Mike Bayer [Mon, 25 Jun 2018 04:23:54 +0000 (00:23 -0400)]
Compare mappers more accurately in Load._chop_path
Fixed bug in cache key generation for baked queries which could cause a
too-short cache key to be generated for the case of eager loads across
subclasses. This could in turn cause the eagerload query to be cached in
place of a non-eagerload query, or vice versa, for a polymorhic "selectin"
load, or possibly for lazy loads or selectin loads as well.
Mike Bayer [Fri, 15 Jun 2018 02:17:00 +0000 (22:17 -0400)]
render WITH clause after INSERT for INSERT..SELECT on Oracle, MySQL
Fixed INSERT FROM SELECT with CTEs for the Oracle and MySQL dialects, where
the CTE was being placed above the entire statement as is typical with
other databases, however Oracle and MariaDB 10.2 wants the CTE underneath
the "INSERT" segment. Note that the Oracle and MySQL dialects don't yet
work when a CTE is applied to a subquery inside of an UPDATE or DELETE
statement, as the CTE is still applied to the top rather than inside the
subquery.
Also adds test suite support CTEs against backends.
Mike Bayer [Fri, 15 Jun 2018 02:56:21 +0000 (22:56 -0400)]
Lookup index columns in parent table by key for copy
Fixed regression in 1.2 due to :ticket:`4147` where a :class:`.Table` that
has had some of its indexed columns redefined with new ones, as would occur
when overriding columns during reflection or when using
:paramref:`.Table.extend_existing`, such that the :meth:`.Table.tometadata`
method would fail when attempting to copy those indexes as they still
referred to the replaced column. The copy logic now accommodates for this
condition.
Mike Bayer [Wed, 6 Jun 2018 20:35:34 +0000 (16:35 -0400)]
Iterate options per path for baked cache key
Fixed an issue that was both a performance regression in 1.2 as well as an
incorrect result regarding the "baked" lazy loader, involving the
generation of cache keys from the original :class:`.Query` object's loader
options. If the loader options were built up in a "branched" style using
common base elements for multiple options, the same options would be
rendered into the cache key repeatedly, causing both a performance issue as
well as generating the wrong cache key. This is fixed, along with a
performance improvement when such "branched" options are applied via
:meth:`.Query.options` to prevent the same option objects from being
applied repeatedly.
Mike Bayer [Mon, 28 May 2018 17:03:14 +0000 (13:03 -0400)]
Add Query.lazy_load_from attribute for sharding
Added new attribute :attr:`.Query.lazy_loaded_from` which is populated
with an :class:`.InstanceState` that is using this :class:`.Query` in
order to lazy load a relationship. The rationale for this is that
it serves as a hint for the horizontal sharding feature to use, such that
the identity token of the state can be used as the default identity token
to use for the query within id_chooser().
Also repaired an issue in the :meth:`.Result.with_post_criteria`
method added in I899808734458e25a023142c2c5bb37cbed869479
for :ticket:`4128` where the "unbake subquery loaders" version was calling
the post crtieria functions given the :class:`.Result` as the argument
rather than applying them to the :class:`.Query`.
Mike Bayer [Fri, 1 Jun 2018 21:54:11 +0000 (16:54 -0500)]
Support undocumented non-entity sequence Query argument
Fixed regression caused by :ticket:`4256` (itself a regression fix for
:ticket:`4228`) which breaks an undocumented behavior which converted for a
non-sequence of entities passed directly to the :class:`.Query` constructor
into a single-element sequence. While this behavior was never supported or
documented, it's already in use so has been added as a behavioral contract
to :class:`.Query`.
Mike Bayer [Sun, 27 May 2018 17:45:20 +0000 (13:45 -0400)]
Mutex on _CONFIGURE_MUTEX in automap.prepare()
Fixed a race condition which could occur if automap
:meth:`.AutomapBase.prepare` were used within a multi-threaded context
against other threads which may call :func:`.configure_mappers` as a
result of use of other mappers. The unfinished mapping work of automap
is particularly sensitive to being pulled in by a
:func:`.configure_mappers` step leading to errors.
