CHANGES_PRE_06

-*- coding: utf-8; fill-column: 68 -*-

===============
CHANGES PRE-0.6
===============

0.5.9
=====
- sql
    - Fixed erroneous self_group() call in expression package.
      [ticket:1661]

0.5.8
=====
- sql
    - The copy() method on Column now supports uninitialized,
      unnamed Column objects. This allows easy creation of 
      declarative helpers which place common columns on multiple 
      subclasses.

    - Default generators like Sequence() translate correctly
      across a copy() operation.

    - Sequence() and other DefaultGenerator objects are accepted 
      as the value for the "default" and "onupdate" keyword
      arguments of Column, in addition to being accepted
      positionally.

    - Fixed a column arithmetic bug that affected column
      correspondence for cloned selectables which contain
      free-standing column expressions.   This bug is
      generally only noticeable when exercising newer 
      ORM behavior only availble in 0.6 via [ticket:1568], 
      but is more correct at the SQL expression level 
      as well. [ticket:1617]

- postgresql
    - The extract() function, which was slightly improved in
      0.5.7, needed a lot more work to generate the correct
      typecast (the typecasts appear to be necessary in PG's
      EXTRACT quite a lot of the time).  The typecast is 
      now generated using a rule dictionary based 
      on PG's documentation for date/time/interval arithmetic.
      It also accepts text() constructs again, which was broken
      in 0.5.7. [ticket:1647]

- firebird 
    - Recognize more errors as disconnections. [ticket:1646]

0.5.7
=====
- orm
    - contains_eager() now works with the automatically 
      generated subquery that results when you say 
      "query(Parent).join(Parent.somejoinedsubclass)", i.e. 
      when Parent joins to a joined-table-inheritance subclass.
      Previously contains_eager() would erroneously add the 
      subclass table to the query separately producing a 
      cartesian product.  An example is in the ticket
      description.  [ticket:1543]

    - query.options() now only propagate to loaded objects
      for potential further sub-loads only for options where
      such behavior is relevant, keeping
      various unserializable options like those generated
      by contains_eager() out of individual instance states.
      [ticket:1553]

    - Session.execute() now locates table- and 
      mapper-specific binds based on a passed
      in expression which is an insert()/update()/delete() 
      construct. [ticket:1054]

    - Session.merge() now properly overwrites a many-to-one or 
      uselist=False attribute to None if the attribute 
      is also None in the given object to be merged.

    - Fixed a needless select which would occur when merging
      transient objects that contained a null primary key
      identifier.  [ticket:1618]

    - Mutable collection passed to the "extension" attribute
      of relation(), column_property() etc. will not be mutated
      or shared among multiple instrumentation calls, preventing
      duplicate extensions, such as backref populators,
      from being inserted into the list.
      [ticket:1585]

    - Fixed the call to get_committed_value() on CompositeProperty.
      [ticket:1504]

    - Fixed bug where Query would crash if a join() with no clear
      "left" side were called when a non-mapped column entity
      appeared in the columns list. [ticket:1602]

    - Fixed bug whereby composite columns wouldn't load properly
      when configured on a joined-table subclass, introduced in
      version 0.5.6 as a result of the fix for [ticket:1480].
      [ticket:1616] thx to Scott Torborg.

    - The "use get" behavior of many-to-one relations, i.e. that a 
      lazy load will fallback to the possibly cached query.get() 
      value, now works across join conditions where the two compared 
      types are not exactly the same class, but share the same 
      "affinity" - i.e. Integer and SmallInteger.  Also allows
      combinations of reflected and non-reflected types to work
      with 0.5 style type reflection, such as PGText/Text (note 0.6 
      reflects types as their generic versions).   [ticket:1556]

    - Fixed bug in query.update() when passing Cls.attribute
      as keys in the value dict and using synchronize_session='expire'
      ('fetch' in 0.6). [ticket:1436]

- sql
    - Fixed bug in two-phase transaction whereby commit() method
      didn't set the full state which allows subsequent close()
      call to succeed. [ticket:1603]

    - Fixed the "numeric" paramstyle, which apparently is the
      default paramstyle used by Informixdb.

    - Repeat expressions in the columns clause of a select
      are deduped based on the identity of each clause element,
      not the actual string.  This allows positional 
      elements to render correctly even if they all render
      identically, such as "qmark" style bind parameters.
      [ticket:1574]

    - The cursor associated with connection pool connections
      (i.e. _CursorFairy) now proxies `__iter__()` to the 
      underlying cursor correctly. [ticket:1632]

    - types now support an "affinity comparison" operation, i.e.
      that an Integer/SmallInteger are "compatible", or 
      a Text/String, PickleType/Binary, etc.  Part of 
      [ticket:1556].

    - Fixed bug preventing alias() of an alias() from being
      cloned or adapted (occurs frequently in ORM operations).
      [ticket:1641]

- sqlite
    - sqlite dialect properly generates CREATE INDEX for a table
      that is in an alternate schema.  [ticket:1439]

- postgresql
    - Added support for reflecting the DOUBLE PRECISION type,
      via a new postgres.PGDoublePrecision object.
      This is postgresql.DOUBLE_PRECISION in 0.6.
      [ticket:1085]

    - Added support for reflecting the INTERVAL YEAR TO MONTH
      and INTERVAL DAY TO SECOND syntaxes of the INTERVAL 
      type.  [ticket:460]

    - Corrected the "has_sequence" query to take current schema,
      or explicit sequence-stated schema, into account.
      [ticket:1576]

    - Fixed the behavior of extract() to apply operator
      precedence rules to the "::" operator when applying
      the "timestamp" cast - ensures proper parenthesization.
      [ticket:1611]

- mssql
    - Changed the name of TrustedConnection to
      Trusted_Connection when constructing pyodbc connect
      arguments [ticket:1561]

- oracle
     - The "table_names" dialect function, used by MetaData
       .reflect(), omits "index overflow tables", a system
       table generated by Oracle when "index only tables"
       with overflow are used.  These tables aren't accessible
       via SQL and can't be reflected.  [ticket:1637]

- ext 
    - A column can be added to a joined-table declarative 
      superclass after the class has been constructed 
      (i.e. via class-level attribute assignment), and
      the column will be propagated down to 
      subclasses. [ticket:1570]  This is the reverse
      situation as that of [ticket:1523], fixed in 0.5.6.

    - Fixed a slight inaccuracy in the sharding example.
      Comparing equivalence of columns in the ORM is best 
      accomplished using col1.shares_lineage(col2).
      [ticket:1491]

    - Removed unused `load()` method from ShardedQuery.
      [ticket:1606]

0.5.6
=====
- orm
    - Fixed bug whereby inheritance discriminator part of a
      composite primary key would fail on updates.
      Continuation of [ticket:1300].

    - Fixed bug which disallowed one side of a many-to-many 
      bidirectional reference to declare itself as "viewonly"
      [ticket:1507]

    - Added an assertion that prevents a @validates function
      or other AttributeExtension from loading an unloaded
      collection such that internal state may be corrupted.
      [ticket:1526]

    - Fixed bug which prevented two entities from mutually
      replacing each other's primary key values within a single
      flush() for some orderings of operations.  [ticket:1519]

    - Fixed an obscure issue whereby a joined-table subclass
      with a self-referential eager load on the base class
      would populate the related object's "subclass" table with
      data from the "subclass" table of the parent.
      [ticket:1485]

    - relations() now have greater ability to be "overridden",
      meaning a subclass that explicitly specifies a relation()
      overriding that of the parent class will be honored
      during a flush.  This is currently to support 
      many-to-many relations from concrete inheritance setups.
      Outside of that use case, YMMV.  [ticket:1477]

    - Squeezed a few more unnecessary "lazy loads" out of 
      relation().  When a collection is mutated, many-to-one
      backrefs on the other side will not fire off to load
      the "old" value, unless "single_parent=True" is set.
      A direct assignment of a many-to-one still loads 
      the "old" value in order to update backref collections 
      on that value, which may be present in the session 
      already, thus maintaining the 0.5 behavioral contract.
      [ticket:1483]

    - Fixed bug whereby a load/refresh of joined table
      inheritance attributes which were based on 
      column_property() or similar would fail to evaluate.
      [ticket:1480]

    - Improved support for MapperProperty objects overriding
      that of an inherited mapper for non-concrete 
      inheritance setups - attribute extensions won't randomly
      collide with each other.  [ticket:1488]

    - UPDATE and DELETE do not support ORDER BY, LIMIT, OFFSET,
      etc. in standard SQL.  Query.update() and Query.delete()
      now raise an exception if any of limit(), offset(),
      order_by(), group_by(), or distinct() have been
      called. [ticket:1487]

    - Added AttributeExtension to sqlalchemy.orm.__all__

    - Improved error message when query() is called with
      a non-SQL /entity expression. [ticket:1476]

    - Using False or 0 as a polymorphic discriminator now
      works on the base class as well as a subclass.
      [ticket:1440]

    - Added enable_assertions(False) to Query which disables
      the usual assertions for expected state - used
      by Query subclasses to engineer custom state.
      [ticket:1424].  See
      http://www.sqlalchemy.org/trac/wiki/UsageRecipes/PreFilteredQuery
      for an example.

    - Fixed recursion issue which occured if a mapped object's
      `__len__()` or `__nonzero__()` method resulted in state
      changes.  [ticket:1501]

    - Fixed incorrect exception raise in 
      Weak/StrongIdentityMap.add() 
      [ticket:1506]

    - Fixed the error message for "could not find a FROM clause"
      in query.join() which would fail to issue correctly
      if the query was against a pure SQL construct.
      [ticket:1522]

    - Fixed a somewhat hypothetical issue which would result
      in the wrong primary key being calculated for a mapper
      using the old polymorphic_union function - but this
      is old stuff.  [ticket:1486]

- sql
    - Fixed column.copy() to copy defaults and onupdates.
      [ticket:1373]

    - Fixed a bug in extract() introduced in 0.5.4 whereby
      the string "field" argument was getting treated as a 
      ClauseElement, causing various errors within more 
      complex SQL transformations.

    - Unary expressions such as DISTINCT propagate their 
      type handling to result sets, allowing conversions like
      unicode and such to take place.  [ticket:1420]

    - Fixed bug in Table and Column whereby passing empty
      dict for "info" argument would raise an exception.
      [ticket:1482]

- oracle
    - Backported 0.6 fix for Oracle alias names not getting
      truncated.  [ticket:1309]

- ext
   - The collection proxies produced by associationproxy are now
     pickleable.  A user-defined proxy_factory however
     is still not pickleable unless it defines __getstate__
     and __setstate__. [ticket:1446]

   - Declarative will raise an informative exception if 
     __table_args__ is passed as a tuple with no dict argument.
     Improved documentation.  [ticket:1468]

   - Table objects declared in the MetaData can now be used
     in string expressions sent to primaryjoin/secondaryjoin/
     secondary - the name is pulled from the MetaData of the
     declarative base.  [ticket:1527]

   - A column can be added to a joined-table subclass after
     the class has been constructed (i.e. via class-level
     attribute assignment).  The column is added to the underlying
     Table as always, but now the mapper will rebuild its
     "join" to include the new column, instead of raising
     an error about "no such column, use column_property()
     instead".  [ticket:1523]

- test
   - Added examples into the test suite so they get exercised
     regularly and cleaned up a couple deprecation warnings.


0.5.5
=======
- general
    - unit tests have been migrated from unittest to nose.  See
      README.unittests for information on how to run the tests.
      [ticket:970]

- orm
    - The "foreign_keys" argument of relation() will now propagate
      automatically to the backref in the same way that primaryjoin
      and secondaryjoin do.  For the extremely rare use case where
      the backref of a relation() has intentionally different
      "foreign_keys" configured, both sides now need to be
      configured explicity (if they do in fact require this setting,
      see the next note...).

    - ...the only known (and really, really rare) use case where a
      different foreign_keys setting was used on the
      forwards/backwards side, a composite foreign key that
      partially points to its own columns, has been enhanced such
      that the fk->itself aspect of the relation won't be used to
      determine relation direction.

    - Session.mapper is now *deprecated*.

      Call session.add() if you'd like a free-standing object to be
      part of your session.  Otherwise, a DIY version of
      Session.mapper is now documented at
      http://www.sqlalchemy.org/trac/wiki/UsageRecipes/SessionAwareMapper
      The method will remain deprecated throughout 0.6.

    - Fixed Query being able to join() from individual columns of a
      joined-table subclass entity, i.e.  query(SubClass.foo,
      SubcClass.bar).join(<anything>).  In most cases, an error
      "Could not find a FROM clause to join from" would be
      raised. In a few others, the result would be returned in terms
      of the base class rather than the subclass - so applications
      which relied on this erroneous result need to be
      adjusted. [ticket:1431]

    - Fixed a bug involving contains_eager(), which would apply
      itself to a secondary (i.e. lazy) load in a particular rare
      case, producing cartesian products.  improved the targeting of
      query.options() on secondary loads overall [ticket:1461].

    - Fixed bug introduced in 0.5.4 whereby Composite types fail
      when default-holding columns are flushed.

    - Fixed another 0.5.4 bug whereby mutable attributes
      (i.e. PickleType) wouldn't be deserialized correctly when the
      whole object was serialized.  [ticket:1426]

    - Fixed bug whereby session.is_modified() would raise an
      exception if any synonyms were in use.

    - Fixed potential memory leak whereby previously pickled objects
      placed back in a session would not be fully garbage collected
      unless the Session were explicitly closed out.

    - Fixed bug whereby list-based attributes, like pickletype and
      PGArray, failed to be merged() properly.

    - Repaired non-working attributes.set_committed_value function.

    - Trimmed the pickle format for InstanceState which should
      further reduce the memory footprint of pickled instances.  The
      format should be backwards compatible with that of 0.5.4 and
      previous.

    - sqlalchemy.orm.join and sqlalchemy.orm.outerjoin are now
      added to __all__ in sqlalchemy.orm.*. [ticket:1463]

    - Fixed bug where Query exception raise would fail when
      a too-short composite primary key value were passed to
      get().  [ticket:1458]

- sql
    - Removed an obscure feature of execute() (including connection,
      engine, Session) whereby a bindparam() construct can be sent
      as a key to the params dictionary.  This usage is undocumented
      and is at the core of an issue whereby the bindparam() object
      created implicitly by a text() construct may have the same
      hash value as a string placed in the params dictionary and may
      result in an inappropriate match when computing the final bind
      parameters.  Internal checks for this condition would add
      significant latency to the critical task of parameter
      rendering, so the behavior is removed.  This is a backwards
      incompatible change for any application that may have been
      using this feature, however the feature has never been
      documented.

- engine/pool
    - Implemented recreate() for StaticPool.


0.5.4p2
=======

- sql
    - Repaired the printing of SQL exceptions which are not 
      based on parameters or are not executemany() style.

- postgresql
    - Deprecated the hardcoded TIMESTAMP function, which when
      used as func.TIMESTAMP(value) would render "TIMESTAMP value".
      This breaks on some platforms as PostgreSQL doesn't allow
      bind parameters to be used in this context.  The hard-coded
      uppercase is also inappropriate and there's lots of other
      PG casts that we'd need to support.  So instead, use
      text constructs i.e. select(["timestamp '12/05/09'"]).


0.5.4p1
=======

- orm
    - Fixed an attribute error introduced in 0.5.4 which would 
      occur when merge() was used with an incomplete object.

0.5.4
=====

- orm
    - Significant performance enhancements regarding Sessions/flush()
      in conjunction with large mapper graphs, large numbers of 
      objects:

      - Removed all* O(N) scanning behavior from the flush() process,
        i.e. operations that were scanning the full session, 
        including an extremely expensive one that was erroneously
        assuming primary key values were changing when this 
        was not the case.

        * one edge case remains which may invoke a full scan,
          if an existing primary key attribute is modified
          to a new value.

      - The Session's "weak referencing" behavior is now *full* -
        no strong references whatsoever are made to a mapped object
        or related items/collections in its __dict__.  Backrefs and 
        other cycles in objects no longer affect the Session's ability 
        to lose all references to unmodified objects.  Objects with 
        pending changes still are maintained strongly until flush.
        [ticket:1398]

        The implementation also improves performance by moving
        the "resurrection" process of garbage collected items
        to only be relevant for mappings that map "mutable" 
        attributes (i.e. PickleType, composite attrs).  This removes
        overhead from the gc process and simplifies internal 
        behavior.

        If a "mutable" attribute change is the sole change on an object 
        which is then dereferenced, the mapper will not have access to 
        other attribute state when the UPDATE is issued.  This may present 
        itself differently to some MapperExtensions.

        The change also affects the internal attribute API, but not
        the AttributeExtension interface nor any of the publically
        documented attribute functions.

      - The unit of work no longer genererates a graph of "dependency"
        processors for the full graph of mappers during flush(), instead
        creating such processors only for those mappers which represent
        objects with pending changes.  This saves a tremendous number
        of method calls in the context of a large interconnected 
        graph of mappers.

      - Cached a wasteful "table sort" operation that previously
        occured multiple times per flush, also removing significant
        method call count from flush().

      - Other redundant behaviors have been simplified in 
        mapper._save_obj().

    - Modified query_cls on DynamicAttributeImpl to accept a full
      mixin version of the AppenderQuery, which allows subclassing
      the AppenderMixin.

    - The "polymorphic discriminator" column may be part of a 
      primary key, and it will be populated with the correct 
      discriminator value.  [ticket:1300]

    - Fixed the evaluator not being able to evaluate IS NULL clauses.

    - Fixed the "set collection" function on "dynamic" relations to
      initiate events correctly.  Previously a collection could only
      be assigned to a pending parent instance, otherwise modified
      events would not be fired correctly.  Set collection is now
      compatible with merge(), fixes [ticket:1352].

