Mike Bayer [Wed, 18 Mar 2015 22:57:13 +0000 (18:57 -0400)]
- 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.
fixes #3331
Mike Bayer [Tue, 17 Mar 2015 16:32:33 +0000 (12:32 -0400)]
- The "auto close" for :class:`.ResultProxy` is now a "soft" close.
That is, after exhausing 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 if :meth:`.ResultProxy.close`
is called explicitly will these methods raise the "result is closed"
error.
fixes #3330 fixes #3329
Mike Bayer [Thu, 12 Mar 2015 14:14:52 +0000 (10:14 -0400)]
- try to document how to get columns from constraints.
unfortunately Sphinx refuses to work correctly for the columns
attribute so we just add a lame message to contains_column().
Mike Bayer [Thu, 12 Mar 2015 00:22:42 +0000 (20:22 -0400)]
- Added a new extension suite :mod:`sqlalchemy.ext.baked`. This
simple but unusual system allows for a dramatic savings in Python
overhead for the construction and processing of orm :class:`.Query`
objects, from query construction up through rendering of a string
SQL statement.
fixes #3054
Mike Bayer [Wed, 11 Mar 2015 18:46:52 +0000 (14:46 -0400)]
- Added a new entry ``"entity"`` to the dictionaries returned by
:attr:`.Query.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.
references #3320
Mike Bayer [Wed, 11 Mar 2015 15:41:52 +0000 (11:41 -0400)]
- The Postgresql :class:`.postgresql.ENUM` type will emit a
DROP TYPE instruction when a plain ``table.drop()`` is called,
assuming the object is not associated directly with a
:class:`.MetaData` object. In order to accomodate the use case of
an enumerated type shared between multiple tables, the type should
be associated directly with the :class:`.MetaData` 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.
fixes #3319
Mike Bayer [Tue, 10 Mar 2015 23:56:59 +0000 (19:56 -0400)]
- Added a new event suite :class:`.QueryEvents`. The
:meth:`.QueryEvents.before_compile` event allows the creation
of functions which may place additional modifications to
:class:`.Query` 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 against
:class:`.Query` objects in an automated fashion.
fixes #3317
Mike Bayer [Tue, 10 Mar 2015 22:23:23 +0000 (18:23 -0400)]
- 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
:paramref:`.relationship.uselist` set to False. This will produce
more efficient queries in these cases.
fixes #3249
Mike Bayer [Tue, 10 Mar 2015 21:51:35 +0000 (17:51 -0400)]
- 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.
fixes #3301
Mike Bayer [Tue, 10 Mar 2015 21:21:46 +0000 (17:21 -0400)]
- 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.
fixes #3219 fixes #3240
Mike Bayer [Mon, 9 Mar 2015 19:24:37 +0000 (15:24 -0400)]
- fix a potential race condition where the per-mapper LRUCache used by
persistence.py could theoretically hit the limit of the cache
(100 items by default) and at some points fail to have a key that
we check for, due to the cleanup. This has never been observed
so its likely that so far, the total number of INSERT, UPDATE and
DELETE statement structures in real apps has not exceeded 100
on a per-mapper basis; this could happen for apps that run a
very wide variety of attribute modified combinations into the unit
of work, *and* which have very high concurrency going on.
This change will be a lot more significant when we open up
use of LRUCache + compiled cache with the baked query extension.
Mike Bayer [Mon, 9 Mar 2015 19:24:01 +0000 (15:24 -0400)]
- reduce some callcounts related to the lightweight tuple,
basically by caching them.
- improve LRUCache so that size management only proceeds
in one thread, add a real get() method
Mike Bayer [Mon, 9 Mar 2015 16:26:30 +0000 (12:26 -0400)]
- we really don't want to force lib here, but we do need
it for the plain "pytest" case, so do an append instead of an insert
- add a new tox file to do a full regen of callcounts
Mike Bayer [Sun, 8 Mar 2015 03:51:12 +0000 (22:51 -0500)]
- the change for #918 was of course not nearly that simple.
The "wrapping" employed by the mssql and oracle dialects using the
"iswrapper" argument was not being used intelligently by the compiler,
and the result map was being written incorrectly, using
*more* columns in the result map than were actually returned by
the statement, due to "row number" columns that are inside the
subquery. The compiler now writes out result map on the
"top level" select in all cases
fully, and for the mssql/oracle wrapping case extracts out
the "proxied" columns in a second step, which only includes
those columns that are proxied outwards to the top level.
This change might have implications for 3rd party dialects that
might be imitating oracle's approach. They can safely continue
to use the "iswrapper" kw which is now ignored, but they may
need to also add the _select_wraps argument as well.
Mike Bayer [Sat, 7 Mar 2015 17:48:13 +0000 (12:48 -0500)]
- 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 textual elements modify the result
map, or 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. fixes #918
- callcounts still need to be adjusted down for this so zoomark
tests won't pass at the moment
Mike Bayer [Sun, 1 Mar 2015 21:09:11 +0000 (16:09 -0500)]
- squash-merge the final row_proc integration branch. this is
a much more modest outcome than what we started with. The
work of create_row_processor() for ColumnProperty objects
is essentially done at query setup time combined with some
lookups in _instance_processor().
