanics to behave more consistently; in particular, getting an attribute with no value does not squash the event that should proceed if the value is actually set to None.
where bound parameters are rendered inline as strings based on a compile-time option. Work on this feature is courtesy of Dobes Vandermeer.
References: #3061
orm declarative¶
The
declared_attr
construct has newly improved behaviors and features in conjunction with declarative. The decorated function will now have access to the final column copies present on the local mixin when invoked, and will also be invoked exactly once for each mapped class, the returned result being memoized. A new modifierdeclared_attr.cascading
is added as well.References: #3150
Fixed “‘NoneType’ object has no attribute ‘concrete’” error when using
AbstractConcreteBase
in conjunction with a subclass that declares__abstract__
.This change is also backported to: 0.9.8
References: #3185
Fixed bug where using an
__abstract__
mixin in the middle of a declarative inheritance hierarchy would prevent attributes and configuration being correctly propagated from the base class to the inheriting class.A relationship set up with
declared_attr
on aAbstractConcreteBase
base class will now be configured on the abstract base mapping automatically, in addition to being set up on descendant concrete classes as usual.References: #2670
examples¶
Added a new example illustrating materialized paths, using the latest relationship features. Example courtesy Jack Zhou.
This change is also backported to: 0.9.5
A new suite of examples dedicated to providing a detailed study into performance of SQLAlchemy ORM and Core, as well as the DBAPI, from multiple perspectives. The suite runs within a container that provides built in profiling displays both through console output as well as graphically via the RunSnake tool.
See also
Updated the Versioning with a History Table example such that mapped columns are re-mapped to match column names as well as grouping of columns; in particular, this allows columns that are explicitly grouped in a same-column-named joined inheritance scenario to be mapped in the same way in the history mappings, avoiding warnings added in the 0.9 series regarding this pattern and allowing the same view of attribute keys.
This change is also backported to: 0.9.9
Fixed a bug in the examples/generic_associations/discriminator_on_association.py example, where the subclasses of AddressAssociation were not being mapped as “single table inheritance”, leading to problems when trying to use the mappings further.
This change is also backported to: 0.9.9
engine¶
Added new user-space accessors for viewing transaction isolation levels;
Connection.get_isolation_level()
,Connection.default_isolation_level
.This change is also backported to: 0.9.9
Added new event
ConnectionEvents.handle_error()
, a more fully featured and comprehensive replacement forConnectionEvents.dbapi_error()
.This change is also backported to: 0.9.7
References: #3076
A new style of warning can be emitted which will “filter” up to N occurrences of a parameterized string. This allows parameterized warnings that can refer to their arguments to be delivered a fixed number of times until allowing Python warning filters to squelch them, and prevents memory from growing unbounded within Python’s warning registries.
References: #3178
Fixed bug in
Connection
and pool where theConnection.invalidate()
method, or an invalidation due to a database disconnect, would fail if theisolation_level
parameter had been used withConnection.execution_options()
; the “finalizer” that resets the isolation level would be called on the no longer opened connection.This change is also backported to: 0.9.9
References: #3302
A warning is emitted if the
isolation_level
parameter is used withConnection.execution_options()
when aTransaction
is in play; DBAPIs and/or SQLAlchemy dialects such as psycopg2, MySQLdb may implicitly rollback or commit the transaction, or not change the setting til next transaction, so this is never safe.This change is also backported to: 0.9.9
References: #3296
The execution options passed to an
Engine
either viacreate_engine.execution_options
orEngine.update_execution_options()
are not passed to the specialConnection
used to initialize the dialect within the “first connect” event; dialects will usually perform their own queries in this phase, and none of the current available options should be applied here. In particular, the “autocommit” option was causing an attempt to autocommit within this initial connect which would fail with an AttributeError due to the non-standard state of theConnection
.This change is also backported to: 0.9.8
References: #3200
The string keys that are used to determine the columns impacted for an INSERT or UPDATE are now sorted when they contribute towards the “compiled cache” cache key. These keys were previously not deterministically ordered, meaning the same statement could be cached multiple times on equivalent keys, costing both in terms of memory as well as performance.
This change is also backported to: 0.9.8
References: #3165
Fixed bug which would occur if a DBAPI exception occurs when the engine first connects and does its initial checks, and the exception is not a disconnect exception, yet the cursor raises an error when we try to close it. In this case the real exception would be quashed as we tried to log the cursor close exception via the connection pool and failed, as we were trying to access the pool’s logger in a way that is inappropriate in this very specific scenario.
This change is also backported to: 0.9.5
References: #3063
Fixed some “double invalidate” situations were detected where a connection invalidation could occur within an already critical section like a connection.close(); ultimately, these conditions are caused by the change in #2907, in that the “reset on return” feature calls out to the Connection/Transaction in order to handle it, where “disconnect detection” might be caught. However, it’s possible that the more recent change in #2985 made it more likely for this to be seen as the “connection invalidate” operation is much quicker, as the issue is more reproducible on 0.9.4 than 0.9.3.
Checks are now added within any section that an invalidate might occur to halt further disallowed operations on the invalidated connection. This includes two fixes both at the engine level and at the pool level. While the issue was observed with highly concurrent gevent cases, it could in theory occur in any kind of scenario where a disconnect occurs within the connection close operation.