    - Allowed pickling of PropertyOption objects constructed with
      instrumented descriptors; previously, pickle errors would occur 
      when pickling an object which was loaded with a descriptor-based 
      option, such as query.options(eagerload(MyClass.foo)).

    - Lazy loader will not use get() if the "lazy load" SQL clause
      matches the clause used by get(), but contains some parameters
      hardcoded.  Previously the lazy strategy would fail with the
      get().  Ideally get() would be used with the hardcoded
      parameters but this would require further development.
      [ticket:1357]

    - MapperOptions and other state associated with query.options()
      is no longer bundled within callables associated with each 
      lazy/deferred-loading attribute during a load.
      The options are now associated with the instance's
      state object just once when it's populated.  This removes
      the need in most cases for per-instance/attribute loader 
      objects, improving load speed and memory overhead for
      individual instances. [ticket:1391]

    - Fixed another location where autoflush was interfering
      with session.merge().  autoflush is disabled completely
      for the duration of merge() now. [ticket:1360]

    - Fixed bug which prevented "mutable primary key" dependency
      logic from functioning properly on a one-to-one
      relation().  [ticket:1406]

    - Fixed bug in relation(), introduced in 0.5.3, 
      whereby a self referential relation
      from a base class to a joined-table subclass would 
      not configure correctly.

    - Fixed obscure mapper compilation issue when inheriting
      mappers are used which would result in un-initialized
      attributes.

    - Fixed documentation for session weak_identity_map - 
      the default value is True, indicating a weak
      referencing map in use.

    - Fixed a unit of work issue whereby the foreign
      key attribute on an item contained within a collection 
      owned by an object being deleted would not be set to
      None if the relation() was self-referential. [ticket:1376]

    - Fixed Query.update() and Query.delete() failures with eagerloaded
      relations. [ticket:1378]

    - It is now an error to specify both columns of a binary primaryjoin
      condition in the foreign_keys or remote_side collection.  Whereas
      previously it was just nonsensical, but would succeed in a 
      non-deterministic way.

- schema
    - Added a quote_schema() method to the IdentifierPreparer class
      so that dialects can override how schemas get handled. This
      enables the MSSQL dialect to treat schemas as multipart
      identifiers, such as 'database.owner'. [ticket: 594, 1341]

- sql
    - Back-ported the "compiler" extension from SQLA 0.6.  This
      is a standardized interface which allows the creation of custom
      ClauseElement subclasses and compilers.  In particular it's
      handy as an alternative to text() when you'd like to 
      build a construct that has database-specific compilations.
      See the extension docs for details.

    - Exception messages are truncated when the list of bound 
      parameters is larger than 10, preventing enormous
      multi-page exceptions from filling up screens and logfiles
      for large executemany() statements. [ticket:1413]

    - ``sqlalchemy.extract()`` is now dialect sensitive and can
      extract components of timestamps idiomatically across the
      supported databases, including SQLite.

    - Fixed __repr__() and other _get_colspec() methods on
      ForeignKey constructed from __clause_element__() style
      construct (i.e. declarative columns).  [ticket:1353]

- mysql
    - Reflecting a FOREIGN KEY construct will take into account
      a dotted schema.tablename combination, if the foreign key
      references a table in a remote schema. [ticket:1405]

- mssql
    - Modified how savepoint logic works to prevent it from
      stepping on non-savepoint oriented routines. Savepoint
      support is still very experimental.

    - Added in reserved words for MSSQL that covers version 2008
      and all prior versions. [ticket:1310]

    - Corrected problem with information schema not working with a
      binary collation based database. Cleaned up information schema
      since it is only used by mssql now. [ticket:1343]

- sqlite
    - Corrected the SLBoolean type so that it properly treats only 1
      as True. [ticket:1402]

    - Corrected the float type so that it correctly maps to a
      SLFloat type when being reflected. [ticket:1273]

- extensions

    - Fixed adding of deferred or other column properties to a
      declarative class. [ticket:1379]

0.5.3
=====
- orm
    - The "objects" argument to session.flush() is deprecated.
      State which represents the linkage between a parent and
      child object does not support "flushed" status on
      one side of the link and not the other, so supporting
      this operation leads to misleading results.
      [ticket:1315]

    - Query now implements __clause_element__() which produces
      its selectable, which means a Query instance can be accepted 
      in many SQL expressions, including col.in_(query), 
      union(query1, query2), select([foo]).select_from(query), 
      etc.

    - Query.join() can now construct multiple FROM clauses, if 
      needed.  Such as, query(A, B).join(A.x).join(B.y)
      might say SELECT A.*, B.* FROM A JOIN X, B JOIN Y.
      Eager loading can also tack its joins onto those 
      multiple FROM clauses.  [ticket:1337]

    - Fixed bug in dynamic_loader() where append/remove events
      after construction time were not being propagated to the 
      UOW to pick up on flush(). [ticket:1347]

    - Fixed bug where column_prefix wasn't being checked before
      not mapping an attribute that already had class-level 
      name present.

    - a session.expire() on a particular collection attribute
      will clear any pending backref additions as well, so that
      the next access correctly returns only what was present
      in the database.  Presents some degree of a workaround for 
      [ticket:1315], although we are considering removing the 
      flush([objects]) feature altogether.

    - Session.scalar() now converts raw SQL strings to text()
      the same way Session.execute() does and accepts same 
      alternative **kw args.

    - improvements to the "determine direction" logic of 
      relation() such that the direction of tricky situations 
      like mapper(A.join(B)) -> relation-> mapper(B) can be
      determined.

    - When flushing partial sets of objects using session.flush([somelist]),
      pending objects which remain pending after the operation won't
      inadvertently be added as persistent. [ticket:1306]

    - Added "post_configure_attribute" method to InstrumentationManager,
      so that the "listen_for_events.py" example works again.
      [ticket:1314]

    - a forward and complementing backwards reference which are both
      of the same direction, i.e. ONETOMANY or MANYTOONE, 
      is now detected, and an error message is raised.
      Saves crazy CircularDependencyErrors later on.

    - Fixed bugs in Query regarding simultaneous selection of 
      multiple joined-table inheritance entities with common base 
      classes:

      - previously the adaption applied to "B" on 
        "A JOIN B" would be erroneously partially applied 
        to "A".

      - comparisons on relations (i.e. A.related==someb)
        were not getting adapted when they should.

      - Other filterings, like 
        query(A).join(A.bs).filter(B.foo=='bar'), were erroneously
        adapting "B.foo" as though it were an "A".

     - Fixed adaptation of EXISTS clauses via any(), has(), etc.
       in conjunction with an aliased object on the left and 
       of_type() on the right.  [ticket:1325]

     - Added an attribute helper method ``set_committed_value`` in 
       sqlalchemy.orm.attributes.  Given an object, attribute name,
       and value, will set the value on the object as part of its
       "committed" state, i.e. state that is understood to have 
       been loaded from the database.   Helps with the creation of 
       homegrown collection loaders and such.

     - Query won't fail with weakref error when a non-mapper/class
       instrumented descriptor is passed, raises 
       "Invalid column expession".

     - Query.group_by() properly takes into account aliasing applied
       to the FROM clause, such as with select_from(), using
       with_polymorphic(), or using from_self().

- sql
    - An alias() of a select() will convert to a "scalar subquery"
      when used in an unambiguously scalar context, i.e. it's used 
      in a comparison operation.  This applies to
      the ORM when using query.subquery() as well.

    - Fixed missing _label attribute on Function object, others
      when used in a select() with use_labels (such as when used
      in an ORM column_property()).  [ticket:1302]

    - anonymous alias names now truncate down to the max length
      allowed by the dialect.  More significant on DBs like
      Oracle with very small character limits. [ticket:1309]

    - the __selectable__() interface has been replaced entirely
      by __clause_element__().

    - The per-dialect cache used by TypeEngine to cache
      dialect-specific types is now a WeakKeyDictionary.
      This to prevent dialect objects from 
      being referenced forever for an application that 
      creates an arbitrarily large number of engines
      or dialects.   There is a small performance penalty
      which will be resolved in 0.6.  [ticket:1299]

- sqlite
    - Fixed SQLite reflection methods so that non-present
      cursor.description, which triggers an auto-cursor
      close, will be detected so that no results doesn't
      fail on recent versions of pysqlite which raise 
      an error when fetchone() called with no rows present.

- postgresql
    - Index reflection won't fail when an index with 
      multiple expressions is encountered.

    - Added PGUuid and PGBit types to 
      sqlalchemy.databases.postgres. [ticket:1327]

    - Refection of unknown PG types won't crash when those
      types are specified within a domain.  [ticket:1327]

- mssql
    - Preliminary support for pymssql 1.0.1

    - Corrected issue on mssql where max_identifier_length was
      not being respected.

- extensions

    - Fixed a recursive pickling issue in serializer, triggered
      by an EXISTS or other embedded FROM construct.

    - Declarative locates the "inherits" class using a search
      through __bases__, to skip over mixins that are local
      to subclasses.

    - Declarative figures out joined-table inheritance primary join
      condition even if "inherits" mapper argument is given 
      explicitly. 

    - Declarative will properly interpret the "foreign_keys" argument
      on a backref() if it's a string.

    - Declarative will accept a table-bound column as a property
      when used in conjunction with __table__, if the column is already
      present in __table__.  The column will be remapped to the given 
      key the same way as when added to the mapper() properties dict.

0.5.2
======

- orm
    - Further refined 0.5.1's warning about delete-orphan cascade
      placed on a many-to-many relation.   First, the bad news:
      the warning will apply to both many-to-many as well as
      many-to-one relations.  This is necessary since in both 
      cases, SQLA does not scan the full set of potential parents
      when determining "orphan" status - for a persistent object
      it only detects an in-python de-association event to establish
      the object as an "orphan".  Next, the good news: to support 
      one-to-one via a foreign key or assocation table, or to 
      support one-to-many via an association table, a new flag 
      single_parent=True may be set which indicates objects 
      linked to the relation are only meant to have a single parent.
      The relation will raise an error if multiple parent-association
      events occur within Python.

    - Adjusted the attribute instrumentation change from 0.5.1 to 
      fully establish instrumentation for subclasses where the mapper
      was created after the superclass had already been fully
      instrumented. [ticket:1292]

    - Fixed bug in delete-orphan cascade whereby two one-to-one
      relations from two different parent classes to the same target 
      class would prematurely expunge the instance.

    - Fixed an eager loading bug whereby self-referential eager 
      loading would prevent other eager loads, self referential or not,
      from joining to the parent JOIN properly.  Thanks to Alex K
      for creating a great test case.

    - session.expire() and related methods will not expire() unloaded
      deferred attributes.  This prevents them from being needlessly
      loaded when the instance is refreshed.

    - query.join()/outerjoin() will now properly join an aliased()
      construct to the existing left side, even if query.from_self()
      or query.select_from(someselectable) has been called.
      [ticket:1293]

- sql
    - Further fixes to the "percent signs and spaces in column/table
       names" functionality. [ticket:1284]

- mssql
    - Restored convert_unicode handling. Results were being passed
      on through without conversion. [ticket:1291]

    - Really fixing the decimal handling this time. [ticket:1282].

    - Modified table reflection code to use only kwargs when 
      constructing tables. [Ticket:1289]

0.5.1
========

- orm
    - Removed an internal join cache which could potentially leak 
      memory when issuing query.join() repeatedly to ad-hoc 
      selectables.

    - The "clear()", "save()", "update()", "save_or_update()"
      Session methods have been deprecated, replaced by 
      "expunge_all()" and "add()".  "expunge_all()" has also
      been added to ScopedSession.

    - Modernized the "no mapped table" exception and added a more
      explicit __table__/__tablename__ exception to declarative.

    - Concrete inheriting mappers now instrument attributes which
      are inherited from the superclass, but are not defined for
      the concrete mapper itself, with an InstrumentedAttribute that
      issues a descriptive error when accessed. [ticket:1237]

    - Added a new `relation()` keyword `back_populates`. This
      allows configuation of backreferences using explicit
      relations. [ticket:781] This is required when creating
      bidirectional relations between a hierarchy of concrete
      mappers and another class. [ticket:1237]

    - Test coverage added for `relation()` objects specified on
      concrete mappers. [ticket:1237]

    - Query.from_self() as well as query.subquery() both disable 
      the rendering of eager joins inside the subquery produced.
      The "disable all eager joins" feature is available publically
      via a new query.enable_eagerloads() generative. [ticket:1276]

    - Added a rudimental series of set operations to Query that
      receive Query objects as arguments, including union(),
      union_all(), intersect(), except_(), insertsect_all(),
      except_all().  See the API documentation for
      Query.union() for examples. 

    - Fixed bug that prevented Query.join() and eagerloads from 
      attaching to a query that selected from a union or aliased union.

    - A short documentation example added for bidirectional
      relations specified on concrete mappers. [ticket:1237]

    - Mappers now instrument class attributes upon construction
      with the final InstrumentedAttribute object which remains
      persistent. The `_CompileOnAttr`/`__getattribute__()`
      methodology has been removed. The net effect is that
      Column-based mapped class attributes can now be used fully
      at the class level without invoking a mapper compilation
      operation, greatly simplifying typical usage patterns
      within declarative. [ticket:1269]

    - ColumnProperty (and front-end helpers such as ``deferred``) no
      longer ignores unknown **keyword arguments.

    - Fixed a bug with the unitofwork's "row switch" mechanism,
      i.e. the conversion of INSERT/DELETE into an UPDATE, when
      combined with joined-table inheritance and an object
      which contained no defined values for the child table where
      an UPDATE with no SET clause would be rendered.

    - Using delete-orphan on a many-to-many relation is deprecated.
      This produces misleading or erroneous results since SQLA does 
      not retrieve the full list of "parents" for m2m.  To get delete-orphan
      behavior with an m2m table, use an explcit association class
      so that the individual association row is treated as a parent.
      [ticket:1281]

    - delete-orphan cascade always requires delete cascade.  Specifying
      delete-orphan without delete now raises a deprecation warning.
      [ticket:1281]

- sql
    - Improved the methodology to handling percent signs in column
      names from [ticket:1256].  Added more tests.  MySQL and
      PostgreSQL dialects still do not issue correct CREATE TABLE
      statements for identifiers with percent signs in them.

- schema
    - Index now accepts column-oriented InstrumentedAttributes
      (i.e. column-based mapped class attributes) as column
      arguments. [ticket:1214]

    - Column with no name (as in declarative) won't raise a 
      NoneType error when it's string output is requsted
      (such as in a stack trace).

    - Fixed bug when overriding a Column with a ForeignKey
      on a reflected table, where derived columns (i.e. the 
      "virtual" columns of a select, etc.) would inadvertently
      call upon schema-level cleanup logic intended only
      for the original column. [ticket:1278]

- declarative
    - Can now specify Column objects on subclasses which have no
      table of their own (i.e. use single table inheritance).
      The columns will be appended to the base table, but only
      mapped by the subclass.

    - For both joined and single inheriting subclasses, the subclass
      will only map those columns which are already mapped on the 
      superclass and those explicit on the subclass.  Other 
      columns that are present on the `Table` will be excluded
      from the mapping by default, which can be disabled
      by passing a blank `exclude_properties` collection to the
      `__mapper_args__`.  This is so that single-inheriting
      classes which define their own columns are the only classes
      to map those columns.   The effect is actually a more organized
      mapping than you'd normally get with explicit `mapper()`
      calls unless you set up the `exclude_properties` arguments
      explicitly.

    - It's an error to add new Column objects to a declarative class
      that specified an existing table using __table__.

- mysql
    - Added the missing keywords from MySQL 4.1 so they get escaped
      properly.

- mssql
    - Corrected handling of large decimal values with more robust
      tests. Removed string manipulation on floats. [ticket:1280]

    - Modified the do_begin handling in mssql to use the Cursor not
      the Connection so it is DBAPI compatible.

    - Corrected SAVEPOINT support on adodbapi by changing the
      handling of savepoint_release, which is unsupported on mssql.

0.5.0
========

- new features
- general
    - Documentation has been converted to Sphinx.  In particular,
      the generated API documentation has been constructed into a
      full blown "API Reference" section which organizes editorial
      documentation combined with generated docstrings.  Cross
      linking between sections and API docs are vastly improved, a
      javascript-powered search feature is provided, and a full
      index of all classes, functions and members is provided.

    - setup.py now imports setuptools only optionally.  If not
      present, distutils is used.  The new "pip" installer is
      recommended over easy_install as it installs in a more
      simplified way.

    - added an extremely basic illustration of a PostGIS integration
      to the examples folder.

- orm
    - Query.with_polymorphic() now accepts a third argument
      "discriminator" which will replace the value of
      mapper.polymorphic_on for that query.  Mappers themselves no
      longer require polymorphic_on to be set, even if the mapper
      has a polymorphic_identity.  When not set, the mapper will
      load non-polymorphically by default. Together, these two
      features allow a non-polymorphic concrete inheritance setup to
      use polymorphic loading on a per-query basis, since concrete
      setups are prone to many issues when used polymorphically in
      all cases.

    - dynamic_loader accepts a query_class= to customize the Query
      classes used for both the dynamic collection and the queries
      built from it.

    - query.order_by() accepts None which will remove any pending
      order_by state from the query, as well as cancel out any
      mapper/relation configured ordering. This is primarily useful
      for overriding the ordering specified on a dynamic_loader().
      [ticket:1079]

- sql
    - RowProxy objects can be used in place of dictionary arguments
      sent to connection.execute() and friends.  [ticket:935]

- dialect
    - Added a new description_encoding attribute on the dialect that
      is used for encoding the column name when processing the
      metadata. This usually defaults to utf-8.