Mike Bayer [Wed, 23 May 2018 20:22:48 +0000 (16:22 -0400)]
Turn oracle BINARY_DOUBLE, BINARY_FLOAT, DOUBLE_PRECISION into floats
The Oracle BINARY_FLOAT and BINARY_DOUBLE datatypes now participate within
cx_Oracle.setinputsizes(), passing along NATIVE_FLOAT, so as to support the
NaN value. Additionally, :class:`.oracle.BINARY_FLOAT`,
:class:`.oracle.BINARY_DOUBLE` and :class:`.oracle.DOUBLE_PRECISION` now
subclass :class:`.Float`, since these are floating point datatypes, not
decimal. These datatypes were already defaulting the
:paramref:`.Float.asdecimal` flag to False in line with what
:class:`.Float` already does.
Added reflection capabilities for the :class:`.oracle.BINARY_FLOAT`,
:class:`.oracle.BINARY_DOUBLE` datatypes.
Mike Bayer [Fri, 18 May 2018 16:51:40 +0000 (12:51 -0400)]
call setinputsizes() for integer types
Altered the Oracle dialect such that when an :class:`.Integer` type is in
use, the cx_Oracle.NUMERIC type is set up for setinputsizes(). In
SQLAlchemy 1.1 and earlier, cx_Oracle.NUMERIC was passed for all numeric
types unconditionally, and in 1.2 this was removed to allow for better
numeric precision. However, for integers, some database/client setups
will fail to coerce boolean values True/False into integers which introduces
regressive behavior when using SQLAlchemy 1.2. Overall, the setinputsizes
logic seems like it will need a lot more flexibility going forward so this
is a start for that.
Mike Bayer [Thu, 17 May 2018 23:57:01 +0000 (19:57 -0400)]
Skip for SQL Server on non-native boolean unconstrained
As SQL Server is now non-native boolean as of
I4765d2a2a00b0d14f50282603cc4d48d4739dac1 but uses the BIT
type, we need to constrain this test to continue to not
run against SQL Server.
Mike Bayer [Wed, 16 May 2018 20:31:40 +0000 (16:31 -0400)]
Update dogpile.cache example to be compatible with baked query.
Updated the dogpile.caching example to include new structures that
accommodate for the "baked" query system, which is used by default within
lazy loaders and some eager relationship loaders. The dogpile.caching
"relationship_caching" and "advanced" examples were also broken due to
:ticket:`4256`. The issue here is also worked-around by the fix in
:ticket:`4128`.
Note that this recipe requires
I3f86fcb12a6a9a89aa308b335e75c25969bcc30e in order for the
"advanced" example to work.
Miguel Ventura [Mon, 14 May 2018 20:56:58 +0000 (16:56 -0400)]
Fix string formatting TypeError if tuple is passed
Fixed issue where the "ambiguous literal" error message used when
interpreting literal values as SQL expression values would encounter a
tuple value, and fail to format the message properly. Pull request courtesy
Miguel Ventura.
Mike Bayer [Wed, 16 May 2018 19:47:50 +0000 (15:47 -0400)]
Clarify dogpile.cache differences which occur via #4128.
The fix for :ticket:`4128` which failed to be included
in the 1.2.5 release also works around an issue in the dogpile.cache
"advanced" example, which will also be fixed independently.
Mike Bayer [Wed, 16 May 2018 15:16:57 +0000 (11:16 -0400)]
Change query._identity_lookup into a normal instance method
Fixed regression in 1.2.7 caused by :ticket:`4228`, which itself was fixing
a 1.2-level regression, where the ``query_cls`` callable passed to a
:class:`.Session` was assumed to be a subclass of :class:`.Query` with
class method availability, as opposed to an arbitrary callable. In
particular, the dogpile caching example illustrates ``query_cls`` as a
function and not a :class:`.Query` subclass.