This change is also backported to: 0.9.5
References: #3043
The engine-level error handling and wrapping routines will now take effect in all engine connection use cases, including when user-custom connect routines are used via the
create_engine.creator
parameter, as well as when theConnection
encounters a connection error on revalidation.References: #3266
Removing (or adding) an event listener at the same time that the event is being run itself, either from inside the listener or from a concurrent thread, now raises a RuntimeError, as the collection used is now an instance of
collections.deque()
and does not support changes while being iterated. Previously, a plain Python list was used where removal from inside the event itself would produce silent failures.References: #3163
sql¶
Liberalized the contract for
Index
a bit in that you can specify atext()
expression as the target; the index no longer needs to have a table-bound column present if the index is to be manually added to the table, either via inline declaration or viaTable.append_constraint()
.This change is also backported to: 0.9.5
References: #3028
Added new flag
between.symmetric
, when set to True renders “BETWEEN SYMMETRIC”. Also added a new negation operator “notbetween_op”, which now allows an expression like~col.between(x, y)
to render as “col NOT BETWEEN x AND y”, rather than a parenthesized NOT string.This change is also backported to: 0.9.5
References: #2990
The SQL compiler now generates the mapping of expected columns such that they are matched to the received result set positionally, rather than by name. Originally, this was seen as a way to handle cases where we had columns returned with difficult-to-predict names, though in modern use that issue has been overcome by anonymous labeling. In this version, the approach basically reduces function call count per-result by a few dozen calls, or more for larger sets of result columns. The approach still degrades into a modern version of the old approach if any discrepancy in size exists between the compiled set of columns versus what was received, so there’s no issue for partially or fully textual compilation scenarios where these lists might not line up.
References: #918
Literal values within a
DefaultClause
, which is invoked when using theColumn.server_default
parameter, will now be rendered using the “inline” compiler, so that they are rendered as-is, rather than as bound parameters.References: #3087
The type of expression is reported when an object passed to a SQL expression unit can’t be interpreted as a SQL fragment; pull request courtesy Ryan P. Kelly.
Added a new parameter
Table.tometadata.name
to theTable.tometadata()
method. Similar toTable.tometadata.schema
, this argument causes the newly copiedTable
to take on the new name instead of the existing one. An interesting capability this adds is that of copying aTable
object to the sameMetaData
target with a new name. Pull request courtesy n.d. parker.Exception messages have been spiffed up a bit. The SQL statement and parameters are not displayed if None, reducing confusion for error messages that weren’t related to a statement. The full module and classname for the DBAPI-level exception is displayed, making it clear that this is a wrapped DBAPI exception. The statement and parameters themselves are bounded within a bracketed sections to better isolate them from the error message and from each other.
References: #3172
Insert.from_select()
now includes Python and SQL-expression defaults if otherwise unspecified; the limitation where non- server column defaults aren’t included in an INSERT FROM SELECT is now lifted and these expressions are rendered as constants into the SELECT statement.The
UniqueConstraint
construct is now included when reflecting aTable
object, for databases where this is applicable. In order to achieve this with sufficient accuracy, MySQL and PostgreSQL now contain features that correct for the duplication of indexes and unique constraints when reflecting tables, indexes, and constraints. In the case of MySQL, there is not actually a “unique constraint” concept independent of a “unique index”, so for this backendUniqueConstraint
continues to remain non-present for a reflectedTable
. For PostgreSQL, the query used to detect indexes againstpg_index
has been improved to check for the same construct inpg_constraint
, and the implicitly constructed unique index is not included with a reflectedTable
.In both cases, the
Inspector.get_indexes()
and theInspector.get_unique_constraints()
methods return both constructs individually, but include a new tokenduplicates_constraint
in the case of PostgreSQL orduplicates_index
in the case of MySQL to indicate when this condition is detected. Pull request courtesy Johannes Erdfelt.References: #3184
Added new method
Select.with_statement_hint()
and ORM methodQuery.with_statement_hint()
to support statement-level hints that are not specific to a table.References: #3206
The
info
parameter has been added as a constructor argument to all schema constructs includingMetaData
,Index
,ForeignKey
,ForeignKeyConstraint
,UniqueConstraint
,PrimaryKeyConstraint
,CheckConstraint
.References: #2963
The
Table.autoload_with
flag now implies thatTable.autoload
should beTrue
. Pull request courtesy Malik Diarra.References: #3027
The
Select.limit()
andSelect.offset()
methods now accept any SQL expression, in addition to integer values, as arguments. Typically this is used to allow a bound parameter to be passed, which can be substituted with a value later thus allowing Python-side caching of the SQL query. The implementation here is fully backwards compatible with existing third party dialects, however those dialects which implement special LIMIT/OFFSET systems will need modification in order to take advantage of the new capabilities. Limit and offset also support “literal_binds” mode,References: #3034
The
column()
andtable()
constructs are now importable from the “from sqlalchemy” namespace, just like every other Core construct.The implicit conversion of strings to
text()
constructs when passed to most builder methods ofselect()
as well asQuery
now emits a warning with just the plain string sent. The textual conversion still proceeds normally, however. The only method that accepts a string without a warning are the “label reference” methods like order_by(), group_by(); these functions will now at compile time attempt to resolve a single string argument to a column or label expression present in the selectable; if none is located, the expression still renders, but you get the warning again. The rationale here is that the implicit conversion from string to text is more unexpected than not these days, and it is better that the user send more direction to the Core / ORM when passing a raw string as to what direction should be taken. Core/ORM tutorials have been updated to go more in depth as to how text is handled.References: #2992
Fixed bug in
Enum
and otherSchemaType
subclasses where direct association of the type with aMetaData
would lead to a hang when events (like create events) were emitted on theMetaData
.This change is also backported to: 0.9.7, 0.8.7
References: #3124
Fixed a bug within the custom operator plus
TypeEngine.with_variant()
system, whereby using aTypeDecorator
in conjunction with variant would fail with an MRO error when a comparison operator was used.This change is also backported to: 0.9.7, 0.8.7
References: #3102
Fixed bug in INSERT..FROM SELECT construct where selecting from a UNION would wrap the union in an anonymous (e.g. unlabeled) subquery.