- mssql
    - Added in a new MSGenericBinary type. This maps to the Binary
      type so it can implement the specialized behavior of treating
      length specified types as fixed-width Binary types and
      non-length types as an unbound variable length Binary type.

    - Added in new types: MSVarBinary and MSImage. [ticket:1249]

    - Added in the MSReal, MSNText, MSSmallDateTime, MSTime,
      MSDateTimeOffset, and MSDateTime2 types

- sqlite
    - Table reflection now stores the actual DefaultClause value for
      the column. [ticket:1266]

- bugfixes, behavioral changes
- general
- orm
    - Exceptions raised during compile_mappers() are now preserved
      to provide "sticky behavior" - if a hasattr() call on a
      pre-compiled mapped attribute triggers a failing compile and
      suppresses the exception, subsequent compilation is blocked
      and the exception will be reiterated on the next compile()
      call.  This issue occurs frequently when using declarative.

    - property.of_type() is now recognized on a single-table
      inheriting target, when used in the context of
      prop.of_type(..).any()/has(), as well as
      query.join(prop.of_type(...)).

    - query.join() raises an error when the target of the join
      doesn't match the property-based attribute - while it's
      unlikely anyone is doing this, the SQLAlchemy author was
      guilty of this particular loosey-goosey behavior.

    - Fixed bug when using weak_instance_map=False where modified
      events would not be intercepted for a flush(). [ticket:1272]

    - Fixed some deep "column correspondence" issues which could
      impact a Query made against a selectable containing multiple
      versions of the same table, as well as unions and similar
      which contained the same table columns in different column
      positions at different levels.  [ticket:1268]

    - Custom comparator classes used in conjunction with
      column_property(), relation() etc. can define new comparison
      methods on the Comparator, which will become available via
      __getattr__() on the InstrumentedAttribute.  In the case of
      synonym() or comparable_property(), attributes are resolved
      first on the user-defined descriptor, then on the user-defined
      comparator.

    - Added ScopedSession.is_active accessor. [ticket:976]

    - Can pass mapped attributes and column objects as keys to
      query.update({}).  [ticket:1262]

    - Mapped attributes passed to the values() of an expression
      level insert() or update() will use the keys of the mapped
      columns, not that of the mapped attribute.

    - Corrected problem with Query.delete() and Query.update() not
      working properly with bind parameters. [ticket:1242]

    - Query.select_from(), from_statement() ensure that the given
      argument is a FromClause, or Text/Select/Union, respectively.

    - Query() can be passed a "composite" attribute as a column
      expression and it will be expanded.  Somewhat related to
      [ticket:1253].

    - Query() is a little more robust when passed various column
      expressions such as strings, clauselists, text() constructs
      (which may mean it just raises an error more nicely).

    - first() works as expected with Query.from_statement().

    - Fixed bug introduced in 0.5rc4 involving eager loading not
      functioning for properties which were added to a mapper
      post-compile using add_property() or equivalent.

    - Fixed bug where many-to-many relation() with viewonly=True
      would not correctly reference the link between
      secondary->remote.

    - Duplicate items in a list-based collection will be maintained
      when issuing INSERTs to a "secondary" table in a many-to-many
      relation.  Assuming the m2m table has a unique or primary key
      constraint on it, this will raise the expected constraint
      violation instead of silently dropping the duplicate
      entries. Note that the old behavior remains for a one-to-many
      relation since collection entries in that case don't result in
      INSERT statements and SQLA doesn't manually police
      collections. [ticket:1232]

    - Query.add_column() can accept FromClause objects in the same
      manner as session.query() can.

    - Comparison of many-to-one relation to NULL is properly
      converted to IS NOT NULL based on not_().

    - Extra checks added to ensure explicit
      primaryjoin/secondaryjoin are ClauseElement instances, to
      prevent more confusing errors later on. [ticket:1087]

    - Improved mapper() check for non-class classes.  [ticket:1236]

    - comparator_factory argument is now documented and supported by
      all MapperProperty types, including column_property(),
      relation(), backref(), and synonym() [ticket:5051].

    - Changed the name of PropertyLoader to RelationProperty, to be
      consistent with all the other names.  PropertyLoader is still
      present as a synonym.

    - fixed "double iter()" call causing bus errors in shard API,
      removed errant result.close() left over from the 0.4
      version. [ticket:1099] [ticket:1228]

    - made Session.merge cascades not trigger autoflush.  Fixes
      merged instances getting prematurely inserted with missing
      values.

    - Two fixes to help prevent out-of-band columns from being
      rendered in polymorphic_union inheritance scenarios (which
      then causes extra tables to be rendered in the FROM clause
      causing cartesian products):

        - improvements to "column adaption" for a->b->c inheritance
          situations to better locate columns that are related to
          one another via multiple levels of indirection, rather
          than rendering the non-adapted column.

        - the "polymorphic discriminator" column is only rendered
          for the actual mapper being queried against. The column
          won't be "pulled in" from a subclass or superclass mapper
          since it's not needed.

    - Fixed shard_id argument on ShardedSession.execute().
      [ticket:1072]

- sql
    - Columns can again contain percent signs within their
      names. [ticket:1256]

    - sqlalchemy.sql.expression.Function is now a public class.  It
      can be subclassed to provide user-defined SQL functions in an
      imperative style, including with pre-established behaviors.
      The postgis.py example illustrates one usage of this.

    - PickleType now favors == comparison by default, if the
      incoming object (such as a dict) implements __eq__().  If the
      object does not implement __eq__() and mutable=True, a
      deprecation warning is raised.

    - Fixed the import weirdness in sqlalchemy.sql to not export
      __names__ [ticket:1215].

    - Using the same ForeignKey object repeatedly raises an error
      instead of silently failing later. [ticket:1238]

    - Added NotImplementedError for params() method on
      Insert/Update/Delete constructs.  These items currently don't
      support this functionality, which also would be a little
      misleading compared to values().

    - Reflected foreign keys will properly locate their referenced
      column, even if the column was given a "key" attribute
      different from the reflected name.  This is achieved via a new
      flag on ForeignKey/ForeignKeyConstraint called "link_to_name",
      if True means the given name is the referred-to column's name,
      not its assigned key.  [ticket:650]

    - select() can accept a ClauseList as a column in the same way
      as a Table or other selectable and the interior expressions
      will be used as column elements. [ticket:1253]

    - the "passive" flag on session.is_modified() is correctly
      propagated to the attribute manager.

    - union() and union_all() will not whack any order_by() that has
      been applied to the select()s inside.  If you union() a
      select() with order_by() (presumably to support LIMIT/OFFSET),
      you should also call self_group() on it to apply parenthesis.

- engine/pool

    - Connection.invalidate() checks for closed status to avoid
      attribute errors. [ticket:1246]

    - NullPool supports reconnect on failure behavior.
      [ticket:1094]

    - Added a mutex for the initial pool creation when using
      pool.manage(dbapi).  This prevents a minor case of "dogpile"
      behavior which would otherwise occur upon a heavy load
      startup.  [ticket:799]

    - _execute_clauseelement() goes back to being a private method.
      Subclassing Connection is not needed now that ConnectionProxy
      is available.

- documentation
    - Tickets [ticket:1200] [ticket:1149].

    - Added note about create_session() defaults.

    - Added section about metadata.reflect().

    - Updated `TypeDecorator` section.

    - Rewrote the "threadlocal" strategy section of the docs due to
      recent confusion over this feature.

    - Removed badly out of date 'polymorphic_fetch' and
      'select_table' docs from inheritance, reworked the second half
      of "joined table inheritance".

    - Documented `comparator_factory` kwarg, added new doc section
      "Custom Comparators".

- mssql
    - Refactored the Date/Time types. The ``smalldatetime`` data
      type no longer truncates to a date only, and will now be
      mapped to the MSSmallDateTime type. [ticket:1254]

    - Corrected an issue with Numerics to accept an int.

    - Mapped ``char_length`` to the ``LEN()`` function.

    - If an ``INSERT`` includes a subselect the ``INSERT`` is
      converted from an ``INSERT INTO VALUES`` construct to a
      ``INSERT INTO SELECT`` construct.

    - If the column is part of a ``primary_key`` it will be ``NOT
      NULL`` since MSSQL doesn't allow ``NULL`` in primary_key
      columns.

    - ``MSBinary`` now returns a ``BINARY`` instead of an
      ``IMAGE``. This is a backwards incompatible change in that
      ``BINARY`` is a fixed length data type whereas ``IMAGE`` is a
      variable length data type. [ticket:1249]

    - ``get_default_schema_name`` is now reflected from the database
      based on the user's default schema. This only works with MSSQL
      2005 and later. [ticket:1258]

    - Added collation support through the use of a new collation
      argument. This is supported on the following types: char,
      nchar, varchar, nvarchar, text, ntext. [ticket:1248]

    - Changes to the connection string parameters favor DSN as the
      default specification for pyodbc. See the mssql.py docstring
      for detailed usage instructions.

    - Added experimental support of savepoints. It currently does
      not work fully with sessions.

    - Support for three levels of column nullability: NULL, NOT
      NULL, and the database's configured default.  The default
      Column configuration (nullable=True) will now generate NULL in
      the DDL. Previously no specification was emitted and the
      database default would take effect (usually NULL, but not
      always).  To explicitly request the database default,
      configure columns with nullable=None and no specification will
      be emitted in DDL. This is backwards incompatible
      behavior. [ticket:1243]

- postgres
    - "%" signs in text() constructs are automatically escaped to
      "%%".  Because of the backwards incompatible nature of this
      change, a warning is emitted if '%%' is detected in the
      string.  [ticket:1267]

    - Calling alias.execute() in conjunction with
      server_side_cursors won't raise AttributeError.

    - Added Index reflection support to PostgreSQL, using a great
      patch we long neglected, submitted by Ken
      Kuhlman. [ticket:714]

- oracle
    - Adjusted the format of create_xid() to repair two-phase
      commit.  We now have field reports of Oracle two-phase commit
      working properly with this change.

    - Added OracleNVarchar type, produces NVARCHAR2, and also
      subclasses Unicode so that convert_unicode=True by default.
      NVARCHAR2 reflects into this type automatically so these
      columns pass unicode on a reflected table with no explicit
      convert_unicode=True flags.  [ticket:1233]

    - Fixed bug which was preventing out params of certain types
      from being received; thanks a ton to huddlej at wwu.edu !
      [ticket:1265]

- mysql
    - "%" signs in text() constructs are automatically escaped to
      "%%".  Because of the backwards incompatible nature of this
      change, a warning is emitted if '%%' is detected in the
      string.

    - Fixed bug in exception raise when FK columns not present
      during reflection. [ticket:1241]

    - Fixed bug involving reflection of a remote-schema table with a
      foreign key ref to another table in that schema.

- associationproxy
    - The association proxy properties are make themselves available
      at the class level, e.g. MyClass.aproxy.  Previously this
      evaluated to None.

- declarative
    - The full list of arguments accepted as string by backref()
      includes 'primaryjoin', 'secondaryjoin', 'secondary',
      'foreign_keys', 'remote_side', 'order_by'.

0.5.0rc4
========
- features
- orm
    - Query.count() has been enhanced to do the "right thing" in a
      wider variety of cases. It can now count multiple-entity
      queries, as well as column-based queries. Note that this means
      if you say query(A, B).count() without any joining criterion,
      it's going to count the cartesian product of A*B. Any query
      which is against column-based entities will automatically
      issue "SELECT count(1) FROM (SELECT...)" so that the real
      rowcount is returned, meaning a query such as
      query(func.count(A.name)).count() will return a value of one,
      since that query would return one row.

    - Lots of performance tuning.  A rough guesstimate over various
      ORM operations places it 10% faster over 0.5.0rc3, 25-30% over
      0.4.8.

- bugfixes and behavioral changes
- general:
    - global "propigate"->"propagate" change.

- orm
    - Adjustments to the enhanced garbage collection on
      InstanceState to better guard against errors due to lost
      state.

    - Query.get() returns a more informative error message when
      executed against multiple entities.  [ticket:1220]

    - Restored NotImplementedError on Cls.relation.in_()
      [ticket:1140] [ticket:1221]

    - Fixed PendingDeprecationWarning involving order_by parameter
      on relation(). [ticket:1226]

- sql
    - Removed the 'properties' attribute of the Connection object,
      Connection.info should be used.

    - Restored "active rowcount" fetch before ResultProxy autocloses
      the cursor.  This was removed in 0.5rc3.

    - Rearranged the `load_dialect_impl()` method in `TypeDecorator`
      such that it will take effect even if the user-defined
      `TypeDecorator` uses another `TypeDecorator` as its impl.

- access
    - Added support for Currency type.

    - Functions were not return their result. [ticket:1017]

    - Corrected problem with joins. Access only support LEFT OUTER
      or INNER not just JOIN by itself.  [ticket:1017]

- mssql
    - Lots of cleanup and fixes to correct problems with limit and
      offset.

    - Correct situation where subqueries as part of a binary
      expression need to be translated to use the IN and NOT IN
      syntax.

    - Fixed E Notation issue that prevented the ability to insert
      decimal values less than 1E-6. [ticket:1216]

    - Corrected problems with reflection when dealing with schemas,
      particularly when those schemas are the default
      schema. [ticket:1217]

    - Corrected problem with casting a zero length item to a
      varchar. It now correctly adjusts the CAST.

- ext
    - Can now use a custom "inherit_condition" in __mapper_args__
      when using declarative.

    - fixed string-based "remote_side", "order_by" and others not
      propagating correctly when used in backref().

0.5.0rc3
========
- features
- orm
    - Added two new hooks to SessionExtension: after_bulk_delete()
      and after_bulk_update().  after_bulk_delete() is called after
      a bulk delete() operation on a query. after_bulk_update() is
      called after a bulk update() operation on a query.

- sql
    - SQL compiler optimizations and complexity reduction. The call
      count for compiling a typical select() construct is 20% less
      versus 0.5.0rc2.

    - Dialects can now generate label names of adjustable
      length. Pass in the argument "label_length=<value>" to
      create_engine() to adjust how many characters max will be
      present in dynamically generated column labels, i.e.
      "somecolumn AS somelabel". Any value less than 6 will result
      in a label of minimal size, consisting of an underscore and a
      numeric counter. The compiler uses the value of
      dialect.max_identifier_length as a default.  [ticket:1211]

- ext
    - Added a new extension sqlalchemy.ext.serializer.  Provides
      Serializer/Deserializer "classes" which mirror
      Pickle/Unpickle, as well as dumps() and loads(). This
      serializer implements an "external object" pickler which keeps
      key context-sensitive objects, including engines, sessions,
      metadata, Tables/Columns, and mappers, outside of the pickle
      stream, and can later restore the pickle using any
      engine/metadata/session provider. This is used not for
      pickling regular object instances, which are pickleable
      without any special logic, but for pickling expression objects
      and full Query objects, such that all mapper/engine/session
      dependencies can be restored at unpickle time.

- oracle
    - Wrote a docstring for Oracle dialect. Apparently that Ohloh
      "few source code comments" label is starting to sting :).

    - Removed FIRST_ROWS() optimize flag when using LIMIT/OFFSET,
      can be reenabled with optimize_limits=True create_engine()
      flag.  [ticket:536]

- bugfixes and behavioral changes
- orm
    - "not equals" comparisons of simple many-to-one relation to an
      instance will not drop into an EXISTS clause and will compare
      foreign key columns instead.

    - Removed not-really-working use cases of comparing a collection
      to an iterable. Use contains() to test for collection
      membership.

    - Improved the behavior of aliased() objects such that they more
      accurately adapt the expressions generated, which helps
      particularly with self-referential comparisons. [ticket:1171]

    - Fixed bug involving primaryjoin/secondaryjoin conditions
      constructed from class-bound attributes (as often occurs when
      using declarative), which later would be inappropriately
      aliased by Query, particularly with the various EXISTS based
      comparators.

    - Fixed bug when using multiple query.join() with an
      aliased-bound descriptor which would lose the left alias.

    - Improved weakref identity map memory management to no longer
      require mutexing, resurrects garbage collected instance on a
      lazy basis for an InstanceState with pending changes.

    - InstanceState object now removes circular references to itself
      upon disposal to keep it outside of cyclic garbage collection.

    - relation() won't hide unrelated ForeignKey errors inside of
      the "please specify primaryjoin" message when determining join
      condition.

    - Fixed bug in Query involving order_by() in conjunction with
      multiple aliases of the same class (will add tests in
      [ticket:1218])

    - When using Query.join() with an explicit clause for the ON
      clause, the clause will be aliased in terms of the left side
      of the join, allowing scenarios like query(Source).
      from_self().join((Dest, Source.id==Dest.source_id)) to work
      properly.

    - polymorphic_union() function respects the "key" of each Column
      if they differ from the column's name.

    - Repaired support for "passive-deletes" on a many-to-one
      relation() with "delete" cascade.  [ticket:1183]

    - Fixed bug in composite types which prevented a primary-key
      composite type from being mutated [ticket:1213].

    - Added more granularity to internal attribute access, such that
      cascade and flush operations will not initialize unloaded
      attributes and collections, leaving them intact for a
      lazy-load later on. Backref events still initialize attrbutes
      and collections for pending instances.  [ticket:1202]

- sql
    - Simplified the check for ResultProxy "autoclose without
      results" to be based solely on presence of
      cursor.description. All the regexp-based guessing about
      statements returning rows has been removed [ticket:1212].

    - Direct execution of a union() construct will properly set up
      result-row processing.  [ticket:1194]

    - The internal notion of an "OID" or "ROWID" column has been
      removed. It's basically not used by any dialect, and the
      possibility of its usage with psycopg2's cursor.lastrowid is
      basically gone now that INSERT..RETURNING is available.