Mike Bayer [Mon, 26 Feb 2018 00:54:37 +0000 (19:54 -0500)]
Restore missing changelog files for #4128
The fix merged in 4a31c30fa5ebd6af0e72937b21b2e5ee79e12631
failed to get backported to 1.2 and the changelog files
got blown away in the release process. Restore the
changelog files to master before cherry-picking
the whole thing to 1.2.
Mike Bayer [Tue, 15 May 2018 20:15:51 +0000 (16:15 -0400)]
Prevent double-checkins and guard during reset-on-return invalidations
Fixed connection pool issue whereby if a disconnection error were raised
during the connection pool's "reset on return" sequence in conjunction with
an explicit transaction opened against the enclosing :class:`.Connection`
object (such as from calling :meth:`.Session.close` without a rollback or
commit, or calling :meth:`.Connection.close` without first closing a
transaction declared with :meth:`.Connection.begin`), a double-checkin would
result, which could then lead towards concurrent checkouts of the same
connection. The double-checkin condition is now prevented overall by an
assertion, as well as the specific double-checkin scenario has been
fixed.
Mike Bayer [Fri, 11 May 2018 13:19:10 +0000 (09:19 -0400)]
Default server_version_info to (0, )
Fixed a bug in the test suite where if an external dialect returned
``None`` for ``server_version_info``, the exclusion logic would raise an
``AttributeError``.
Mike Bayer [Thu, 10 May 2018 15:39:06 +0000 (11:39 -0400)]
SQL Server is not native boolean; add new flag for CHECK constraint
Fixed a 1.2 regression caused by :ticket:`4061` where the SQL Server
"BIT" type would be considered to be "native boolean". The goal here
was to avoid creating a CHECK constraint on the column, however the bigger
issue is that the BIT value does not behave like a true/false constant
and cannot be interpreted as a standalone expression, e.g.
"WHERE <column>". The SQL Server dialect now goes back to being
non-native boolean, but with an extra flag that still avoids creating
the CHECK constraint.
Mike Bayer [Thu, 3 May 2018 16:35:23 +0000 (12:35 -0400)]
Use identity_token for refresh(), unexpire, undefer
The horizontal sharding extension now makes use of the identity token
added to ORM identity keys as part of :ticket:`4137`, when an object
refresh or column-based deferred load or unexpiration operation occurs.
Since we know the "shard" that the object originated from, we make
use of this value when refreshing, thereby avoiding queries against
other shards that don't match this object's identity in any case.
Mike Bayer [Mon, 30 Apr 2018 15:31:48 +0000 (11:31 -0400)]
Render FOR UPDATE on the inner subquery as well as the outer
The ORM now doubles the "FOR UPDATE" clause within the subquery that
renders in conjunction with joined eager loading in some cases, as it has
been observed that MySQL does not lock the rows from a subquery. This
means the query renders with two FOR UPDATE clauses; note that on some
backends such as Oracle, FOR UPDATE clauses on subqueries are silently
ignored since they are unnecessary. Additionally, in the case of the "OF"
clause used primarily with Postgresql, the FOR UPDATE is rendered only on
the inner subquery when this is used so that the selectable can be targeted
to the table within the SELECT statement.
Fixed a reference leak issue where the values of the parameter dictionary
used in a statement execution would remain referenced by the "compiled
cache", as a result of storing the key view used by Python 3 dictionary
keys(). Pull request courtesy Olivier Grisel.
Mike Bayer [Fri, 20 Apr 2018 15:44:09 +0000 (11:44 -0400)]
Refactor "get" to allow for pluggable identity token schemes
Fixed regression in 1.2 within sharded query feature where the
new "identity_token" element was not being correctly considered within
the scope of a lazy load operation, when searching the identity map
for a related many-to-one element. The new behavior will allow for
making use of the "id_chooser" in order to determine the best identity
key to retrieve from the identity map. In order to achieve this, some
refactoring of 1.2's "identity_token" approach has made some slight changes
to the implementation of ``ShardedQuery`` which should be noted for other
derivations of this class.