This change is also backported to: 0.9.5, 0.8.7
References: #3044
Fixed bug where
Table.update()
andTable.delete()
would produce an empty WHERE clause when an emptyand_()
oror_()
or other blank expression were applied. This is now consistent with that ofselect()
.This change is also backported to: 0.9.5, 0.8.7
References: #3045
Added the
native_enum
flag to the__repr__()
output ofEnum
, which is mostly important when using it with Alembic autogenerate. Pull request courtesy Dimitris Theodorou.This change is also backported to: 0.9.9
Fixed bug where using a
TypeDecorator
that implemented a type that was also aTypeDecorator
would fail with Python’s “Cannot create a consistent method resolution order (MRO)” error, when any kind of SQL comparison expression were used against an object using this type.This change is also backported to: 0.9.9
References: #3278
Fixed issue where the columns from a SELECT embedded in an INSERT, either through the values clause or as a “from select”, would pollute the column types used in the result set produced by the RETURNING clause when columns from both statements shared the same name, leading to potential errors or mis-adaptation when retrieving the returning rows.
This change is also backported to: 0.9.9
References: #3248
Fixed bug where a fair number of SQL elements within the sql package would fail to
__repr__()
successfully, due to a missingdescription
attribute that would then invoke a recursion overflow when an internal AttributeError would then re-invoke__repr__()
.This change is also backported to: 0.9.8
References: #3195
An adjustment to table/index reflection such that if an index reports a column that isn’t found to be present in the table, a warning is emitted and the column is skipped. This can occur for some special system column situations as has been observed with Oracle.
This change is also backported to: 0.9.8
References: #3180
Fixed bug in CTE where
literal_binds
compiler argument would not be always be correctly propagated when one CTE referred to another aliased CTE in a statement.This change is also backported to: 0.9.8
References: #3154
Fixed 0.9.7 regression caused by #3067 in conjunction with a mis-named unit test such that so-called “schema” types like
Boolean
andEnum
could no longer be pickled.This change is also backported to: 0.9.8
Fix bug in naming convention feature where using a check constraint convention that includes
constraint_name
would then force allBoolean
andEnum
types to require names as well, as these implicitly create a constraint, even if the ultimate target backend were one that does not require generation of the constraint such as PostgreSQL. The mechanics of naming conventions for these particular constraints has been reorganized such that the naming determination is done at DDL compile time, rather than at constraint/table construction time.This change is also backported to: 0.9.7
References: #3067
Fixed bug in common table expressions whereby positional bound parameters could be expressed in the wrong final order when CTEs were nested in certain ways.
This change is also backported to: 0.9.7
References: #3090
Fixed bug where multi-valued
Insert
construct would fail to check subsequent values entries beyond the first one given for literal SQL expressions.This change is also backported to: 0.9.7
References: #3069
Added a “str()” step to the dialect_kwargs iteration for Python version < 2.6.5, working around the “no unicode keyword arg” bug as these args are passed along as keyword args within some reflection processes.
This change is also backported to: 0.9.7
References: #3123
The
TypeEngine.with_variant()
method will now accept a type class as an argument which is internally converted to an instance, using the same convention long established by other constructs such asColumn
.This change is also backported to: 0.9.7
References: #3122
The
Column.nullable
flag is implicitly set toFalse
when thatColumn
is referred to in an explicitPrimaryKeyConstraint
for that table. This behavior now matches that of when theColumn
itself has theColumn.primary_key
flag set toTrue
, which is intended to be an exactly equivalent case.This change is also backported to: 0.9.5
References: #3023
Fixed bug where the
Operators.__and__()
,Operators.__or__()
andOperators.__invert__()
operator overload methods could not be overridden within a customComparator
implementation.This change is also backported to: 0.9.5
References: #3012
Fixed bug in new
DialectKWArgs.argument_for()
method where adding an argument for a construct not previously included for any special arguments would fail.This change is also backported to: 0.9.5
References: #3024
Fixed regression introduced in 0.9 where new “ORDER BY <labelname>” feature from #1068 would not apply quoting rules to the label name as rendered in the ORDER BY.