    - Removed "default_order_by()" method on all FromClause objects.

    - Repaired the table.tometadata() method so that a passed-in
      schema argument is propagated to ForeignKey constructs.

    - Slightly changed behavior of IN operator for comparing to
      empty collections. Now results in inequality comparison
      against self. More portable, but breaks with stored procedures
      that aren't pure functions.

- oracle
    - Setting the auto_convert_lobs to False on create_engine() will
      also instruct the OracleBinary type to return the cx_oracle
      LOB object unchanged.

- mysql
    - Fixed foreign key reflection in the edge case where a Table's
      explicit schema= is the same as the schema (database) the
      connection is attached to.

    - No longer expects include_columns in table reflection to be
      lower case.

- ext
    - Fixed bug preventing declarative-bound "column" objects from
      being used in column_mapped_collection(). [ticket:1174]

- misc
    - util.flatten_iterator() func doesn't interpret strings with
      __iter__() methods as iterators, such as in pypy
      [ticket:1077].

0.5.0rc2
========
- orm
    - Fixed bug involving read/write relation()s that contain
      literal or other non-column expressions within their
      primaryjoin condition equated to a foreign key column.

    - "non-batch" mode in mapper(), a feature which allows mapper
      extension methods to be called as each instance is
      updated/inserted, now honors the insert order of the objects
      given.

    - Fixed RLock-related bug in mapper which could deadlock upon
      reentrant mapper compile() calls, something that occurs when
      using declarative constructs inside of ForeignKey objects.

    - ScopedSession.query_property now accepts a query_cls factory,
      overriding the session's configured query_cls.

    - Fixed shared state bug interfering with ScopedSession.mapper's
      ability to apply default __init__ implementations on object
      subclasses.

    - Fixed up slices on Query (i.e. query[x:y]) to work properly
      for zero length slices, slices with None on either end.
      [ticket:1177]

    - Added an example illustrating Celko's "nested sets" as a 
      SQLA mapping.

    - contains_eager() with an alias argument works even when 
      the alias is embedded in a SELECT, as when sent to the
      Query via query.select_from().

    - contains_eager() usage is now compatible with a Query that
      also contains a regular eager load and limit/offset, in that
      the columns are added to the Query-generated subquery.
      [ticket:1180]

    - session.execute() will execute a Sequence object passed to
      it (regression from 0.4).

    - Removed the "raiseerror" keyword argument from object_mapper()
      and class_mapper().  These functions raise in all cases
      if the given class/instance is not mapped.

    - Fixed session.transaction.commit() on a autocommit=False
      session not starting a new transaction.

    - Some adjustments to Session.identity_map's weak referencing
      behavior to reduce asynchronous GC side effects.

    - Adjustment to Session's post-flush accounting of newly
      "clean" objects to better protect against operating on
      objects as they're asynchronously gc'ed. [ticket:1182]

- sql
    - column.in_(someselect) can now be used as a columns-clause
      expression without the subquery bleeding into the FROM clause
      [ticket:1074]

- sqlite
    - Overhauled SQLite date/time bind/result processing to use
      regular expressions and format strings, rather than
      strptime/strftime, to generically support pre-1900 dates,
      dates with microseconds.  [ticket:968]

    - String's (and Unicode's, UnicodeText's, etc.) convert_unicode 
      logic disabled in the sqlite dialect, to adjust for pysqlite 
      2.5.0's new requirement that only Python unicode objects are 
      accepted;
      http://itsystementwicklung.de/pipermail/list-pysqlite/2008-March/000018.html

- mysql
    - Temporary tables are now reflectable.

- oracle
    - Oracle will detect string-based statements which contain 
      comments at the front before a SELECT as SELECT statements.
      [ticket:1187]

0.5.0rc1
========

- orm
    - Query now has delete() and update(values) methods. This allows
      to perform bulk deletes/updates with the Query object.

    - The RowTuple object returned by Query(*cols) now features
      keynames which prefer mapped attribute names over column keys,
      column keys over column names, i.e.  Query(Class.foo,
      Class.bar) will have names "foo" and "bar" even if those are
      not the names of the underlying Column objects.  Direct Column
      objects such as Query(table.c.col) will return the "key"
      attribute of the Column.

    - Added scalar() and value() methods to Query, each return a
      single scalar value.  scalar() takes no arguments and is
      roughly equivalent to first()[0], value()
      takes a single column expression and is roughly equivalent to
      values(expr).next()[0].

    - Improved the determination of the FROM clause when placing SQL
      expressions in the query() list of entities.  In particular
      scalar subqueries should not "leak" their inner FROM objects
      out into the enclosing query.

    - Joins along a relation() from a mapped class to a mapped
      subclass, where the mapped subclass is configured with single
      table inheritance, will include an IN clause which limits the
      subtypes of the joined class to those requested, within the ON
      clause of the join.  This takes effect for eager load joins as
      well as query.join().  Note that in some scenarios the IN
      clause will appear in the WHERE clause of the query as well
      since this discrimination has multiple trigger points.

    - AttributeExtension has been refined such that the event
      is fired before the mutation actually occurs.  Additionally,
      the append() and set() methods must now return the given value,
      which is used as the value to be used in the mutation operation.
      This allows creation of validating AttributeListeners which
      raise before the action actually occurs, and which can change
      the given value into something else before its used.

    - column_property(), composite_property(), and relation() now 
      accept a single or list of AttributeExtensions using the 
      "extension" keyword argument.

    - query.order_by().get() silently drops the "ORDER BY" from 
      the query issued by GET but does not raise an exception.

    - Added a Validator AttributeExtension, as well as a 
      @validates decorator which is used in a similar fashion
      as @reconstructor, and marks a method as validating
      one or more mapped attributes.

    - class.someprop.in_() raises NotImplementedError pending the
      implementation of "in_" for relation [ticket:1140]

    - Fixed primary key update for many-to-many collections where
      the collection had not been loaded yet [ticket:1127]

    - Fixed bug whereby deferred() columns with a group in conjunction
      with an otherwise unrelated synonym() would produce 
      an AttributeError during deferred load.

    - The before_flush() hook on SessionExtension takes place before
      the list of new/dirty/deleted is calculated for the final
      time, allowing routines within before_flush() to further
      change the state of the Session before the flush proceeds.
      [ticket:1128]

    - The "extension" argument to Session and others can now
      optionally be a list, supporting events sent to multiple
      SessionExtension instances.  Session places SessionExtensions
      in Session.extensions.

    - Reentrant calls to flush() raise an error.  This also serves
      as a rudimentary, but not foolproof, check against concurrent
      calls to Session.flush().

    - Improved the behavior of query.join() when joining to
      joined-table inheritance subclasses, using explicit join
      criteria (i.e. not on a relation).

    - @orm.attributes.reconstitute and
      MapperExtension.reconstitute have been renamed to
      @orm.reconstructor and MapperExtension.reconstruct_instance

    - Fixed @reconstructor hook for subclasses which inherit from a
      base class.  [ticket:1129]

    - The composite() property type now supports a
      __set_composite_values__() method on the composite class which
      is required if the class represents state using attribute
      names other than the column's keynames; default-generated
      values now get populated properly upon flush.  Also,
      composites with attributes set to None compare correctly.
      [ticket:1132]

    - The 3-tuple of iterables returned by attributes.get_history()
      may now be a mix of lists and tuples.  (Previously members
      were always lists.)

    - Fixed bug whereby changing a primary key attribute on an 
      entity where the attribute's previous value had been expired 
      would produce an error upon flush(). [ticket:1151]

    - Fixed custom instrumentation bug whereby get_instance_dict()
      was not called for newly constructed instances not loaded
      by the ORM.

    - Session.delete() adds the given object to the session if 
      not already present.  This was a regression bug from 0.4.
      [ticket:1150]

    - The `echo_uow` flag on `Session` is deprecated, and unit-of-work
      logging is now application-level only, not per-session level.

    - Removed conflicting `contains()` operator from 
      `InstrumentedAttribute` which didn't accept `escape` kwaarg
      [ticket:1153].

- declarative
    - Fixed bug whereby mapper couldn't initialize if a composite
      primary key referenced another table that was not defined
      yet. [ticket:1161]

    - Fixed exception throw which would occur when string-based
      primaryjoin condition was used in conjunction with backref.

- schema
    - Added "sorted_tables" accessor to MetaData, which returns
      Table objects sorted in order of dependency as a list.
      This deprecates the MetaData.table_iterator() method.
      The "reverse=False" keyword argument has also been
      removed from util.sort_tables(); use the Python
      'reversed' function to reverse the results.
      [ticket:1033]

    - The 'length' argument to all Numeric types has been renamed
      to 'scale'.  'length' is deprecated and is still accepted
      with a warning.

    - Dropped 0.3-compatibility for user defined types
      (convert_result_value, convert_bind_param).

- sql
    - Temporarily rolled back the "ORDER BY" enhancement from
      [ticket:1068].  This feature is on hold pending further
      development.

    - The exists() construct won't "export" its contained list 
      of elements as FROM clauses, allowing them to be used more
      effectively in the columns clause of a SELECT.

    - and_() and or_() now generate a ColumnElement, allowing
      boolean expressions as result columns, i.e.
      select([and_(1, 0)]).  [ticket:798]

    - Bind params now subclass ColumnElement which allows them to be
      selectable by orm.query (they already had most ColumnElement
      semantics).

    - Added select_from() method to exists() construct, which becomes
      more and more compatible with a regular select().

    - Added func.min(), func.max(), func.sum() as "generic functions",
      which basically allows for their return type to be determined
      automatically.  Helps with dates on SQLite, decimal types, 
      others. [ticket:1160]

    - added decimal.Decimal as an "auto-detect" type; bind parameters
      and generic functions will set their type to Numeric when a 
      Decimal is used.

- mysql
    - The 'length' argument to MSInteger, MSBigInteger, MSTinyInteger,
      MSSmallInteger and MSYear has been renamed to 'display_width'.

    - Added MSMediumInteger type [ticket:1146].

    - the function func.utc_timestamp() compiles to UTC_TIMESTAMP, without
      the parenthesis, which seem to get in the way when using in 
      conjunction with executemany().

- oracle
    - limit/offset no longer uses ROW NUMBER OVER to limit rows,
      and instead uses subqueries in conjunction with a special
      Oracle optimization comment.  Allows LIMIT/OFFSET to work
      in conjunction with DISTINCT. [ticket:536]
    - has_sequence() now takes the current "schema" argument into
      account [ticket:1155]
    - added BFILE to reflected type names [ticket:1121]

0.5beta3
========

0.5beta3 includes all bugfixes listed under release "0.4.7".

- orm
    - The "entity_name" feature of SQLAlchemy mappers has been
      removed.  For rationale, see http://tinyurl.com/6nm2ne

    - the "autoexpire" flag on Session, sessionmaker(), and
      scoped_session() has been renamed to "expire_on_commit".  It
      does not affect the expiration behavior of rollback().

    - fixed endless loop bug which could occur within a mapper's
      deferred load of inherited attributes.

    - a legacy-support flag "_enable_transaction_accounting" flag
      added to Session which when False, disables all
      transaction-level object accounting, including expire on
      rollback, expire on commit, new/deleted list maintenance, and
      autoflush on begin.

    - The 'cascade' parameter to relation() accepts None as a value,
      which is equivalent to no cascades.

    - A critical fix to dynamic relations allows the "modified"
      history to be properly cleared after a flush().

    - user-defined @properties on a class are detected and left in
      place during mapper initialization.  This means that a
      table-bound column of the same name will not be mapped at all
      if a @property is in the way (and the column is not remapped
      to a different name), nor will an instrumented attribute from
      an inherited class be applied.  The same rules apply for names
      excluded using the include_properties/exclude_properties
      collections.

    - Added a new SessionExtension hook called after_attach().  This
      is called at the point of attachment for objects via add(),
      add_all(), delete(), and merge().

    - A mapper which inherits from another, when inheriting the
      columns of its inherited mapper, will use any reassigned
      property names specified in that inheriting mapper.
      Previously, if "Base" had reassigned "base_id" to the name
      "id", "SubBase(Base)" would still get an attribute called
      "base_id".  This could be worked around by explicitly stating
      the column in each submapper as well but this is fairly
      unworkable and also impossible when using declarative
      [ticket:1111].

    - Fixed a series of potential race conditions in Session whereby
      asynchronous GC could remove unmodified, no longer referenced
      items from the session as they were present in a list of items
      to be processed, typically during session.expunge_all() and
      dependent methods.

    - Some improvements to the _CompileOnAttr mechanism which should
      reduce the probability of "Attribute x was not replaced during
      compile" warnings. (this generally applies to SQLA hackers,
      like Elixir devs).

    - Fixed bug whereby the "unsaved, pending instance" FlushError
      raised for a pending orphan would not take superclass mappers
      into account when generating the list of relations responsible
      for the error.

- sql
    - func.count() with no arguments renders as COUNT(*), equivalent
      to func.count(text('*')).

    - simple label names in ORDER BY expressions render as
      themselves, and not as a re-statement of their corresponding
      expression.  This feature is currently enabled only for
      SQLite, MySQL, and PostgreSQL.  It can be enabled on other
      dialects as each is shown to support this
      behavior. [ticket:1068]

- ext
    - Class-bound attributes sent as arguments to relation()'s
      remote_side and foreign_keys parameters are now accepted,
      allowing them to be used with declarative.  Additionally fixed
      bugs involving order_by being specified as a class-bound
      attribute in conjunction with eager loading.

    - declarative initialization of Columns adjusted so that
      non-renamed columns initialize in the same way as a non
      declarative mapper.  This allows an inheriting mapper to set
      up its same-named "id" columns in particular such that the
      parent "id" column is favored over the child column, reducing
      database round trips when this value is requested.

- mysql
    - Quoting of MSEnum values for use in CREATE TABLE is now
      optional & will be quoted on demand as required.  (Quoting was
      always optional for use with existing tables.) [ticket:1110]

0.5beta2
========

0.5beta2 includes some of the bugfixes listed under release "0.4.7".

- orm
    - In addition to expired attributes, deferred attributes also
      load if their data is present in the result set.  [ticket:870]

    - session.refresh() raises an informative error message if the
      list of attributes does not include any column-based
      attributes.

    - query() raises an informative error message if no columns or
      mappers are specified.

    - lazy loaders now trigger autoflush before proceeding.  This
      allows expire() of a collection or scalar relation to function
      properly in the context of autoflush.

    - column_property() attributes which represent SQL expressions
      or columns that are not present in the mapped tables (such as
      those from views) are automatically expired after an INSERT or
      UPDATE, assuming they have not been locally modified, so that
      they are refreshed with the most recent data upon access.
      [ticket:887]

    - Fixed explicit, self-referential joins between two
      joined-table inheritance mappers when using query.join(cls,
      aliased=True).  [ticket:1082]

    - Fixed query.join() when used in conjunction with a
      columns-only clause and an SQL-expression ON clause in the
      join.

    - The "allow_column_override" flag from mapper() has been
      removed.  This flag is virtually always misunderstood.  Its
      specific functionality is available via the
      include_properties/exclude_properties mapper arguments.

    - Repaired `__str__()` method on Query. [ticket:1066]

    - Session.bind gets used as a default even when table/mapper
      specific binds are defined.

- schema
    - Added prefixes option to `Table` that accepts a list of
      strings to insert after CREATE in the CREATE TABLE statement.
      [ticket:1075]

    - Unicode, UnicodeText types now set "assert_unicode" and
      "convert_unicode" by default, but accept overriding
      **kwargs for these values.

- sql
    - Added new match() operator that performs a full-text search.
      Supported on PostgreSQL, SQLite, MySQL, MS-SQL, and Oracle
      backends.

- sqlite
    - Modified SQLite's representation of "microseconds" to match
      the output of str(somedatetime), i.e. in that the microseconds
      are represented as fractional seconds in string format.  This
      makes SQLA's SQLite date type compatible with datetimes that
      were saved directly using Pysqlite (which just calls str()).
      Note that this is incompatible with the existing microseconds
      values in a SQLA 0.4 generated SQLite database file.

      To get the old behavior globally:

           from sqlalchemy.databases.sqlite import DateTimeMixin
           DateTimeMixin.__legacy_microseconds__ = True

      To get the behavior on individual DateTime types:

            t = sqlite.SLDateTime()
            t.__legacy_microseconds__ = True

      Then use "t" as the type on the Column.

      [ticket:1090]

    - SQLite Date, DateTime, and Time types only accept Python
      datetime objects now, not strings.  If you'd like to format
      dates as strings yourself with SQLite, use a String type.  If
      you'd like them to return datetime objects anyway despite
      their accepting strings as input, make a TypeDecorator around
      String - SQLA doesn't encourage this pattern.

- extensions
    - Declarative supports a __table_args__ class variable, which is
      either a dictionary, or tuple of the form (arg1, arg2, ...,
      {kwarg1:value, ...}) which contains positional + kw arguments
      to be passed to the Table constructor. [ticket:1096]

0.5beta1
========

An ongoing document describing the changes from 0.4 to 0.5 is at:
http://www.sqlalchemy.org/trac/wiki/05Migration

For this section, the list below primarily lists changes within the
attribute instrumentation API and is only a small fraction of the
full changes between 0.4/0.5.

-orm
    - The "__init__" trigger/decorator added by mapper now attempts
      to exactly mirror the argument signature of the original
      __init__.  The pass-through for '_sa_session' is no longer
      implicit- you must allow for this keyword argument in your
      constructor.

    - ClassState is renamed to ClassManager.