Mike Bayer [Fri, 20 Apr 2018 17:31:45 +0000 (13:31 -0400)]
Document how to opt-out of NCHAR for cx_Oracle
Unfortunately, we need to bind Python unicode values as
NCHAR as in the case where non-ascii characters are present,
it's necessary. We can't know in all cases how this value is being
used, so in those cases where Oracle will not accept NCHAR the
user should explicitly cast a value down to String.
Mike Bayer [Thu, 19 Apr 2018 21:07:32 +0000 (17:07 -0400)]
Ensure select_from_entity adapter is used in adjust_for_single_inheritance
Fixed issue in single-inheritance loading where the use of an aliased
entity against a single-inheritance subclass in conjunction with the
:meth:`.Query.select_from` method would cause the SQL to be rendered with
the unaliased table mixed in to the query, causing a cartesian product. In
particular this was affecting the new "selectin" loader when used against a
single-inheritance subclass.
Mike Bayer [Tue, 17 Apr 2018 17:22:13 +0000 (13:22 -0400)]
Render and reflect MySQL WITH PARSER index options
Support added for the "WITH PARSER" syntax of CREATE FULLTEXT INDEX
in MySQL, using the ``mysql_with_parser`` keyword argument. Reflection
is also supported, which accommodates MySQL's special comment format
for reporting on this option as well. Additionally, the "FULLTEXT" and
"SPATIAL" index prefixes are now reflected back into the ``mysql_prefix``
index option.
Kent Bower [Wed, 11 Apr 2018 21:21:26 +0000 (17:21 -0400)]
Reflect Oracle NUMBER(NULL, 0) as INTEGER
The Oracle NUMBER datatype is reflected as INTEGER if the precision is NULL
and the scale is zero, as this is how INTEGER values come back when
reflected from Oracle's tables. Pull request courtesy Kent Bower.
Mike Bayer [Wed, 11 Apr 2018 13:29:00 +0000 (09:29 -0400)]
Correct join for FKs with schema in SQL Server
Fixed 1.2 regression caused by :ticket:`4060` where the query used to
reflect SQL Server cross-schema foreign keys was limiting the criteria
incorrectly.
Additionally, added some rework of the inter-schema reflection tests
so that MySQL, MSSQL can be included, breaking out some of the
Postgresql-specific behaviors into separate requirements.
Mike Bayer [Fri, 6 Apr 2018 15:39:15 +0000 (11:39 -0400)]
Raise informative exception for non-sortable PK
An informative exception is re-raised when a primary key value is not
sortable in Python during an ORM flush under Python 3, such as an ``Enum``
that has no ``__lt__()`` method; normally Python 3 raises a ``TypeError``
in this case. The flush process sorts persistent objects by primary key
in Python so the values must be sortable.
Mike Bayer [Wed, 4 Apr 2018 17:36:28 +0000 (13:36 -0400)]
Ensure all visit_sequence accepts **kw args
Fixed issue where the compilation of an INSERT statement with the
"literal_binds" option that also uses an explicit sequence and "inline"
generation, as on Postgresql and Oracle, would fail to accommodate the
extra keyword argument within the sequence processing routine.
Mike Bayer [Tue, 3 Apr 2018 19:35:00 +0000 (15:35 -0400)]
Use base __ne__ implementation for range types w/ None
Fixed bug where the special "not equals" operator for the Postgresql
"range" datatypes such as DATERANGE would fail to render "IS NOT NULL" when
compared to the Python ``None`` value.
Also break up range tests into backend round trip and straight
compilation suites.