This change is also backported to: 0.9.5
Restored the import for
Function
to thesqlalchemy.sql.expression
import namespace, which was removed at the beginning of 0.9.This change is also backported to: 0.9.5
The multi-values version of
Insert.values()
has been repaired to work more usefully with tables that have Python- side default values and/or functions, as well as server-side defaults. The feature will now work with a dialect that uses “positional” parameters; a Python callable will also be invoked individually for each row just as is the case with an “executemany” style invocation; a server- side default column will no longer implicitly receive the value explicitly specified for the first row, instead refusing to invoke without an explicit value.References: #3288
Fixed bug in
Table.tometadata()
method where theCheckConstraint
associated with aBoolean
orEnum
type object would be doubled in the target table. The copy process now tracks the production of this constraint object as local to a type object.References: #3260
The behavioral contract of the
ForeignKeyConstraint.columns
collection has been made consistent; this attribute is now aColumnCollection
like that of all other constraints and is initialized at the point when the constraint is associated with aTable
.References: #3243
The
Column.key
attribute is now used as the source of anonymous bound parameter names within expressions, to match the existing use of this value as the key when rendered in an INSERT or UPDATE statement. This allowsColumn.key
to be used as a “substitute” string to work around a difficult column name that doesn’t translate well into a bound parameter name. Note that the paramstyle is configurable oncreate_engine()
in any case, and most DBAPIs today support a named and positional style.References: #3245
Fixed the name of the
PoolEvents.reset.dbapi_connection
parameter as passed to this event; in particular this affects usage of the “named” argument style for this event. Pull request courtesy Jason Goldberger.Reversing a change that was made in 0.9, the “singleton” nature of the “constants”
null()
,true()
, andfalse()
has been reverted. These functions returning a “singleton” object had the effect that different instances would be treated as the same regardless of lexical use, which in particular would impact the rendering of the columns clause of a SELECT statement.References: #3170
Fixed bug where a “branched” connection, that is the kind you get when you call
Connection.connect()
, would not share invalidation status with the parent. The architecture of branching has been tweaked a bit so that the branched connection defers to the parent for all invalidation status and operations.References: #3215
Fixed bug where a “branched” connection, that is the kind you get when you call
Connection.connect()
, would not share transaction status with the parent. The architecture of branching has been tweaked a bit so that the branched connection defers to the parent for all transactional status and operations.References: #3190
Using
Insert.from_select()
now impliesinline=True
oninsert()
. This helps to fix a bug where an INSERT…FROM SELECT construct would inadvertently be compiled as “implicit returning” on supporting backends, which would cause breakage in the case of an INSERT that inserts zero rows (as implicit returning expects a row), as well as arbitrary return data in the case of an INSERT that inserts multiple rows (e.g. only the first row of many). A similar change is also applied to an INSERT..VALUES with multiple parameter sets; implicit RETURNING will no longer emit for this statement either. As both of these constructs deal with variable numbers of rows, theResultProxy.inserted_primary_key
accessor does not apply. Previously, there was a documentation note that one may preferinline=True
with INSERT..FROM SELECT as some databases don’t support returning and therefore can’t do “implicit” returning, but there’s no reason an INSERT…FROM SELECT needs implicit returning in any case. Regular explicitInsert.returning()
should be used to return variable numbers of result rows if inserted data is needed.References: #3169
Custom dialects that implement
GenericTypeCompiler
can now be constructed such that the visit methods receive an indication of the owning expression object, if any. Any visit method that accepts keyword arguments (e.g.**kw
) will in most cases receive a keyword argumenttype_expression
, referring to the expression object that the type is contained within. For columns in DDL, the dialect’s compiler class may need to alter itsget_column_specification()
method to support this as well. TheUserDefinedType.get_col_spec()
method will also receivetype_expression
if it provides**kw
in its argument signature.References: #3074
schema¶
The DDL generation system of
MetaData.create_all()
andMetaData.drop_all()
has been enhanced to in most cases automatically handle the case of mutually dependent foreign key constraints; the need for theForeignKeyConstraint.use_alter
flag is greatly reduced. The system also works for constraints which aren’t given a name up front; only in the case of DROP is a name required for at least one of the constraints involved in the cycle.References: #3282
Added a new accessor
Table.foreign_key_constraints
to complement theTable.foreign_keys
collection, as well asForeignKeyConstraint.referred_table
.The
CheckConstraint
construct now supports naming conventions that include the token%(column_0_name)s
; the constraint expression is scanned for columns. Additionally, naming conventions for check constraints that don’t include the%(constraint_name)s
token will now work forSchemaType
- generated constraints, such as those ofBoolean
andEnum
; this stopped working in 0.9.7 due to #3067.
postgresql¶
Added support for the
CONCURRENTLY
keyword with PostgreSQL indexes, established usingpostgresql_concurrently
. Pull request courtesy Iuri de Silvio.See also
postgresql_index_concurrently
This change is also backported to: 0.9.9
Support is added for “sane multi row count” with the pg8000 driver, which applies mostly to when using versioning with the ORM. The feature is version-detected based on pg8000 1.9.14 or greater in use. Pull request courtesy Tony Locke.
This change is also backported to: 0.9.8
Added kw argument
postgresql_regconfig
to theColumnOperators.match()
operator, allows the “reg config” argument to be specified to theto_tsquery()
function emitted. Pull request courtesy Jonathan Vanasco.This change is also backported to: 0.9.7
References: #3078
Added support for PostgreSQL JSONB via
JSONB
. Pull request courtesy Damian Dimmich.This change is also backported to: 0.9.7
Added support for AUTOCOMMIT isolation level when using the pg8000 DBAPI. Pull request courtesy Tony Locke.