    - Classes may supply their own InstrumentationManager by
      providing a __sa_instrumentation_manager__ property.

    - Custom instrumentation may use any mechanism to associate a
      ClassManager with a class and an InstanceState with an
      instance.  Attributes on those objects are still the default
      association mechanism used by SQLAlchemy's native
      instrumentation.

    - Moved entity_name, _sa_session_id, and _instance_key from the
      instance object to the instance state.  These values are still
      available in the old way, which is now deprecated, using
      descriptors attached to the class.  A deprecation warning will
      be issued when accessed.

    - The _prepare_instrumentation alias for prepare_instrumentation
      has been removed.

    - sqlalchemy.exceptions has been renamed to sqlalchemy.exc.  The
      module may be imported under either name.

    - ORM-related exceptions are now defined in sqlalchemy.orm.exc.
      ConcurrentModificationError, FlushError, and
      UnmappedColumnError compatibility aliases are installed in
      sqlalchemy.exc during the import of sqlalchemy.orm.

    - sqlalchemy.logging has been renamed to sqlalchemy.log.

    - The transitional sqlalchemy.log.SADeprecationWarning alias for
      the warning's definition in sqlalchemy.exc has been removed.

    - exc.AssertionError has been removed and usage replaced with
      Python's built-in AssertionError.

    - The behavior of MapperExtensions attached to multiple,
      entity_name= primary mappers for a single class has been
      altered.  The first mapper() defined for a class is the only
      mapper eligible for the MapperExtension 'instrument_class',
      'init_instance' and 'init_failed' events.  This is backwards
      incompatible; previously the extensions of last mapper defined
      would receive these events.

- firebird
    - Added support for returning values from inserts (2.0+ only),
      updates and deletes (2.1+ only).

0.4.7
=====
- orm
    - The contains() operator when used with many-to-many
      will alias() the secondary (association) table so
      that multiple contains() calls will not conflict
      with each other [ticket:1058]

    - fixed bug preventing merge() from functioning in
      conjunction with a comparable_property()

    - the enable_typechecks=False setting on relation()
      now only allows subtypes with inheriting mappers.
      Totally unrelated types, or subtypes not set up with
      mapper inheritance against the target mapper are
      still not allowed.

    - Added is_active flag to Sessions to detect when
      a transaction is in progress [ticket:976].  This
      flag is always True with a "transactional"
      (in 0.5 a non-"autocommit") Session.

- sql
    - Fixed bug when calling select([literal('foo')])
      or select([bindparam('foo')]).

- schema
    - create_all(), drop_all(), create(), drop() all raise
      an error if the table name or schema name contains
      more characters than that dialect's configured
      character limit.  Some DB's can handle too-long
      table names during usage, and SQLA can handle this
      as well. But various reflection/
      checkfirst-during-create scenarios fail since we are
      looking for the name within the DB's catalog tables.
      [ticket:571]

    - The index name generated when you say "index=True"
      on a Column is truncated to the length appropriate
      for the dialect. Additionally, an Index with a too-
      long name cannot be explicitly dropped with
      Index.drop(), similar to [ticket:571].
      [ticket:820]

- postgres
    - Repaired server_side_cursors to properly detect
      text() clauses.

    - Added PGCidr type. [ticket:1092]

- mysql
    - Added 'CALL' to the list of SQL keywords which return
      result rows.

- oracle
    - Oracle get_default_schema_name() "normalizes" the name
      before returning, meaning it returns a lower-case name
      when the identifier is detected as case insensitive.

    - creating/dropping tables takes schema name into account
      when searching for the existing table, so that tables
      in other owner namespaces with the same name do not
      conflict [ticket:709]

    - Cursors now have "arraysize" set to 50 by default on
      them, the value of which is configurable using the
      "arraysize" argument to create_engine() with the
      Oracle dialect.  This to account for cx_oracle's default
      setting of "1", which has the effect of many round trips
      being sent to Oracle.  This actually works well in
      conjunction with BLOB/CLOB-bound cursors, of which
      there are any number available but only for the life of
      that row request (so BufferedColumnRow is still needed,
      but less so). [ticket:1062]

- sqlite
      - add SLFloat type, which matches the SQLite REAL
        type affinity.  Previously, only SLNumeric was provided
        which fulfills NUMERIC affinity, but that's not the
        same as REAL.

0.4.6
=====
- orm
    - Fix to the recent relation() refactoring which fixes
      exotic viewonly relations which join between local and
      remote table multiple times, with a common column shared
      between the joins.

    - Also re-established viewonly relation() configurations
      that join across multiple tables.

    - Added experimental relation() flag to help with
      primaryjoins across functions, etc.,
      _local_remote_pairs=[tuples].  This complements a complex
      primaryjoin condition allowing you to provide the
      individual column pairs which comprise the relation's
      local and remote sides.  Also improved lazy load SQL
      generation to handle placing bind params inside of
      functions and other expressions.  (partial progress
      towards [ticket:610])

    - repaired single table inheritance such that you
      can single-table inherit from a joined-table inherting
      mapper without issue [ticket:1036].

    - Fixed "concatenate tuple" bug which could occur with
      Query.order_by() if clause adaption had taken place.
      [ticket:1027]

    - Removed ancient assertion that mapped selectables require
      "alias names" - the mapper creates its own alias now if
      none is present.  Though in this case you need to use the
      class, not the mapped selectable, as the source of column
      attributes - so a warning is still issued.

    - fixes to the "exists" function involving inheritance (any(),
      has(), ~contains()); the full target join will be rendered
      into the EXISTS clause for relations that link to subclasses.

    - restored usage of append_result() extension method for primary
      query rows, when the extension is present and only a single-
      entity result is being returned.

    - Also re-established viewonly relation() configurations that
      join across multiple tables.

    - removed ancient assertion that mapped selectables require
      "alias names" - the mapper creates its own alias now if
      none is present.  Though in this case you need to use
      the class, not the mapped selectable, as the source of
      column attributes - so a warning is still issued.

    - refined mapper._save_obj() which was unnecessarily calling
      __ne__() on scalar values during flush [ticket:1015]

    - added a feature to eager loading whereby subqueries set
      as column_property() with explicit label names (which is not
      necessary, btw) will have the label anonymized when
      the instance is part of the eager join, to prevent
      conflicts with a subquery or column of the same name
      on the parent object.  [ticket:1019]

    - set-based collections |=, -=, ^= and &= are stricter about
      their operands and only operate on sets, frozensets or
      subclasses of the collection type. Previously, they would
      accept any duck-typed set.

    - added an example dynamic_dict/dynamic_dict.py, illustrating
      a simple way to place dictionary behavior on top of
      a dynamic_loader.

- declarative extension
    - Joined table inheritance mappers use a slightly relaxed
      function to create the "inherit condition" to the parent
      table, so that other foreign keys to not-yet-declared
      Table objects don't trigger an error.

    - fixed reentrant mapper compile hang when
      a declared attribute is used within ForeignKey,
      ie. ForeignKey(MyOtherClass.someattribute)

- sql
    - Added COLLATE support via the .collate(<collation>)
      expression operator and collate(<expr>, <collation>) sql
      function.

    - Fixed bug with union() when applied to non-Table connected
      select statements

    - improved behavior of text() expressions when used as
      FROM clauses, such as select().select_from(text("sometext"))
      [ticket:1014]

    - Column.copy() respects the value of "autoincrement",
      fixes usage with Migrate [ticket:1021]

- engines
    - Pool listeners can now be provided as a dictionary of
      callables or a (possibly partial) duck-type of
      PoolListener, your choice.

    - added "rollback_returned" option to Pool which will
      disable the rollback() issued when connections are
      returned.  This flag is only safe to use with a database
      which does not support transactions (i.e. MySQL/MyISAM).

- ext
    - set-based association proxies |=, -=, ^= and &= are
      stricter about their operands and only operate on sets,
      frozensets or other association proxies. Previously, they
      would accept any duck-typed set.

- mssql
    - Added "odbc_autotranslate" parameter to engine / dburi
      parameters. Any given string will be passed through to the
      ODBC connection string as:

            "AutoTranslate=%s" % odbc_autotranslate

      [ticket:1005]

    - Added "odbc_options" parameter to engine / dburi
      parameters. The given string is simply appended to the
      SQLAlchemy-generated odbc connection string.

      This should obviate the need of adding a myriad of ODBC
      options in the future.

- firebird
    - Handle the "SUBSTRING(:string FROM :start FOR :length)"
      builtin.

0.4.5
=====
- orm
    - A small change in behavior to session.merge() - existing
      objects are checked for based on primary key attributes, not
      necessarily _instance_key.  So the widely requested
      capability, that:

            x = MyObject(id=1)
            x = sess.merge(x)

      will in fact load MyObject with id #1 from the database if
      present, is now available.  merge() still copies the state
      of the given object to the persistent one, so an example
      like the above would typically have copied "None" from all
      attributes of "x" onto the persistent copy.  These can be
      reverted using session.expire(x).

    - Also fixed behavior in merge() whereby collection elements
      present on the destination but not the merged collection
      were not being removed from the destination.

    - Added a more aggressive check for "uncompiled mappers",
      helps particularly with declarative layer [ticket:995]

    - The methodology behind "primaryjoin"/"secondaryjoin" has
      been refactored.  Behavior should be slightly more
      intelligent, primarily in terms of error messages which
      have been pared down to be more readable.  In a slight
      number of scenarios it can better resolve the correct
      foreign key than before.

    - Added comparable_property(), adds query Comparator
      behavior to regular, unmanaged Python properties

    - the functionality of query.with_polymorphic() has
      been added to mapper() as a configuration option.

      It's set via several forms:
            with_polymorphic='*'
            with_polymorphic=[mappers]
            with_polymorphic=('*', selectable)
            with_polymorphic=([mappers], selectable)

      This controls the default polymorphic loading strategy
      for inherited mappers. When a selectable is not given,
      outer joins are created for all joined-table inheriting
      mappers requested. Note that the auto-create of joins
      is not compatible with concrete table inheritance.

      The existing select_table flag on mapper() is now
      deprecated and is synonymous with
      with_polymorphic('*', select_table).  Note that the
      underlying "guts" of select_table have been
      completely removed and replaced with the newer,
      more flexible approach.

      The new approach also automatically allows eager loads
      to work for subclasses, if they are present, for
      example
        sess.query(Company).options(
         eagerload_all(
          [Company.employees.of_type(Engineer), 'machines']
        ))
      to load Company objects, their employees, and the
      'machines' collection of employees who happen to be
      Engineers. A "with_polymorphic" Query option should be
      introduced soon as well which would allow per-Query
      control of with_polymorphic() on relations.

    - added two "experimental" features to Query,
      "experimental" in that their specific name/behavior
      is not carved in stone just yet:  _values() and
      _from_self().  We'd like feedback on these.

      - _values(*columns) is given a list of column
        expressions, and returns a new Query that only
        returns those columns. When evaluated, the return
        value is a list of tuples just like when using
        add_column() or add_entity(), the only difference is
        that "entity zero", i.e. the mapped class, is not
        included in the results. This means it finally makes
        sense to use group_by() and having() on Query, which
        have been sitting around uselessly until now.

        A future change to this method may include that its
        ability to join, filter and allow other options not
        related to a "resultset" are removed, so the feedback
        we're looking for is how people want to use
        _values()...i.e. at the very end, or do people prefer
        to continue generating after it's called.

      - _from_self() compiles the SELECT statement for the
        Query (minus any eager loaders), and returns a new
        Query that selects from that SELECT. So basically you
        can query from a Query without needing to extract the
        SELECT statement manually. This gives meaning to
        operations like query[3:5]._from_self().filter(some
        criterion). There's not much controversial here
        except that you can quickly create highly nested
        queries that are less efficient, and we want feedback
        on the naming choice.

    - query.order_by() and query.group_by() will accept
      multiple arguments using *args (like select()
      already does).

    - Added some convenience descriptors to Query:
      query.statement returns the full SELECT construct,
      query.whereclause returns just the WHERE part of the
      SELECT construct.

    - Fixed/covered case when using a False/0 value as a
      polymorphic discriminator.

    - Fixed bug which was preventing synonym() attributes from
      being used with inheritance

    - Fixed SQL function truncation of trailing underscores
      [ticket:996]

    - When attributes are expired on a pending instance, an
      error will not be raised when the "refresh" action is
      triggered and no result is found.

    - Session.execute can now find binds from metadata

    - Adjusted the definition of "self-referential" to be any
      two mappers with a common parent (this affects whether or
      not aliased=True is required when joining with Query).

    - Made some fixes to the "from_joinpoint" argument to
      query.join() so that if the previous join was aliased and
      this one isn't, the join still happens successfully.

    - Assorted "cascade deletes" fixes:
        - Fixed "cascade delete" operation of dynamic relations,
          which had only been implemented for foreign-key
          nulling behavior in 0.4.2 and not actual cascading
          deletes [ticket:895]

        - Delete cascade without delete-orphan cascade on a
          many-to-one will not delete orphans which were
          disconnected from the parent before session.delete()
          is called on the parent (one-to-many already had
          this).

        - Delete cascade with delete-orphan will delete orphans
          whether or not it remains attached to its also-deleted
          parent.

        - delete-orphan casacde is properly detected on relations
          that are present on superclasses when using inheritance.

    - Fixed order_by calculation in Query to properly alias
      mapper-config'ed order_by when using select_from()

    - Refactored the diffing logic that kicks in when replacing
      one collection with another into collections.bulk_replace,
      useful to anyone building multi-level collections.

    - Cascade traversal algorithm converted from recursive to
      iterative to support deep object graphs.

- sql
    - schema-qualified tables now will place the schemaname
      ahead of the tablename in all column expressions as well
      as when generating column labels.  This prevents cross-
      schema name collisions in all cases [ticket:999]

    - can now allow selects which correlate all FROM clauses
      and have no FROM themselves.  These are typically
      used in a scalar context, i.e. SELECT x, (SELECT x WHERE y)
      FROM table.  Requires explicit correlate() call.

    - 'name' is no longer a required constructor argument for
      Column().  It (and .key) may now be deferred until the
      column is added to a Table.

    - like(), ilike(), contains(), startswith(), endswith() take
      an optional keyword argument "escape=<somestring>", which
      is set as the escape character using the syntax "x LIKE y
      ESCAPE '<somestring>'" [ticket:993], [ticket:791]

    - random() is now a generic sql function and will compile to
      the database's random implementation, if any.

    - update().values() and insert().values() take keyword
      arguments.

    - Fixed an issue in select() regarding its generation of
      FROM clauses, in rare circumstances two clauses could be
      produced when one was intended to cancel out the other.
      Some ORM queries with lots of eager loads might have seen
      this symptom.

    - The case() function now also takes a dictionary as its
      whens parameter.  It also interprets the "THEN"
      expressions as values by default, meaning case([(x==y,
      "foo")]) will interpret "foo" as a bound value, not a SQL
      expression.  use text(expr) for literal SQL expressions in
      this case.  For the criterion itself, these may be literal
      strings only if the "value" keyword is present, otherwise
      SA will force explicit usage of either text() or
      literal().

- oracle
    - The "owner" keyword on Table is now deprecated, and is
      exactly synonymous with the "schema" keyword.  Tables can
      now be reflected with alternate "owner" attributes,
      explicitly stated on the Table object or not using
      "schema".

    - All of the "magic" searching for synonyms, DBLINKs etc.
      during table reflection are disabled by default unless you
      specify "oracle_resolve_synonyms=True" on the Table
      object.  Resolving synonyms necessarily leads to some
      messy guessing which we'd rather leave off by default.
      When the flag is set, tables and related tables will be
      resolved against synonyms in all cases, meaning if a
      synonym exists for a particular table, reflection will use
      it when reflecting related tables.  This is stickier
      behavior than before which is why it's off by default.

- declarative extension
    - The "synonym" function is now directly usable with
      "declarative".  Pass in the decorated property using the
      "descriptor" keyword argument, e.g.: somekey =
      synonym('_somekey', descriptor=property(g, s))

    - The "deferred" function is usable with "declarative".
      Simplest usage is to declare deferred and Column together,
      e.g.: data = deferred(Column(Text))

    - Declarative also gained @synonym_for(...) and
      @comparable_using(...), front-ends for synonym and
      comparable_property.

    - Improvements to mapper compilation when using declarative;
      already-compiled mappers will still trigger compiles of
      other uncompiled mappers when used [ticket:995]

    - Declarative will complete setup for Columns lacking names,
      allows a more DRY syntax.

        class Foo(Base):
            __tablename__ = 'foos'
            id = Column(Integer, primary_key=True)

     - inheritance in declarative can be disabled when sending
       "inherits=None" to __mapper_args__.

     - declarative_base() takes optional kwarg "mapper", which
       is any callable/class/method that produces a mapper,
       such as declarative_base(mapper=scopedsession.mapper).
       This property can also be set on individual declarative
       classes using the "__mapper_cls__" property.

- postgres
    - Got PG server side cursors back into shape, added fixed
      unit tests as part of the default test suite.  Added
      better uniqueness to the cursor ID [ticket:1001]

- oracle
    - The "owner" keyword on Table is now deprecated, and is
      exactly synonymous with the "schema" keyword.  Tables can
      now be reflected with alternate "owner" attributes,
      explicitly stated on the Table object or not using
      "schema".

    - All of the "magic" searching for synonyms, DBLINKs etc.
      during table reflection are disabled by default unless you
      specify "oracle_resolve_synonyms=True" on the Table
      object.  Resolving synonyms necessarily leads to some
      messy guessing which we'd rather leave off by default.
      When the flag is set, tables and related tables will be
      resolved against synonyms in all cases, meaning if a
      synonym exists for a particular table, reflection will use
      it when reflecting related tables.  This is stickier
      behavior than before which is why it's off by default.