Mike Bayer [Fri, 30 Mar 2018 15:22:30 +0000 (11:22 -0400)]
Filter non-integer characters from pyodbc SQL Server version
Adjusted the SQL Server version detection for pyodbc to only allow for
numeric tokens, filtering out non-integers, since the dialect does tuple-
numeric comparisons with this value. This is normally true for all known
SQL Server / pyodbc drivers in any case.
Mike Bayer [Wed, 28 Mar 2018 15:58:47 +0000 (11:58 -0400)]
Invalidate on failed connect handler
Fixed bug in connection pool where a connection could be present in the
pool without all of its "connect" event handlers called, if a previous
"connect" handler threw an exception; note that the dialects themselves
have connect handlers that emit SQL, such as those which set transaction
isolation, which can fail if the database is in a non-available state, but
still allows a connection. The connection is now invalidated first if any
of the connect handlers fail.
Mike Bayer [Wed, 21 Mar 2018 13:59:46 +0000 (09:59 -0400)]
Don't warn for mixin-based __table_args__, __mapper_args__ declared_attr
Removed a warning that would be emitted when calling upon
``__table_args__``, ``__mapper_args__`` as named with a ``@declared_attr``
method, when called from a non-mapped declarative mixin. Calling these
directly is documented as the approach to use when one is overidding one
of these methods on a mapped class. The warning still emits for regular
attribute names.
Mike Bayer [Wed, 14 Mar 2018 15:04:43 +0000 (11:04 -0400)]
Track if we're rendering within the CTE recursively
Fixed a regression that occurred from the previous fix to :ticket:`4204` in
version 1.2.5, where a CTE that refers to itself after the
:meth:`.CTE.alias` method has been called would not refer to iself
correctly.
Mike Bayer [Mon, 12 Mar 2018 16:50:52 +0000 (12:50 -0400)]
Ignore non-primary mappers within mutable instrumentation
Fixed bug where using :meth:`.Mutable.associate_with` or
:meth:`.Mutable.as_mutable` in conjunction with a class that has non-
primary mappers set up with alternatively-named attributes would produce an
attribute error. Since non-primary mappers are not used for persistence,
the mutable extension now excludes non-primary mappers from its
instrumentation steps.
Mike Bayer [Wed, 7 Mar 2018 21:27:26 +0000 (16:27 -0500)]
Raise cx_Oracle minimum version to 5.2
The minimum cx_Oracle version supported is 5.2 (June 2015). Previously,
the dialect asserted against version 5.0 but as of 1.2.2 we are using some
symbols that did not appear until 5.2.
Mike Bayer [Wed, 7 Mar 2018 21:03:07 +0000 (16:03 -0500)]
Add notes regarding flat=True
Start documenting that flat=True and aliased=True don't work
with selectable particularly when selectable is an aliased select
already. References #4212
Eric Atkin [Wed, 28 Feb 2018 21:00:38 +0000 (16:00 -0500)]
Add Query.enable_single_entity()
Added new feature :meth:`.Query.only_return_tuples`. Causes the
:class:`.Query` object to return keyed tuple objects unconditionally even
if the query is against a single entity. Pull request courtesy Eric
Atkin.
Mike Bayer [Tue, 6 Mar 2018 02:36:18 +0000 (21:36 -0500)]
Clone _cte_alias instead of assigning "self"
Fixed bug in :class:.`CTE` construct along the same lines as that of
:ticket:`4204` where a :class:`.CTE` that was aliased would not copy itself
correctly during a "clone" operation as is frequent within the ORM as well
as when using the :meth:`.ClauseElement.params` method.
Mike Bayer [Mon, 5 Mar 2018 19:48:00 +0000 (14:48 -0500)]
Don't include AliasedClass when pickling options
Fixed 1.2 regression where a mapper option that contains an
:class:`.AliasedClass` object, as is typical when using the
:meth:`.QueryableAttribute.of_type` method, could not be pickled. 1.1's
behavior was to omit the aliased class objects from the path, so this
behavior is restored.
projects / thirdparty / sqlalchemy / sqlalchemy.git / log