This change is also backported to: 0.9.5
Added a new flag
ARRAY.zero_indexes
to the PostgreSQLARRAY
type. When set toTrue
, a value of one will be added to all array index values before passing to the database, allowing better interoperability between Python style zero-based indexes and PostgreSQL one-based indexes. Pull request courtesy Alexey Terentev.This change is also backported to: 0.9.5
References: #2785
The PG8000 dialect now supports the
create_engine.encoding
parameter, by setting up the client encoding on the connection which is then intercepted by pg8000. Pull request courtesy Tony Locke.Added support for PG8000’s native JSONB feature. Pull request courtesy Tony Locke.
Added support for the psycopg2cffi DBAPI on pypy. Pull request courtesy shauns.
See also
sqlalchemy.dialects.postgresql.psycopg2cffi
References: #3052
Added support for the FILTER keyword as applied to aggregate functions, supported by PostgreSQL 9.4. Pull request courtesy Ilja Everilä.
See also
Support has been added for reflection of materialized views and foreign tables, as well as support for materialized views within
Inspector.get_view_names()
, and a new methodPGInspector.get_foreign_table_names()
available on the PostgreSQL version ofInspector
. Pull request courtesy Rodrigo Menezes.References: #2891
Added support for PG table options TABLESPACE, ON COMMIT, WITH(OUT) OIDS, and INHERITS, when rendering DDL via the
Table
construct. Pull request courtesy malikdiarra.See also
postgresql_table_options
References: #2051
Added new method
PGInspector.get_enums()
, when using the inspector for PostgreSQL will provide a list of ENUM types. Pull request courtesy Ilya Pekelny.Added the
hashable=False
flag to the PGHSTORE
type, which is needed to allow the ORM to skip over trying to “hash” an ORM-mapped HSTORE column when requesting it in a mixed column/entity list. Patch courtesy Gunnlaugur Þór Briem.This change is also backported to: 0.9.5, 0.8.7
References: #3053
Added a new “disconnect” message “connection has been closed unexpectedly”. This appears to be related to newer versions of SSL. Pull request courtesy Antti Haapala.
This change is also backported to: 0.9.5, 0.8.7
Repaired support for PostgreSQL UUID types in conjunction with the ARRAY type when using psycopg2. The psycopg2 dialect now employs use of the psycopg2.extras.register_uuid() hook so that UUID values are always passed to/from the DBAPI as UUID() objects. The
UUID.as_uuid
flag is still honored, except with psycopg2 we need to convert returned UUID objects back into strings when this is disabled.This change is also backported to: 0.9.9
References: #2940
Added support for the
postgresql.JSONB
datatype when using psycopg2 2.5.4 or greater, which features native conversion of JSONB data so that SQLAlchemy’s converters must be disabled; additionally, the newly added psycopg2 extensionextras.register_default_jsonb
is used to establish a JSON deserializer passed to the dialect via thejson_deserializer
argument. Also repaired the PostgreSQL integration tests which weren’t actually round-tripping the JSONB type as opposed to the JSON type. Pull request courtesy Mateusz Susik.This change is also backported to: 0.9.9
Repaired the use of the “array_oid” flag when registering the HSTORE type with older psycopg2 versions < 2.4.3, which does not support this flag, as well as use of the native json serializer hook “register_default_json” with user-defined
json_deserializer
on psycopg2 versions < 2.5, which does not include native json.This change is also backported to: 0.9.9
Fixed bug where PostgreSQL dialect would fail to render an expression in an
Index
that did not correspond directly to a table-bound column; typically when atext()
construct was one of the expressions within the index; or could misinterpret the list of expressions if one or more of them were such an expression.This change is also backported to: 0.9.9
References: #3174
A revisit to this issue first patched in 0.9.5, apparently psycopg2’s
.closed
accessor is not as reliable as we assumed, so we have added an explicit check for the exception messages “SSL SYSCALL error: Bad file descriptor” and “SSL SYSCALL error: EOF detected” when detecting an is-disconnect scenario. We will continue to consult psycopg2’s connection.closed as a first check.This change is also backported to: 0.9.8
References: #3021
Fixed bug where PostgreSQL JSON type was not able to persist or otherwise render a SQL NULL column value, rather than a JSON-encoded
'null'
. To support this case, changes are as follows:The value
null()
can now be specified, which will always result in a NULL value resulting in the statement.A new parameter
JSON.none_as_null
is added, which when True indicates that the PythonNone
value should be persisted as SQL NULL, rather than JSON-encoded'null'
.
Retrieval of NULL as None is also repaired for DBAPIs other than psycopg2, namely pg8000.
This change is also backported to: 0.9.8
References: #3159
The exception wrapping system for DBAPI errors can now accommodate non-standard DBAPI exceptions, such as the psycopg2 TransactionRollbackError. These exceptions will now be raised using the closest available subclass in
sqlalchemy.exc
, in the case of TransactionRollbackError,sqlalchemy.exc.OperationalError
.This change is also backported to: 0.9.8
References: #3075
Fixed bug in
array
object where comparison to a plain Python list would fail to use the correct array constructor. Pull request courtesy Andrew.This change is also backported to: 0.9.8
References: #3141
Added a supported
FunctionElement.alias()
method to functions, e.g. thefunc
construct. Previously, behavior for this method was undefined. The current behavior mimics that of pre-0.9.4, which is that the function is turned into a single-column FROM clause with the given alias name, where the column itself is anonymously named.This change is also backported to: 0.9.8
References: #3137
Fixed bug introduced in 0.9.5 by new pg8000 isolation level feature where engine-level isolation level parameter would raise an error on connect.