- mssql
     - Reflected tables will now automatically load other tables
       which are referenced by Foreign keys in the auto-loaded
       table, [ticket:979].

     - Added executemany check to skip identity fetch, [ticket:916].

     - Added stubs for small date type, [ticket:884]

     - Added a new 'driver' keyword parameter for the pyodbc dialect.
       Will substitute into the ODBC connection string if given,
       defaults to 'SQL Server'.

     - Added a new 'max_identifier_length' keyword parameter for
       the pyodbc dialect.

     - Improvements to pyodbc + Unix. If you couldn't get that
       combination to work before, please try again.

- mysql
     - The connection.info keys the dialect uses to cache server
       settings have changed and are now namespaced.

0.4.4
------
- sql
    - Can again create aliases of selects against textual FROM
      clauses, [ticket:975]

    - The value of a bindparam() can be a callable, in which
      case it's evaluated at statement execution time to get the
      value.

    - Added exception wrapping/reconnect support to result set
      fetching.  Reconnect works for those databases that raise
      a catchable data error during results (i.e. doesn't work
      on MySQL) [ticket:978]

    - Implemented two-phase API for "threadlocal" engine, via
      engine.begin_twophase(), engine.prepare() [ticket:936]

    - Fixed bug which was preventing UNIONS from being
      cloneable, [ticket:986]

    - Added "bind" keyword argument to insert(), update(),
      delete() and DDL(). The .bind property is now assignable
      on those statements as well as on select().

    - Insert statements can now be compiled with extra "prefix"
      words between INSERT and INTO, for vendor extensions like
      MySQL's INSERT IGNORE INTO table.

- orm
    - any(), has(), contains(), ~contains(), attribute level ==
      and != now work properly with self-referential relations -
      the clause inside the EXISTS is aliased on the "remote"
      side to distinguish it from the parent table.  This
      applies to single table self-referential as well as
      inheritance-based self-referential.

    - Repaired behavior of == and != operators at the relation()
      level when compared against NULL for one-to-one relations
      [ticket:985]

    - Fixed bug whereby session.expire() attributes were not
      loading on an polymorphically-mapped instance mapped by a
      select_table mapper.

    - Added query.with_polymorphic() - specifies a list of
      classes which descend from the base class, which will be
      added to the FROM clause of the query.  Allows subclasses
      to be used within filter() criterion as well as eagerly
      loads the attributes of those subclasses.

    - Your cries have been heard: removing a pending item from
      an attribute or collection with delete-orphan expunges the
      item from the session; no FlushError is raised.  Note that
      if you session.save()'ed the pending item explicitly, the
      attribute/collection removal still knocks it out.

    - session.refresh() and session.expire() raise an error when
      called on instances which are not persistent within the
      session

    - Fixed potential generative bug when the same Query was
      used to generate multiple Query objects using join().

    - Fixed bug which was introduced in 0.4.3, whereby loading
      an already-persistent instance mapped with joined table
      inheritance would trigger a useless "secondary" load from
      its joined table, when using the default "select"
      polymorphic_fetch.  This was due to attributes being
      marked as expired during its first load and not getting
      unmarked from the previous "secondary" load.  Attributes
      are now unexpired based on presence in __dict__ after any
      load or commit operation succeeds.

    - Deprecated Query methods apply_sum(), apply_max(),
      apply_min(), apply_avg().  Better methodologies are
      coming....

    - relation() can accept a callable for its first argument,
      which returns the class to be related.  This is in place
      to assist declarative packages to define relations without
      classes yet being in place.

    - Added a new "higher level" operator called "of_type()":
      used in join() as well as with any() and has(), qualifies
      the subclass which will be used in filter criterion, e.g.:

        query.filter(Company.employees.of_type(Engineer).
          any(Engineer.name=='foo'))

        or

        query.join(Company.employees.of_type(Engineer)).
          filter(Engineer.name=='foo')

    - Preventive code against a potential lost-reference bug in
      flush().

    - Expressions used in filter(), filter_by() and others, when
      they make usage of a clause generated from a relation
      using the identity of a child object (e.g.,
      filter(Parent.child==<somechild>)), evaluate the actual
      primary key value of <somechild> at execution time so that
      the autoflush step of the Query can complete, thereby
      populating the PK value of <somechild> in the case that
      <somechild> was pending.

    - setting the relation()-level order by to a column in the
      many-to-many "secondary" table will now work with eager
      loading, previously the "order by" wasn't aliased against
      the secondary table's alias.

    - Synonyms riding on top of existing descriptors are now
      full proxies to those descriptors.

- dialects
    - Invalid SQLite connection URLs now raise an error.

    - postgres TIMESTAMP renders correctly [ticket:981]

    - postgres PGArray is a "mutable" type by default; when used
      with the ORM, mutable-style equality/ copy-on-write
      techniques are used to test for changes.

- extensions
    - a new super-small "declarative" extension has been added,
      which allows Table and mapper() configuration to take
      place inline underneath a class declaration.  This
      extension differs from ActiveMapper and Elixir in that it
      does not redefine any SQLAlchemy semantics at all; literal
      Column, Table and relation() constructs are used to define
      the class behavior and table definition.

0.4.3
------
- sql
    - Added "schema.DDL", an executable free-form DDL statement.
      DDLs can be executed in isolation or attached to Table or
      MetaData instances and executed automatically when those
      objects are created and/or dropped.

    - Table columns and constraints can be overridden on a an
      existing table (such as a table that was already reflected)
      using the 'useexisting=True' flag, which now takes into
      account the arguments passed along with it.

    - Added a callable-based DDL events interface, adds hooks
      before and after Tables and MetaData create and drop.

    - Added generative where(<criterion>) method to delete() and
      update() constructs which return a new object with criterion
      joined to existing criterion via AND, just like
      select().where().

    - Added "ilike()" operator to column operations.  Compiles to
      ILIKE on postgres, lower(x) LIKE lower(y) on all
      others. [ticket:727]

    - Added "now()" as a generic function; on SQLite, Oracle
      and MSSQL compiles as "CURRENT_TIMESTAMP"; "now()" on
      all others. [ticket:943]

    - The startswith(), endswith(), and contains() operators now
      concatenate the wildcard operator with the given operand in
      SQL, i.e. "'%' || <bindparam>" in all cases, accept
      text('something') operands properly [ticket:962]

    - cast() accepts text('something') and other non-literal
      operands properly [ticket:962]

    - fixed bug in result proxy where anonymously generated
      column labels would not be accessible using their straight
      string name

    - Deferrable constraints can now be defined.

    - Added "autocommit=True" keyword argument to select() and
      text(), as well as generative autocommit() method on
      select(); for statements which modify the database through
      some user-defined means other than the usual INSERT/UPDATE/
      DELETE etc.  This flag will enable "autocommit" behavior
      during execution if no transaction is in progress.
      [ticket:915]

    - The '.c.' attribute on a selectable now gets an entry for
      every column expression in its columns clause.  Previously,
      "unnamed" columns like functions and CASE statements weren't
      getting put there.  Now they will, using their full string
      representation if no 'name' is available.

    - a CompositeSelect, i.e. any union(), union_all(),
      intersect(), etc. now asserts that each selectable contains
      the same number of columns.  This conforms to the
      corresponding SQL requirement.

    - The anonymous 'label' generated for otherwise unlabeled
      functions and expressions now propagates outwards at compile
      time for expressions like select([select([func.foo()])]).

    - Building on the above ideas, CompositeSelects now build up
      their ".c." collection based on the names present in the
      first selectable only; corresponding_column() now works
      fully for all embedded selectables.

    - Oracle and others properly encode SQL used for defaults like
      sequences, etc., even if no unicode idents are used since
      identifier preparer may return a cached unicode identifier.

    - Column and clause comparisons to datetime objects on the
      left hand side of the expression now work (d < table.c.col).
      (datetimes on the RHS have always worked, the LHS exception
      is a quirk of the datetime implementation.)

- orm
    - Every Session.begin() must now be accompanied by a
      corresponding commit() or rollback() unless the session is
      closed with Session.close().  This also includes the begin()
      which is implicit to a session created with
      transactional=True.  The biggest change introduced here is
      that when a Session created with transactional=True raises
      an exception during flush(), you must call
      Session.rollback() or Session.close() in order for that
      Session to continue after an exception.

    - Fixed merge() collection-doubling bug when merging transient
      entities with backref'ed collections.  [ticket:961]

    - merge(dont_load=True) does not accept transient entities,
      this is in continuation with the fact that
      merge(dont_load=True) does not accept any "dirty" objects
      either.

    - Added standalone "query" class attribute generated by a
      scoped_session.  This provides MyClass.query without using
      Session.mapper.  Use via:

        MyClass.query = Session.query_property()

    - The proper error message is raised when trying to access
      expired instance attributes with no session present

    - dynamic_loader() / lazy="dynamic" now accepts and uses
      the order_by parameter in the same way in which it works
      with relation().

    - Added expire_all() method to Session.  Calls expire() for
      all persistent instances.  This is handy in conjunction
      with...

    - Instances which have been partially or fully expired will
      have their expired attributes populated during a regular
      Query operation which affects those objects, preventing a
      needless second SQL statement for each instance.

    - Dynamic relations, when referenced, create a strong
      reference to the parent object so that the query still has a
      parent to call against even if the parent is only created
      (and otherwise dereferenced) within the scope of a single
      expression. [ticket:938]

    - Added a mapper() flag "eager_defaults". When set to True,
      defaults that are generated during an INSERT or UPDATE
      operation are post-fetched immediately, instead of being
      deferred until later.  This mimics the old 0.3 behavior.

    - query.join() can now accept class-mapped attributes as
      arguments. These can be used in place or in any combination
      with strings.  In particular this allows construction of
      joins to subclasses on a polymorphic relation, i.e.:

        query(Company).join(['employees', Engineer.name])

    - query.join() can also accept tuples of attribute name/some
      selectable as arguments.  This allows construction of joins
      *from* subclasses of a polymorphic relation, i.e.:

        query(Company).\
        join(
          [('employees', people.join(engineer)), Engineer.name]
        )

    - General improvements to the behavior of join() in
      conjunction with polymorphic mappers, i.e. joining from/to
      polymorphic mappers and properly applying aliases.

    - Fixed/improved behavior when a mapper determines the natural
      "primary key" of a mapped join, it will more effectively
      reduce columns which are equivalent via foreign key
      relation.  This affects how many arguments need to be sent
      to query.get(), among other things.  [ticket:933]

    - The lazy loader can now handle a join condition where the
      "bound" column (i.e. the one that gets the parent id sent as
      a bind parameter) appears more than once in the join
      condition.  Specifically this allows the common task of a
      relation() which contains a parent-correlated subquery, such
      as "select only the most recent child item". [ticket:946]

    - Fixed bug in polymorphic inheritance where an incorrect
      exception is raised when base polymorphic_on column does not
      correspond to any columns within the local selectable of an
      inheriting mapper more than one level deep

    - Fixed bug in polymorphic inheritance which made it difficult
      to set a working "order_by" on a polymorphic mapper.

    - Fixed a rather expensive call in Query that was slowing down
      polymorphic queries.

    - "Passive defaults" and other "inline" defaults can now be
      loaded during a flush() call if needed; in particular, this
      allows constructing relations() where a foreign key column
      references a server-side-generated, non-primary-key
      column. [ticket:954]

    - Additional Session transaction fixes/changes:
        - Fixed bug with session transaction management: parent
          transactions weren't started on the connection when
          adding a connection to a nested transaction.

        - session.transaction now always refers to the innermost
          active transaction, even when commit/rollback are called
          directly on the session transaction object.

        - Two-phase transactions can now be prepared.

        - When preparing a two-phase transaction fails on one
          connection, all the connections are rolled back.

        - session.close() didn't close all transactions when
          nested transactions were used.

        - rollback() previously erroneously set the current
          transaction directly to the parent of the transaction
          that could be rolled back to. Now it rolls back the next
          transaction up that can handle it, but sets the current
          transaction to it's parent and inactivates the
          transactions in between. Inactive transactions can only
          be rolled back or closed, any other call results in an
          error.

        - autoflush for commit() wasn't flushing for simple
          subtransactions.

        - unitofwork flush didn't close the failed transaction
          when the session was not in a transaction and commiting
          the transaction failed.

    - Miscellaneous tickets: [ticket:940] [ticket:964]

- general
    - Fixed a variety of hidden and some not-so-hidden
      compatibility issues for Python 2.3, thanks to new support
      for running the full test suite on 2.3.

    - Warnings are now issued as type exceptions.SAWarning.

- dialects
    - Better support for schemas in SQLite (linked in by ATTACH
      DATABASE ... AS name).  In some cases in the past, schema
      names were ommitted from generated SQL for SQLite.  This is
      no longer the case.

    - table_names on SQLite now picks up temporary tables as well.

    - Auto-detect an unspecified MySQL ANSI_QUOTES mode during
      reflection operations, support for changing the mode
      midstream.  Manual mode setting is still required if no
      reflection is used.

    - Fixed reflection of TIME columns on SQLite.

    - Finally added PGMacAddr type to postgres [ticket:580]

    - Reflect the sequence associated to a PK field (typically
      with a BEFORE INSERT trigger) under Firebird

    - Oracle assembles the correct columns in the result set
      column mapping when generating a LIMIT/OFFSET subquery,
      allows columns to map properly to result sets even if
      long-name truncation kicks in [ticket:941]

    - MSSQL now includes EXEC in the _is_select regexp, which
      should allow row-returning stored procedures to be used.

    - MSSQL now includes an experimental implementation of
      LIMIT/OFFSET using the ANSI SQL row_number() function, so it
      requires MSSQL-2005 or higher. To enable the feature, add
      "has_window_funcs" to the keyword arguments for connect, or
      add "?has_window_funcs=1" to your dburi query arguments.

- ext
    - Changed ext.activemapper to use a non-transactional session
      for the objectstore.

    - Fixed output order of "['a'] + obj.proxied" binary operation
      on association-proxied lists.

0.4.2p3
------
- general
    - sub version numbering scheme changed to suite
      setuptools version number rules; easy_install -u
      should now get this version over 0.4.2.

- sql
    - Text type is properly exported now and does not
      raise a warning on DDL create; String types with no
      length only raise warnings during CREATE TABLE
      [ticket:912]

    - new UnicodeText type is added, to specify an
      encoded, unlengthed Text type

    - fixed bug in union() so that select() statements
      which don't derive from FromClause objects can be
      unioned

- orm
    - fixed bug with session.dirty when using "mutable
      scalars" (such as PickleTypes)

    - added a more descriptive error message when flushing
      on a relation() that has non-locally-mapped columns
      in its primary or secondary join condition

- dialects
    - Fixed reflection of mysql empty string column
      defaults.

0.4.2b  (0.4.2p2)
------
- sql
    - changed name of TEXT to Text since its a "generic"
      type; TEXT name is deprecated until 0.5. The
      "upgrading" behavior of String to Text when no
      length is present is also deprecated until 0.5; will
      issue a warning when used for CREATE TABLE
      statements (String with no length for SQL expression
      purposes is still fine) [ticket:912]

    - generative select.order_by(None) / group_by(None)
      was not managing to reset order by/group by
      criterion, fixed [ticket:924]

- orm
    - suppressing *all* errors in
      InstanceState.__cleanup() now.

    - fixed an attribute history bug whereby assigning a
      new collection to a collection-based attribute which
      already had pending changes would generate incorrect
      history [ticket:922]

    - fixed delete-orphan cascade bug whereby setting the
      same object twice to a scalar attribute could log it
      as an orphan [ticket:925]

    - Fixed cascades on a += assignment to a list-based
      relation.

    - synonyms can now be created against props that don't
      exist yet, which are later added via add_property().
      This commonly includes backrefs. (i.e. you can make
      synonyms for backrefs without worrying about the
      order of operations) [ticket:919]

    - fixed bug which could occur with polymorphic "union"
      mapper which falls back to "deferred" loading of
      inheriting tables

    - the "columns" collection on a mapper/mapped class
      (i.e. 'c') is against the mapped table, not the
      select_table in the case of polymorphic "union"
      loading (this shouldn't be noticeable).

- ext
    - '+', '*', '+=' and '*=' support for association
      proxied lists.

- dialects
    - mssql - narrowed down the test for "date"/"datetime"
      in MSDate/ MSDateTime subclasses so that incoming
      "datetime" objects don't get mis-interpreted as
      "date" objects and vice versa, [ticket:923]

0.4.2a   (0.4.2p1)
------

- orm
    - fixed fairly critical bug whereby the same instance could be listed
      more than once in the unitofwork.new collection; most typically
      reproduced when using a combination of inheriting mappers and
      ScopedSession.mapper, as the multiple __init__ calls per instance
      could save() the object with distinct _state objects

    - added very rudimentary yielding iterator behavior to Query.  Call
      query.yield_per(<number of rows>) and evaluate the Query in an
      iterative context; every collection of N rows will be packaged up
      and yielded.  Use this method with extreme caution since it does
      not attempt to reconcile eagerly loaded collections across
      result batch boundaries, nor will it behave nicely if the same
      instance occurs in more than one batch.  This means that an eagerly
      loaded collection will get cleared out if it's referenced in more than
      one batch, and in all cases attributes will be overwritten on instances
      that occur in more than one batch.