- to allow this change for deferred columns, deferred columns
no longer search for themselves in the result. If they've been
set up as deferred without any explicit directive to undefer them,
then this is what was asked for. if we don't do this,
then we're stuck with this performance penalty for all deferred
columns which in the vast majority of typical use cases (e.g. loading
large, legacy tables or tables with many/large very seldom
used values) won't be present in the result and won't be accessed at all.
Mike Bayer [Tue, 24 Feb 2015 20:29:30 +0000 (15:29 -0500)]
- repair issue in declared_attr.cascading such that within a
subclass, the value returned by the descriptor is not available
because the superclass is already mapped with the InstrumentedAttribute,
until the subclass is mapped. We add a setattr() to set up that
attribute so that the __mapper_args__ hook and possibly others
have access to the "cascaded" version of the attribute within
the call.
Mike Bayer [Fri, 20 Feb 2015 20:14:08 +0000 (15:14 -0500)]
- 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.
fixes #3310
Mike Bayer [Fri, 20 Feb 2015 05:01:19 +0000 (00:01 -0500)]
- Fixed bug where internal assertion would fail in the case where
an ``after_rollback()`` handler for a :class:`.Session` incorrectly
adds state to that :class:`.Session` within the handler, and the task
to warn and remove this state (established by :ticket:`2389`) attempts
to proceed.
fixes #3309
Mike Bayer [Wed, 18 Feb 2015 21:08:19 +0000 (16:08 -0500)]
- 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 :meth:`.Session.commit` and
for :meth:`.Session.expire_all`, as well as in the "cleanup" step
which occurs when object states are garbage collected.
fixes #3307
Mike Bayer [Mon, 9 Feb 2015 20:29:14 +0000 (15:29 -0500)]
- A warning is emitted when :func:`.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.
fixes #3237
Mike Bayer [Mon, 9 Feb 2015 19:58:26 +0000 (14:58 -0500)]
- Literal values within a :class:`.DefaultClause`, which is invoked
when using the :paramref:`.Column.server_default` parameter, will
now be rendered using the "inline" compiler, so that they are rendered
as-is, rather than as bound parameters.
fixes #3087
Mike Bayer [Thu, 5 Feb 2015 21:17:23 +0000 (16:17 -0500)]
- 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 when
``nullable=False``. Existing applications are unaffected as
SQLAlchemy has always emitted NULL for a TIMESTAMP column that
is ``nullable=True``. fixes #3155
Mike Bayer [Wed, 4 Feb 2015 23:51:24 +0000 (18:51 -0500)]
- Fixed bug in :class:`.Connection` and pool where the
:meth:`.Connection.invalidate` method, or an invalidation due
to a database disconnect, would fail if the
``isolation_level`` parameter had been used with
:meth:`.Connection.execution_options`; the "finalizer" that resets
the isolation level would be called on the no longer opened connection.
fixes #3302
Mike Bayer [Wed, 4 Feb 2015 22:07:15 +0000 (17:07 -0500)]
- A warning is emitted if the ``isolation_level`` parameter is used
with :meth:`.Connection.execution_options` when a :class:`.Transaction`
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.
- Added new parameter :paramref:`.Session.connection.execution_options`
which may be used to set up execution options on a :class:`.Connection`
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.
- added new documentation section
detailing best practices for setting transaction isolation with
sessions.
fixes #3296
Mike Bayer [Tue, 3 Feb 2015 00:46:13 +0000 (19:46 -0500)]
- 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.
fixes #3300
Mike Bayer [Mon, 2 Feb 2015 00:00:07 +0000 (19:00 -0500)]
- 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 :paramref:`.UUID.as_uuid` flag is still
honored, except with psycopg2 we need to convert returned
UUID objects back into strings when this is disabled.
fixes #2940
Mike Bayer [Sun, 1 Feb 2015 00:04:54 +0000 (19:04 -0500)]
- Added support for the :class:`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 extension
``extras.register_default_jsonb`` is used to establish a JSON
deserializer passed to the dialect via the ``json_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.
- 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.
Mike Bayer [Fri, 30 Jan 2015 18:38:51 +0000 (13:38 -0500)]
- The :class:`.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 for :class:`.SchemaType`-
generated constraints, such as those of :class:`.Boolean` and
:class:`.Enum`; this stopped working in 0.9.7 due to :ticket:`3067`.
fixes #3299
Mike Bayer [Wed, 28 Jan 2015 16:48:20 +0000 (11:48 -0500)]
- Fixed bug in 0.9's foreign key setup system, such that
the logic used to link a :class:`.ForeignKey` to its parent could fail
when the foreign key used "link_to_name=True" in conjunction with
a target :class:`.Table` that would not receive its parent column until
later, such as within a reflection + "useexisting" scenario,
if the target column in fact had a key value different from its name,
as would occur in reflection if column reflect events were used to
alter the .key of reflected :class:`.Column` objects so that the
link_to_name becomes significant. Also repaired support for column
type via FK transmission in a similar way when target columns had a
different key and were referenced using link_to_name.
fixes #3298
Mike Bayer [Mon, 26 Jan 2015 23:02:46 +0000 (18:02 -0500)]
- additional test adjustments for pypy / psycopg2cffi. This
consists mainly of adjusting fixtures to ensure connections are closed
explicitly. psycopg2cffi also handles unicode bind parameter
names differently than psycopg2, and seems to possibly have a little less
control over floating point values at least in one test which is
marked as a "fail", though will see if it runs differently on linux
than osx..
- changelog for psycopg2cffi, fixes #3052