This change is also backported to: 0.9.7
References: #3134
The psycopg2
.closed
accessor is now consulted when determining if an exception is a “disconnect” error; ideally, this should remove the need for any other inspection of the exception message to detect disconnect, however we will leave those existing messages in place as a fallback. This should be able to handle newer cases like “SSL EOF” conditions. Pull request courtesy Dirk Mueller.This change is also backported to: 0.9.5
References: #3021
The PostgreSQL
ENUM
type will emit a DROP TYPE instruction when a plaintable.drop()
is called, assuming the object is not associated directly with aMetaData
object. In order to accommodate the use case of an enumerated type shared between multiple tables, the type should be associated directly with theMetaData
object; in this case the type will only be created at the metadata level, or if created directly. The rules for create/drop of PostgreSQL enumerated types have been highly reworked in general.References: #3319
The
PGDialect.has_table()
method will now query againstpg_catalog.pg_table_is_visible(c.oid)
, rather than testing for an exact schema match, when the schema name is None; this so that the method will also illustrate that temporary tables are present. Note that this is a behavioral change, as PostgreSQL allows a non-temporary table to silently overwrite an existing temporary table of the same name, so this changes the behavior ofcheckfirst
in that unusual scenario.References: #3264
Added a new type
OID
to the PostgreSQL dialect. While “oid” is generally a private type within PG that is not exposed in modern versions, there are some PG use cases such as large object support where these types might be exposed, as well as within some user-reported schema reflection use cases.This change is also backported to: 0.9.5
References: #3002
mysql¶
The MySQL dialect now renders TIMESTAMP with NULL / NOT NULL in all cases, so that MySQL 5.6.6 with the
explicit_defaults_for_timestamp
flag enabled will will allow TIMESTAMP to continue to work as expected whennullable=False
. Existing applications are unaffected as SQLAlchemy has always emitted NULL for a TIMESTAMP column that isnullable=True
.References: #3155
Updated the “supports_unicode_statements” flag to True for MySQLdb and Pymysql under Python 2. This refers to the SQL statements themselves, not the parameters, and affects issues such as table and column names using non-ASCII characters. These drivers both appear to support Python 2 Unicode objects without issue in modern versions.
References: #3121
The
gaerdbms
dialect is no longer necessary, and emits a deprecation warning. Google now recommends using the MySQLdb dialect directly.This change is also backported to: 0.9.9
References: #3275
MySQL error 2014 “commands out of sync” appears to be raised as a ProgrammingError, not OperationalError, in modern MySQL-Python versions; all MySQL error codes that are tested for “is disconnect” are now checked within OperationalError and ProgrammingError regardless.
This change is also backported to: 0.9.7, 0.8.7
References: #3101
Fixed bug where column names added to
mysql_length
parameter on an index needed to have the same quoting for quoted names in order to be recognized. The fix makes the quotes optional but also provides the old behavior for backwards compatibility with those using the workaround.This change is also backported to: 0.9.5, 0.8.7
References: #3085
Added support for reflecting tables where an index includes KEY_BLOCK_SIZE using an equal sign. Pull request courtesy Sean McGivern.
This change is also backported to: 0.9.5, 0.8.7
Added a version check to the MySQLdb dialect surrounding the check for ‘utf8_bin’ collation, as this fails on MySQL server < 5.0.
This change is also backported to: 0.9.9
References: #3274
Mysqlconnector as of version 2.0, probably as a side effect of the python 3 merge, now does not expect percent signs (e.g. as used as the modulus operator and others) to be doubled, even when using the “pyformat” bound parameter format (this change is not documented by Mysqlconnector). The dialect now checks for py2k and for mysqlconnector less than version 2.0 when detecting if the modulus operator should be rendered as
%%
or%
.This change is also backported to: 0.9.8
Unicode SQL is now passed for MySQLconnector version 2.0 and above; for Py2k and MySQL < 2.0, strings are encoded.
This change is also backported to: 0.9.8
The MySQL dialect now supports CAST on types that are constructed as
TypeDecorator
objects.A warning is emitted when
cast()
is used with the MySQL dialect on a type where MySQL does not support CAST; MySQL only supports CAST on a subset of datatypes. SQLAlchemy has for a long time just omitted the CAST for unsupported types in the case of MySQL. While we don’t want to change this now, we emit a warning to show that it’s taken place. A warning is also emitted when a CAST is used with an older MySQL version (< 4) that doesn’t support CAST at all, it’s skipped in this case as well.References: #3237
The
SET
type has been overhauled to no longer assume that the empty string, or a set with a single empty string value, is in fact a set with a single empty string; instead, this is by default treated as the empty set. In order to handle persistence of aSET
that actually wants to include the blank value''
as a legitimate value, a new bitwise operational mode is added which is enabled by theSET.retrieve_as_bitwise
flag, which will persist and retrieve values unambiguously using their bitflag positioning. Storage and retrieval of unicode values for driver configurations that aren’t converting unicode natively is also repaired.References: #3283
The
ColumnOperators.match()
operator is now handled such that the return type is not strictly assumed to be boolean; it now returns aBoolean
subclass calledMatchType
. The type will still produce boolean behavior when used in Python expressions, however the dialect can override its behavior at result time. In the case of MySQL, while the MATCH operator is typically used in a boolean context within an expression, if one actually queries for the value of a match expression, a floating point value is returned; this value is not compatible with SQLAlchemy’s C-based boolean processor, so MySQL’s result-set behavior now follows that of theFloat
type. A new operator objectnotmatch_op
is also added to better allow dialects to define the negation of a match operation.References: #3263
MySQL boolean symbols “true”, “false” work again. 0.9’s change in #2682 disallowed the MySQL dialect from making use of the “true” and “false” symbols in the context of “IS” / “IS NOT”, but MySQL supports this syntax even though it has no boolean type. MySQL remains “non native boolean”, but the
true()
andfalse()
symbols again produce the keywords “true” and “false”, so that an expression likecolumn.is_(true())
again works on MySQL.References: #3186
The MySQL dialect will now disable
ConnectionEvents.handle_error()
events from firing for those statements which it uses internally to detect if a table exists or not. This is achieved using an execution optionskip_user_error_events
that disables the handle error event for the scope of that execution. In this way, user code that rewrites exceptions doesn’t need to worry about the MySQL dialect or other dialects that occasionally need to catch SQLAlchemy specific exceptions.Changed the default value of “raise_on_warnings” to False for MySQLconnector. This was set at True for some reason. The “buffered” flag unfortunately must stay at True as MySQLconnector does not allow a cursor to be closed unless all results are fully fetched.