   - Fixed in-place set mutation operators for set collections and association
     proxied sets. [ticket:920]

- dialects
    - Fixed the missing call to subtype result processor for the PGArray
      type. [ticket:913]

0.4.2
-----
- sql
    - generic functions ! we introduce a database of known SQL functions, such
      as current_timestamp, coalesce, and create explicit function objects
      representing them. These objects have constrained argument lists, are
      type aware, and can compile in a dialect-specific fashion. So saying
      func.char_length("foo", "bar") raises an error (too many args),
      func.coalesce(datetime.date(2007, 10, 5), datetime.date(2005, 10, 15))
      knows that its return type is a Date. We only have a few functions
      represented so far but will continue to add to the system [ticket:615]

    - auto-reconnect support improved; a Connection can now automatically
      reconnect after its underlying connection is invalidated, without
      needing to connect() again from the engine.  This allows an ORM session
      bound to a single Connection to not need a reconnect.
      Open transactions on the Connection must be rolled back after an invalidation
      of the underlying connection else an error is raised.  Also fixed
      bug where disconnect detect was not being called for cursor(), rollback(),
      or commit().

    - added new flag to String and create_engine(),
      assert_unicode=(True|False|'warn'|None). Defaults to `False` or `None` on
      create_engine() and String, `'warn'` on the Unicode type. When `True`,
      results in all unicode conversion operations raising an exception when a
      non-unicode bytestring is passed as a bind parameter. 'warn' results
      in a warning. It is strongly advised that all unicode-aware applications
      make proper use of Python unicode objects (i.e. u'hello' and not 'hello')
      so that data round trips accurately.

    - generation of "unique" bind parameters has been simplified to use the same
      "unique identifier" mechanisms as everything else.  This doesn't affect
      user code, except any code that might have been hardcoded against the generated
      names.  Generated bind params now have the form "<paramname>_<num>",
      whereas before only the second bind of the same name would have this form.

    - select().as_scalar() will raise an exception if the select does not have
      exactly one expression in its columns clause.

    - bindparam() objects themselves can be used as keys for execute(), i.e.
      statement.execute({bind1:'foo', bind2:'bar'})

    - added new methods to TypeDecorator, process_bind_param() and
      process_result_value(), which automatically take advantage of the processing
      of the underlying type.  Ideal for using with Unicode or Pickletype.
      TypeDecorator should now be the primary way to augment the behavior of any
      existing type including other TypeDecorator subclasses such as PickleType.

    - selectables (and others) will issue a warning when two columns in
      their exported columns collection conflict based on name.

    - tables with schemas can still be used in sqlite, firebird,
      schema name just gets dropped [ticket:890]

    - changed the various "literal" generation functions to use an anonymous
      bind parameter.  not much changes here except their labels now look
      like ":param_1", ":param_2" instead of ":literal"

    - column labels in the form "tablename.columname", i.e. with a dot, are now
      supported.

    - from_obj keyword argument to select() can be a scalar or a list.

- orm
   - a major behavioral change to collection-based backrefs: they no
     longer trigger lazy loads !  "reverse" adds and removes
     are queued up and are merged with the collection when it is
     actually read from and loaded; but do not trigger a load beforehand.
     For users who have noticed this behavior, this should be much more
     convenient than using dynamic relations in some cases; for those who
     have not, you might notice your apps using a lot fewer queries than
     before in some situations. [ticket:871]

   - mutable primary key support is added. primary key columns can be
     changed freely, and the identity of the instance will change upon
     flush. In addition, update cascades of foreign key referents (primary
     key or not) along relations are supported, either in tandem with the
     database's ON UPDATE CASCADE (required for DB's like Postgres) or
     issued directly by the ORM in the form of UPDATE statements, by setting
     the flag "passive_cascades=False".

   - inheriting mappers now inherit the MapperExtensions of their parent
     mapper directly, so that all methods for a particular MapperExtension
     are called for subclasses as well.  As always, any MapperExtension
     can return either EXT_CONTINUE to continue extension processing
     or EXT_STOP to stop processing.  The order of mapper resolution is:
     <extensions declared on the classes mapper> <extensions declared on the
     classes' parent mapper> <globally declared extensions>.

     Note that if you instantiate the same extension class separately
     and then apply it individually for two mappers in the same inheritance
     chain, the extension will be applied twice to the inheriting class,
     and each method will be called twice.

     To apply a mapper extension explicitly to each inheriting class but
     have each method called only once per operation, use the same
     instance of the extension for both mappers.
     [ticket:490]

   - MapperExtension.before_update() and after_update() are now called
     symmetrically; previously, an instance that had no modified column
     attributes (but had a relation() modification) could be called with
     before_update() but not after_update() [ticket:907]

   - columns which are missing from a Query's select statement
     now get automatically deferred during load.

   - mapped classes which extend "object" and do not provide an
     __init__() method will now raise TypeError if non-empty *args
     or **kwargs are present at instance construction time (and are
     not consumed by any extensions such as the scoped_session mapper),
     consistent with the behavior of normal Python classes [ticket:908]

   - fixed Query bug when filter_by() compares a relation against None
     [ticket:899]

   - improved support for pickling of mapped entities.  Per-instance
     lazy/deferred/expired callables are now serializable so that
     they serialize and deserialize with _state.

   - new synonym() behavior: an attribute will be placed on the mapped
     class, if one does not exist already, in all cases. if a property
     already exists on the class, the synonym will decorate the property
     with the appropriate comparison operators so that it can be used in in
     column expressions just like any other mapped attribute (i.e. usable in
     filter(), etc.) the "proxy=True" flag is deprecated and no longer means
     anything. Additionally, the flag "map_column=True" will automatically
     generate a ColumnProperty corresponding to the name of the synonym,
     i.e.: 'somename':synonym('_somename', map_column=True) will map the
     column named 'somename' to the attribute '_somename'. See the example
     in the mapper docs. [ticket:801]

   - Query.select_from() now replaces all existing FROM criterion with
     the given argument; the previous behavior of constructing a list
     of FROM clauses was generally not useful as is required
     filter() calls to create join criterion, and new tables introduced
     within filter() already add themselves to the FROM clause.  The
     new behavior allows not just joins from the main table, but select
     statements as well.  Filter criterion, order bys, eager load
     clauses will be "aliased" against the given statement.

   - this month's refactoring of attribute instrumentation changes
     the "copy-on-load" behavior we've had since midway through 0.3
     with "copy-on-modify" in most cases.  This takes a sizable chunk
     of latency out of load operations and overall does less work
     as only attributes which are actually modified get their
     "committed state" copied.  Only "mutable scalar" attributes
     (i.e. a pickled object or other mutable item), the reason for
     the copy-on-load change in the first place, retain the old
     behavior.

   - a slight behavioral change to attributes is, del'ing an attribute
     does *not* cause the lazyloader of that attribute to fire off again;
     the "del" makes the effective value of the attribute "None".  To
     re-trigger the "loader" for an attribute, use
     session.expire(instance, [attrname]).

   - query.filter(SomeClass.somechild == None), when comparing
     a many-to-one property to None, properly generates "id IS NULL"
     including that the NULL is on the right side.

   - query.order_by() takes into account aliased joins, i.e.
     query.join('orders', aliased=True).order_by(Order.id)

   - eagerload(), lazyload(), eagerload_all() take an optional
     second class-or-mapper argument, which will select the mapper
     to apply the option towards.  This can select among other
     mappers which were added using add_entity().

   - eagerloading will work with mappers added via add_entity().

   - added "cascade delete" behavior to "dynamic" relations just like
     that of regular relations.  if passive_deletes flag (also just added)
     is not set, a delete of the parent item will trigger a full load of
     the child items so that they can be deleted or updated accordingly.

   - also with dynamic, implemented correct count() behavior as well
     as other helper methods.

   - fix to cascades on polymorphic relations, such that cascades
     from an object to a polymorphic collection continue cascading
     along the set of attributes specific to each element in the collection.

   - query.get() and query.load() do not take existing filter or other
     criterion into account; these methods *always* look up the given id
     in the database or return the current instance from the identity map,
     disregarding any existing filter, join, group_by or other criterion
     which has been configured. [ticket:893]

   - added support for version_id_col in conjunction with inheriting mappers.
     version_id_col is typically set on the base mapper in an inheritance
     relationship where it takes effect for all inheriting mappers.
     [ticket:883]

   - relaxed rules on column_property() expressions having labels; any
     ColumnElement is accepted now, as the compiler auto-labels non-labeled
     ColumnElements now.  a selectable, like a select() statement, still
     requires conversion to ColumnElement via as_scalar() or label().

   - fixed backref bug where you could not del instance.attr if attr
     was None

   - several ORM attributes have been removed or made private:
     mapper.get_attr_by_column(), mapper.set_attr_by_column(),
     mapper.pks_by_table, mapper.cascade_callable(),
     MapperProperty.cascade_callable(), mapper.canload(),
     mapper.save_obj(), mapper.delete_obj(), mapper._mapper_registry,
     attributes.AttributeManager

   - Assigning an incompatible collection type to a relation attribute now
     raises TypeError instead of sqlalchemy's ArgumentError.

   - Bulk assignment of a MappedCollection now raises an error if a key in the
     incoming dictionary does not match the key that the collection's keyfunc
     would use for that value. [ticket:886]

   - Custom collections can now specify a @converter method to translate
     objects used in "bulk" assignment into a stream of values, as in::

        obj.col = [newval1, newval2]
        # or
        obj.dictcol = {'foo': newval1, 'bar': newval2}

     The MappedCollection uses this hook to ensure that incoming key/value
     pairs are sane from the collection's perspective.

   - fixed endless loop issue when using lazy="dynamic" on both
     sides of a bi-directional relationship [ticket:872]

   - more fixes to the LIMIT/OFFSET aliasing applied with Query + eagerloads,
     in this case when mapped against a select statement [ticket:904]

   - fix to self-referential eager loading such that if the same mapped
     instance appears in two or more distinct sets of columns in the same
     result set, its eagerly loaded collection will be populated regardless
     of whether or not all of the rows contain a set of "eager" columns for
     that collection.  this would also show up as a KeyError when fetching
     results with join_depth turned on.

   - fixed bug where Query would not apply a subquery to the SQL when LIMIT
     was used in conjunction with an inheriting mapper where the eager
     loader was only in the parent mapper.

   - clarified the error message which occurs when you try to update()
     an instance with the same identity key as an instance already present
     in the session.

   - some clarifications and fixes to merge(instance, dont_load=True).
     fixed bug where lazy loaders were getting disabled on returned instances.
     Also, we currently do not support merging an instance which has uncommitted
     changes on it, in the case that dont_load=True is used....this will
     now raise an error.  This is due to complexities in merging the
     "committed state" of the given instance to correctly correspond to the
     newly copied instance, as well as other modified state.
     Since the use case for dont_load=True is caching, the given instances
     shouldn't have any uncommitted changes on them anyway.
     We also copy the instances over without using any events now, so that
     the 'dirty' list on the new session remains unaffected.

   - fixed bug which could arise when using session.begin_nested() in conjunction
     with more than one level deep of enclosing session.begin() statements

   - fixed session.refresh() with instance that has custom entity_name
     [ticket:914]

- dialects

   - sqlite SLDate type will not erroneously render "microseconds" portion
     of a datetime or time object.

   - oracle
      - added disconnect detection support for Oracle
      - some cleanup to binary/raw types so that cx_oracle.LOB is detected
        on an ad-hoc basis [ticket:902]

   - MSSQL
      - PyODBC no longer has a global "set nocount on".
      - Fix non-identity integer PKs on autload [ticket:824]
      - Better support for convert_unicode [ticket:839]
      - Less strict date conversion for pyodbc/adodbapi [ticket:842]
      - Schema-qualified tables / autoload [ticket:901]

   - Firebird backend

     - does properly reflect domains (partially fixing [ticket:410]) and
       PassiveDefaults

     - reverted to use default poolclass (was set to SingletonThreadPool in
       0.4.0 [3562] for test purposes)

     - map func.length() to 'char_length' (easily overridable with the UDF
       'strlen' on old versions of Firebird)


0.4.1
-----

- sql

  - the "shortname" keyword parameter on bindparam() has been
    deprecated.

  - Added contains operator (generates a "LIKE %<other>%" clause).

  - anonymous column expressions are automatically labeled.
    e.g. select([x* 5]) produces "SELECT x * 5 AS anon_1".
    This allows the labelname to be present in the cursor.description
    which can then be appropriately matched to result-column processing
    rules. (we can't reliably use positional tracking for result-column
    matches since text() expressions may represent multiple columns).

  - operator overloading is now controlled by TypeEngine objects - the
    one built-in operator overload so far is String types overloading
    '+' to be the string concatenation operator.
    User-defined types can also define their own operator overloading
    by overriding the adapt_operator(self, op) method.

  - untyped bind parameters on the right side of a binary expression
    will be assigned the type of the left side of the operation, to better
    enable the appropriate bind parameter processing to take effect
    [ticket:819]

  - Removed regular expression step from most statement compilations.
    Also fixes [ticket:833]

  - Fixed empty (zero column) sqlite inserts, allowing inserts on
    autoincrementing single column tables.

  - Fixed expression translation of text() clauses; this repairs various
    ORM scenarios where literal text is used for SQL expressions

  - Removed ClauseParameters object; compiled.params returns a regular
    dictionary now, as well as result.last_inserted_params() /
    last_updated_params().

  - Fixed INSERT statements w.r.t. primary key columns that have
    SQL-expression based default generators on them; SQL expression
    executes inline as normal but will not trigger a "postfetch" condition
    for the column, for those DB's who provide it via cursor.lastrowid

  - func. objects can be pickled/unpickled [ticket:844]

  - rewrote and simplified the system used to "target" columns across
    selectable expressions.  On the SQL side this is represented by the
    "corresponding_column()" method. This method is used heavily by the ORM
    to "adapt" elements of an expression to similar, aliased expressions,
    as well as to target result set columns originally bound to a
    table or selectable to an aliased, "corresponding" expression.  The new
    rewrite features completely consistent and accurate behavior.

  - Added a field ("info") for storing arbitrary data on schema items
    [ticket:573]

  - The "properties" collection on Connections has been renamed "info" to
    match schema's writable collections.  Access is still available via
    the "properties" name until 0.5.

  - fixed the close() method on Transaction when using strategy='threadlocal'

  - fix to compiled bind parameters to not mistakenly populate None
    [ticket:853]

  - <Engine|Connection>._execute_clauseelement becomes a public method
    Connectable.execute_clauseelement

- orm
  - eager loading with LIMIT/OFFSET applied no longer adds the primary
    table joined to a limited subquery of itself; the eager loads now
    join directly to the subquery which also provides the primary table's
    columns to the result set.  This eliminates a JOIN from all eager loads
    with LIMIT/OFFSET.  [ticket:843]

  - session.refresh() and session.expire() now support an additional argument
    "attribute_names", a list of individual attribute keynames to be refreshed
    or expired, allowing partial reloads of attributes on an already-loaded
    instance. [ticket:802]

  - added op() operator to instrumented attributes; i.e.
    User.name.op('ilike')('%somename%') [ticket:767]

  - Mapped classes may now define __eq__, __hash__, and __nonzero__ methods
    with arbitrary semantics.  The orm now handles all mapped instances on
    an identity-only basis. (e.g. 'is' vs '==') [ticket:676]

  - the "properties" accessor on Mapper is removed; it now throws an informative
    exception explaining the usage of mapper.get_property() and
    mapper.iterate_properties

  - added having() method to Query, applies HAVING to the generated statement
    in the same way as filter() appends to the WHERE clause.

  - The behavior of query.options() is now fully based on paths, i.e. an
    option such as eagerload_all('x.y.z.y.x') will apply eagerloading to
    only those paths, i.e. and not 'x.y.x'; eagerload('children.children')
    applies only to exactly two-levels deep, etc. [ticket:777]

  - PickleType will compare using `==` when set up with mutable=False,
    and not the `is` operator.  To use `is` or any other comparator, send
    in a custom comparison function using PickleType(comparator=my_custom_comparator).

  - query doesn't throw an error if you use distinct() and an order_by()
    containing UnaryExpressions (or other) together [ticket:848]

  - order_by() expressions from joined tables are properly added to columns
    clause when using distinct() [ticket:786]

  - fixed error where Query.add_column() would not accept a class-bound
    attribute as an argument; Query also raises an error if an invalid
    argument was sent to add_column() (at instances() time) [ticket:858]

  - added a little more checking for garbage-collection dereferences in
    InstanceState.__cleanup() to reduce "gc ignored" errors on app
    shutdown

  - The session API has been solidified:

    - It's an error to session.save() an object which is already
      persistent [ticket:840]

    - It's an error to session.delete() an object which is *not*
      persistent.

    - session.update() and session.delete() raise an error when updating
      or deleting an instance that is already in the session with a
      different identity.

    - The session checks more carefully when determining "object X already
      in another session"; e.g. if you pickle a series of objects and
      unpickle (i.e. as in a Pylons HTTP session or similar), they can go
      into a new session without any conflict

    - merge() includes a keyword argument "dont_load=True".  setting this
      flag will cause the merge operation to not load any data from the
      database in response to incoming detached objects, and will accept
      the incoming detached object as though it were already present in
      that session.  Use this to merge detached objects from external
      caching systems into the session.

  - Deferred column attributes no longer trigger a load operation when the
    attribute is assigned to.  In those cases, the newly assigned value
    will be present in the flushes' UPDATE statement unconditionally.

  - Fixed a truncation error when re-assigning a subset of a collection
    (obj.relation = obj.relation[1:]) [ticket:834]

  - De-cruftified backref configuration code, backrefs which step on
    existing properties now raise an error [ticket:832]

  - Improved behavior of add_property() etc., fixed [ticket:831] involving
    synonym/deferred.

  - Fixed clear_mappers() behavior to better clean up after itself.