References: #2515
sqlite¶
Added support for partial indexes (e.g. with a WHERE clause) on SQLite. Pull request courtesy Kai Groner.
See also
sqlite_partial_index
This change is also backported to: 0.9.9
Added a new SQLite backend for the SQLCipher backend. This backend provides for encrypted SQLite databases using the pysqlcipher Python driver, which is very similar to the pysqlite driver.
See also
pysqlcipher
This change is also backported to: 0.9.9
When selecting from a UNION using an attached database file, the pysqlite driver reports column names in cursor.description as ‘dbname.tablename.colname’, instead of ‘tablename.colname’ as it normally does for a UNION (note that it’s supposed to just be ‘colname’ for both, but we work around it). The column translation logic here has been adjusted to retrieve the rightmost token, rather than the second token, so it works in both cases. Workaround courtesy Tony Roberts.
This change is also backported to: 0.9.8
References: #3211
Fixed a SQLite join rewriting issue where a subquery that is embedded as a scalar subquery such as within an IN would receive inappropriate substitutions from the enclosing query, if the same table were present inside the subquery as were in the enclosing query such as in a joined inheritance scenario.
This change is also backported to: 0.9.7
References: #3130
UNIQUE and FOREIGN KEY constraints are now fully reflected on SQLite both with and without names. Previously, foreign key names were ignored and unnamed unique constraints were skipped. Thanks to Jon Nelson for assistance with this.
The SQLite dialect, when using the
DATE
,TIME
, orDATETIME
types, and given astorage_format
that only renders numbers, will render the types in DDL asDATE_CHAR
,TIME_CHAR
, andDATETIME_CHAR
, so that despite the lack of alpha characters in the values, the column will still deliver the “text affinity”. Normally this is not needed, as the textual values within the default storage formats already imply text.See also
sqlite_datetime
References: #3257
SQLite now supports reflection of unique constraints from temp tables; previously, this would fail with a TypeError. Pull request courtesy Johannes Erdfelt.
See also
SQLite/Oracle have distinct methods for temporary table/view name reporting - changes regarding SQLite temporary table and view reflection.
References: #3203
Added
Inspector.get_temp_table_names()
andInspector.get_temp_view_names()
; currently, only the SQLite and Oracle dialects support these methods. The return of temporary table and view names has been removed from SQLite and Oracle’s version ofInspector.get_table_names()
andInspector.get_view_names()
; other database backends cannot support this information (such as MySQL), and the scope of operation is different in that the tables can be local to a session and typically aren’t supported in remote schemas.References: #3204
mssql¶
Enabled “multivalues insert” for SQL Server 2008. Pull request courtesy Albert Cervin. Also expanded the checks for “IDENTITY INSERT” mode to include when the identity key is present in the VALUEs clause of the statement.
This change is also backported to: 0.9.7
SQL Server 2012 now recommends VARCHAR(max), NVARCHAR(max), VARBINARY(max) for large text/binary types. The MSSQL dialect will now respect this based on version detection, as well as the new
deprecate_large_types
flag.See also
mssql_large_type_deprecation
References: #3039
The hostname-based connection format for SQL Server when using pyodbc will no longer specify a default “driver name”, and a warning is emitted if this is missing. The optimal driver name for SQL Server changes frequently and is per-platform, so hostname based connections need to specify this. DSN-based connections are preferred.
References: #3182
Added statement encoding to the “SET IDENTITY_INSERT” statements which operate when an explicit INSERT is being interjected into an IDENTITY column, to support non-ascii table identifiers on drivers such as pyodbc + unix + py2k that don’t support unicode statements.
This change is also backported to: 0.9.7, 0.8.7
In the SQL Server pyodbc dialect, repaired the implementation for the
description_encoding
dialect parameter, which when not explicitly set was preventing cursor.description from being parsed correctly in the case of result sets that contained names in alternate encodings. This parameter shouldn’t be needed going forward.This change is also backported to: 0.9.7, 0.8.7
References: #3091
Fixed the version string detection in the pymssql dialect to work with Microsoft SQL Azure, which changes the word “SQL Server” to “SQL Azure”.
This change is also backported to: 0.9.8
References: #3151
Revised the query used to determine the current default schema name to use the
database_principal_id()
function in conjunction with thesys.database_principals
view so that we can determine the default schema independently of the type of login in progress (e.g., SQL Server, Windows, etc).This change is also backported to: 0.9.5
References: #3025
oracle¶
Added support for cx_oracle connections to a specific service name, as opposed to a tns name, by passing
?service_name=<name>
to the URL. Pull request courtesy Sławomir Ehlert.New Oracle DDL features for tables, indexes: COMPRESS, BITMAP. Patch courtesy Gabor Gombas.
Added support for CTEs under Oracle. This includes some tweaks to the aliasing syntax, as well as a new CTE feature
CTE.suffix_with()
, which is useful for adding in special Oracle-specific directives to the CTE.See also
References: #3220
Added support for the Oracle table option ON COMMIT.
Fixed long-standing bug in Oracle dialect where bound parameter names that started with numbers would not be quoted, as Oracle doesn’t like numerics in bound parameter names.
This change is also backported to: 0.9.8
References: #2138
Fixed bug in oracle dialect test suite where in one test, ‘username’ was assumed to be in the database URL, even though this might not be the case.
This change is also backported to: 0.9.7
References: #3128
An alias name will be properly quoted when referred to using the
%(name)s
token inside theSelect.with_hint()
method. Previously, the Oracle backend hadn’t implemented this quoting.
tests¶
Fixed bug where “python setup.py test” wasn’t calling into distutils appropriately, and errors would be emitted at the end of the test suite.
This change is also backported to: 0.9.7
Corrected for some deprecation warnings involving the
imp
module and Python 3.3 or greater, when running tests. Pull request courtesy Matt Chisholm.This change is also backported to: 0.9.5
References: #2830
misc¶
Added a new extension suite
sqlalchemy.ext.baked
. This simple but unusual system allows for a dramatic savings in Python overhead for the construction and processing of ormQuery
objects, from query construction up through rendering of a string SQL statement.See also
References: #3054
The
sqlalchemy.ext.automap
extension will now setcascade="all, delete-orphan"
automatically on a one-to-many relationship/backref where the foreign key is detected as containing one or more non-nullable columns. This argument is present in the keywords passed togenerate_relationship()
in this case and can still be overridden. Additionally, if theForeignKeyConstraint
specifiesondelete="CASCADE"
for a non-nullable orondelete="SET NULL"
for a nullable set of columns, the argumentpassive_deletes=True
is also added to the relationship. Note that not all backends support reflection of ondelete, but backends that do include PostgreSQL and MySQL.References: #3210
The
__mapper_args__
dictionary is copied from a declarative mixin or abstract class when accessed, so that modifications made to this dictionary by declarative itself won’t conflict with that of other mappings. The dictionary is modified regarding theversion_id_col
andpolymorphic_on
arguments, replacing the column within with the one that is officially mapped to the local class/table.This change is also backported to: 0.9.5, 0.8.7
References: #3062
Fixed bug in mutable extension where
MutableDict
did not report change events for thesetdefault()
dictionary operation.This change is also backported to: 0.9.5, 0.8.7
Fixed bug where
MutableDict.setdefault()
didn’t return the existing or new value (this bug was not released in any 0.8 version). Pull request courtesy Thomas Hervé.This change is also backported to: 0.9.5, 0.8.7
Fixed bug where the association proxy list class would not interpret slices correctly under Py3K. Pull request courtesy Gilles Dartiguelongue.
This change is also backported to: 0.9.9
Fixed an unlikely race condition observed in some exotic end-user setups, where the attempt to check for “duplicate class name” in declarative would hit upon a not-totally-cleaned-up weak reference related to some other class being removed; the check here now ensures the weakref still references an object before calling upon it further.
This change is also backported to: 0.9.8
References: #3208
Fixed bug in ordering list where the order of items would be thrown off during a collection replace event, if the reorder_on_append flag were set to True. The fix ensures that the ordering list only impacts the list that is explicitly associated with the object.
This change is also backported to: 0.9.8
References: #3191
Fixed bug where
MutableDict
failed to implement theupdate()
dictionary method, thus not catching changes. Pull request courtesy Matt Chisholm.This change is also backported to: 0.9.8
Fixed bug where a custom subclass of
MutableDict
would not show up in a “coerce” operation, and would instead return a plainMutableDict
. Pull request courtesy Matt Chisholm.This change is also backported to: 0.9.8
Fixed bug in connection pool logging where the “connection checked out” debug logging message would not emit if the logging were set up using
logging.setLevel()
, rather than using theecho_pool
flag. Tests to assert this logging have been added. This is a regression that was introduced in 0.9.0.This change is also backported to: 0.9.8
References: #3168
Fixed bug when the declarative
__abstract__
flag was not being distinguished for when it was actually the valueFalse
. The__abstract__
flag needs to actually evaluate to a True value at the level being tested.This change is also backported to: 0.9.7
References: #3097
In public test suite, changed to use of
String(40)
from less-supportedText
inStringTest.test_literal_backslashes
. Pullreq courtesy Jan.This change is also backported to: 0.9.5
The Drizzle dialect has been removed from the Core; it is now available as sqlalchemy-drizzle, an independent, third party dialect. The dialect is still based almost entirely off of the MySQL dialect present in SQLAlchemy.