  - Fix to "row switch" behavior, i.e. when an INSERT/DELETE is combined
    into a single UPDATE; many-to-many relations on the parent object
    update properly.  [ticket:841]

  - Fixed __hash__ for association proxy- these collections are unhashable,
    just like their mutable Python counterparts.

  - Added proxying of save_or_update, __contains__ and __iter__ methods for
    scoped sessions.

  - fixed very hard-to-reproduce issue where by the FROM clause of Query
    could get polluted by certain generative calls [ticket:852]

- dialects

  - Added experimental support for MaxDB (versions >= 7.6.03.007 only).

  - oracle will now reflect "DATE" as an OracleDateTime column, not
    OracleDate

  - added awareness of schema name in oracle table_names() function,
    fixes metadata.reflect(schema='someschema') [ticket:847]

  - MSSQL anonymous labels for selection of functions made deterministic

  - sqlite will reflect "DECIMAL" as a numeric column.

  - Made access dao detection more reliable [ticket:828]

  - Renamed the Dialect attribute 'preexecute_sequences' to
    'preexecute_pk_sequences'.  An attribute porxy is in place for
    out-of-tree dialects using the old name.

  - Added test coverage for unknown type reflection. Fixed sqlite/mysql
    handling of type reflection for unknown types.

  - Added REAL for mysql dialect (for folks exploiting the
    REAL_AS_FLOAT sql mode).

  - mysql Float, MSFloat and MSDouble constructed without arguments
    now produce no-argument DDL, e.g.'FLOAT'.

- misc

  - Removed unused util.hash().


0.4.0
-----

- (see 0.4.0beta1 for the start of major changes against 0.3,
  as well as http://www.sqlalchemy.org/trac/wiki/WhatsNewIn04 )

- Added initial Sybase support (mxODBC so far) [ticket:785]

- Added partial index support for PostgreSQL. Use the postgres_where keyword
  on the Index.

- string-based query param parsing/config file parser understands
  wider range of string values for booleans [ticket:817]

- backref remove object operation doesn't fail if the other-side
  collection doesn't contain the item, supports noload collections
  [ticket:813]

- removed __len__ from "dynamic" collection as it would require issuing
  a SQL "count()" operation, thus forcing all list evaluations to issue
  redundant SQL [ticket:818]

- inline optimizations added to locate_dirty() which can greatly speed up
  repeated calls to flush(), as occurs with autoflush=True [ticket:816]

- The IdentifierPreprarer's _requires_quotes test is now regex based.  Any
  out-of-tree dialects that provide custom sets of legal_characters or
  illegal_initial_characters will need to move to regexes or override
  _requires_quotes.

- Firebird has supports_sane_rowcount and supports_sane_multi_rowcount set
  to False due to ticket #370 (right way).

- Improvements and fixes on Firebird reflection:
  . FBDialect now mimics OracleDialect, regarding case-sensitivity of TABLE and
    COLUMN names (see 'case_sensitive remotion' topic on this current file).
  . FBDialect.table_names() doesn't bring system tables (ticket:796).
  . FB now reflects Column's nullable property correctly.

- Fixed SQL compiler's awareness of top-level column labels as used
  in result-set processing; nested selects which contain the same column
  names don't affect the result or conflict with result-column metadata.

- query.get() and related functions (like many-to-one lazyloading)
  use compile-time-aliased bind parameter names, to prevent
  name conflicts with bind parameters that already exist in the
  mapped selectable.

- Fixed three- and multi-level select and deferred inheritance loading
  (i.e. abc inheritance with no select_table), [ticket:795]

- Ident passed to id_chooser in shard.py always a list.

- The no-arg ResultProxy._row_processor() is now the class attribute
  `_process_row`.

- Added support for returning values from inserts and updates for
  PostgreSQL 8.2+. [ticket:797]

- PG reflection, upon seeing the default schema name being used explicitly
  as the "schema" argument in a Table, will assume that this is the the
  user's desired convention, and will explicitly set the "schema" argument
  in foreign-key-related reflected tables, thus making them match only
  with Table constructors that also use the explicit "schema" argument
  (even though its the default schema).
  In other words, SA assumes the user is being consistent in this usage.

- fixed sqlite reflection of BOOL/BOOLEAN [ticket:808]

- Added support for UPDATE with LIMIT on mysql.

- null foreign key on a m2o doesn't trigger a lazyload [ticket:803]

- oracle does not implicitly convert to unicode for non-typed result
  sets (i.e. when no TypeEngine/String/Unicode type is even being used;
  previously it was detecting DBAPI types and converting regardless).
  should fix [ticket:800]

- fix to anonymous label generation of long table/column names [ticket:806]

- Firebird dialect now uses SingletonThreadPool as poolclass.

- Firebird now uses dialect.preparer to format sequences names

- Fixed breakage with postgres and multiple two-phase transactions. Two-phase
  commits and and rollbacks didn't automatically end up with a new transaction
  as the usual dbapi commits/rollbacks do. [ticket:810]

- Added an option to the _ScopedExt mapper extension to not automatically
  save new objects to session on object initialization.

- fixed Oracle non-ansi join syntax

- PickleType and Interval types (on db not supporting it natively) are now
  slightly faster.

- Added Float and Time types to Firebird (FBFloat and FBTime). Fixed
  BLOB SUB_TYPE for TEXT and Binary types.

- Changed the API for the in_ operator. in_() now accepts a single argument
  that is a sequence of values or a selectable. The old API of passing in
  values as varargs still works but is deprecated.


0.4.0beta6
----------

- The Session identity map is now *weak referencing* by default, use
  weak_identity_map=False to use a regular dict.  The weak dict we are using
  is customized to detect instances which are "dirty" and maintain a
  temporary strong reference to those instances until changes are flushed.

- Mapper compilation has been reorganized such that most compilation occurs
  upon mapper construction.  This allows us to have fewer calls to
  mapper.compile() and also to allow class-based properties to force a
  compilation (i.e. User.addresses == 7 will compile all mappers; this is
  [ticket:758]).  The only caveat here is that an inheriting mapper now
  looks for its inherited mapper upon construction; so mappers within
  inheritance relationships need to be constructed in inheritance order
  (which should be the normal case anyway).

- added "FETCH" to the keywords detected by Postgres to indicate a
  result-row holding statement (i.e. in addition to "SELECT").

- Added full list of SQLite reserved keywords so that they get escaped
  properly.

- Tightened up the relationship between the Query's generation of "eager
  load" aliases, and Query.instances() which actually grabs the eagerly
  loaded rows.  If the aliases were not specifically generated for that
  statement by EagerLoader, the EagerLoader will not take effect when the
  rows are fetched.  This prevents columns from being grabbed accidentally
  as being part of an eager load when they were not meant for such, which
  can happen with textual SQL as well as some inheritance situations.  It's
  particularly important since the "anonymous aliasing" of columns uses
  simple integer counts now to generate labels.

- Removed "parameters" argument from clauseelement.compile(), replaced with
  "column_keys".  The parameters sent to execute() only interact with the
  insert/update statement compilation process in terms of the column names
  present but not the values for those columns.  Produces more consistent
  execute/executemany behavior, simplifies things a bit internally.

- Added 'comparator' keyword argument to PickleType.  By default, "mutable"
  PickleType does a "deep compare" of objects using their dumps()
  representation.  But this doesn't work for dictionaries.  Pickled objects
  which provide an adequate __eq__() implementation can be set up with
  "PickleType(comparator=operator.eq)" [ticket:560]

- Added session.is_modified(obj) method; performs the same "history"
  comparison operation as occurs within a flush operation; setting
  include_collections=False gives the same result as is used when the flush
  determines whether or not to issue an UPDATE for the instance's row.

- Added "schema" argument to Sequence; use this with Postgres /Oracle when
  the sequence is located in an alternate schema.  Implements part of
  [ticket:584], should fix [ticket:761].

- Fixed reflection of the empty string for mysql enums.

- Changed MySQL dialect to use the older LIMIT <offset>, <limit> syntax
  instead of LIMIT <l> OFFSET <o> for folks using 3.23. [ticket:794]

- Added 'passive_deletes="all"' flag to relation(), disables all nulling-out
  of foreign key attributes during a flush where the parent object is
  deleted.

- Column defaults and onupdates, executing inline, will add parenthesis for
  subqueries and other parenthesis-requiring expressions

- The behavior of String/Unicode types regarding that they auto-convert to
  TEXT/CLOB when no length is present now occurs *only* for an exact type of
  String or Unicode with no arguments.  If you use VARCHAR or NCHAR
  (subclasses of String/Unicode) with no length, they will be interpreted by
  the dialect as VARCHAR/NCHAR; no "magic" conversion happens there.  This
  is less surprising behavior and in particular this helps Oracle keep
  string-based bind parameters as VARCHARs and not CLOBs [ticket:793].

- Fixes to ShardedSession to work with deferred columns [ticket:771].

- User-defined shard_chooser() function must accept "clause=None" argument;
  this is the ClauseElement passed to session.execute(statement) and can be
  used to determine correct shard id (since execute() doesn't take an
  instance.)

- Adjusted operator precedence of NOT to match '==' and others, so that
  ~(x <operator> y) produces NOT (x <op> y), which is better compatible
  with older MySQL versions. [ticket:764].  This doesn't apply to "~(x==y)"
  as it does in 0.3 since ~(x==y) compiles to "x != y", but still applies
  to operators like BETWEEN.

- Other tickets: [ticket:768], [ticket:728], [ticket:779], [ticket:757]

0.4.0beta5
----------

- Connection pool fixes; the better performance of beta4 remains but fixes
  "connection overflow" and other bugs which were present (like
  [ticket:754]).

- Fixed bugs in determining proper sync clauses from custom inherit
  conditions. [ticket:769]

- Extended 'engine_from_config' coercion for QueuePool size / overflow.
  [ticket:763]

- mysql views can be reflected again. [ticket:748]

- AssociationProxy can now take custom getters and setters.

- Fixed malfunctioning BETWEEN in orm queries.

- Fixed OrderedProperties pickling [ticket:762]

- SQL-expression defaults and sequences now execute "inline" for all
  non-primary key columns during an INSERT or UPDATE, and for all columns
  during an executemany()-style call. inline=True flag on any insert/update
  statement also forces the same behavior with a single execute().
  result.postfetch_cols() is a collection of columns for which the previous
  single insert or update statement contained a SQL-side default expression.

- Fixed PG executemany() behavior, [ticket:759]

- postgres reflects tables with autoincrement=False for primary key columns
  which have no defaults.

- postgres no longer wraps executemany() with individual execute() calls,
  instead favoring performance.  "rowcount"/"concurrency" checks with
  deleted items (which use executemany) are disabled with PG since psycopg2
  does not report proper rowcount for executemany().

- Tickets fixed:

  - [ticket:742]
  - [ticket:748]
  - [ticket:760]
  - [ticket:762]
  - [ticket:763]

0.4.0beta4
----------

- Tidied up what ends up in your namespace when you 'from sqlalchemy import *':

  - 'table' and 'column' are no longer imported.  They remain available by
    direct reference (as in 'sql.table' and 'sql.column') or a glob import
    from the sql package.  It was too easy to accidentally use a
    sql.expressions.table instead of schema.Table when just starting out
    with SQLAlchemy, likewise column.

  - Internal-ish classes like ClauseElement, FromClause, NullTypeEngine,
    etc., are also no longer imported into your namespace

  - The 'Smallinteger' compatiblity name (small i!) is no longer imported,
    but remains in schema.py for now.  SmallInteger (big I!) is still
    imported.

- The connection pool uses a "threadlocal" strategy internally to return
  the same connection already bound to a thread, for "contextual" connections;
  these are the connections used when you do a "connectionless" execution
  like insert().execute().  This is like a "partial" version of the
  "threadlocal" engine strategy but without the thread-local transaction part
  of it.  We're hoping it reduces connection pool overhead as well as
  database usage.  However, if it proves to impact stability in a negative way,
  we'll roll it right back.

- Fix to bind param processing such that "False" values (like blank strings)
  still get processed/encoded.

- Fix to select() "generative" behavior, such that calling column(),
  select_from(), correlate(), and with_prefix() does not modify the
  original select object [ticket:752]

- Added a "legacy" adapter to types, such that user-defined TypeEngine
  and TypeDecorator classes which define convert_bind_param() and/or
  convert_result_value() will continue to function.  Also supports
  calling the super() version of those methods.

- Added session.prune(), trims away instances cached in a session that
  are no longer referenced elsewhere. (A utility for strong-ref
  identity maps).

- Added close() method to Transaction.  Closes out a transaction using
  rollback if it's the outermost transaction, otherwise just ends
  without affecting the outer transaction.

- Transactional and non-transactional Session integrates better with
  bound connection; a close() will ensure that connection
  transactional state is the same as that which existed on it before
  being bound to the Session.

- Modified SQL operator functions to be module-level operators,
  allowing SQL expressions to be pickleable. [ticket:735]

- Small adjustment to mapper class.__init__ to allow for Py2.6
  object.__init__() behavior.

- Fixed 'prefix' argument for select()

- Connection.begin() no longer accepts nested=True, this logic is now
  all in begin_nested().

- Fixes to new "dynamic" relation loader involving cascades

- Tickets fixed:

  - [ticket:735]
  - [ticket:752]

0.4.0beta3
----------

- SQL types optimization:

  - New performance tests show a combined mass-insert/mass-select test as
    having 68% fewer function calls than the same test run against 0.3.

  - General performance improvement of result set iteration is around 10-20%.

  - In types.AbstractType, convert_bind_param() and convert_result_value()
    have migrated to callable-returning bind_processor() and
    result_processor() methods.  If no callable is returned, no pre/post
    processing function is called.

  - Hooks added throughout base/sql/defaults to optimize the calling of bind
    aram/result processors so that method call overhead is minimized.

  - Support added for executemany() scenarios such that unneeded "last row id"
    logic doesn't kick in, parameters aren't excessively traversed.

- Added 'inherit_foreign_keys' arg to mapper().

- Added support for string date passthrough in sqlite.

- Tickets fixed:

  - [ticket:738]
  - [ticket:739]
  - [ticket:743]
  - [ticket:744]

0.4.0beta2
----------

- mssql improvements.

- oracle improvements.

- Auto-commit after LOAD DATA INFILE for mysql.

- A rudimental SessionExtension class has been added, allowing user-defined
  functionality to take place at flush(), commit(), and rollback() boundaries.

- Added engine_from_config() function for helping to create_engine() from an
  .ini style config.

- base_mapper() becomes a plain attribute.

- session.execute() and scalar() can search for a Table with which to bind from
  using the given ClauseElement.

- Session automatically extrapolates tables from mappers with binds, also uses
  base_mapper so that inheritance hierarchies bind automatically.

- Moved ClauseVisitor traversal back to inlined non-recursive.

- Tickets fixed:

  - [ticket:730]
  - [ticket:732]
  - [ticket:733]
  - [ticket:734]

0.4.0beta1
----------

- orm

  - Speed! Along with recent speedups to ResultProxy, total number of function
    calls significantly reduced for large loads.

  - test/perf/masseagerload.py reports 0.4 as having the fewest number of
    function calls across all SA versions (0.1, 0.2, and 0.3).

  - New collection_class api and implementation [ticket:213]. Collections are
    now instrumented via decorations rather than proxying.  You can now have
    collections that manage their own membership, and your class instance will
    be directly exposed on the relation property.  The changes are transparent
    for most users.

    - InstrumentedList (as it was) is removed, and relation properties no
      longer have 'clear()', '.data', or any other added methods beyond those
      provided by the collection type. You are free, of course, to add them to
      a custom class.

    - __setitem__-like assignments now fire remove events for the existing
      value, if any.

    - dict-likes used as collection classes no longer need to change __iter__
      semantics- itervalues() is used by default instead. This is a backwards
      incompatible change.

    - Subclassing dict for a mapped collection is no longer needed in most
      cases. orm.collections provides canned implementations that key objects
      by a specified column or a custom function of your choice.

    - Collection assignment now requires a compatible type- assigning None to
      clear a collection or assigning a list to a dict collection will now
      raise an argument error.

    - AttributeExtension moved to interfaces, and .delete is now .remove The
      event method signature has also been swapped around.

  - Major overhaul for Query:

    - All selectXXX methods are deprecated.  Generative methods are now the
      standard way to do things, i.e. filter(), filter_by(), all(), one(),
      etc.  Deprecated methods are docstring'ed with their new replacements.

    - Class-level properties are now usable as query elements... no more
      '.c.'!  "Class.c.propname" is now superceded by "Class.propname".  All
      clause operators are supported, as well as higher level operators such
      as Class.prop==<some instance> for scalar attributes,
      Class.prop.contains(<some instance>) and Class.prop.any(<some
      expression>) for collection-based attributes (all are also
      negatable).  Table-based column expressions as well as columns mounted
      on mapped classes via 'c' are of course still fully available and can be
      freely mixed with the new attributes.  [ticket:643]

    - Removed ancient query.select_by_attributename() capability.

    - The aliasing logic used by eager loading has been generalized, so that
      it also adds full automatic aliasing support to Query.  It's no longer
      necessary to create an explicit Alias to join to the same tables
      multiple times; *even for self-referential relationships*.

      - join() and outerjoin() take arguments "aliased=True".  Yhis causes
        their joins to be built on aliased tables; subsequent calls to
        filter() and filter_by() will translate all table expressions (yes,
        real expressions using the original mapped Table) to be that of the
        Alias for the duration of that join() (i.e. until reset_joinpoint() or
        another join() is called).

      - join() and outerjoin() take arguments "id=<somestring>".  When used
        with "aliased=True", the id can be referenced by add_entity(cls,
        id=<somestring>) so that you can select the joined instances even if
        they're from an alias.

      - join() and outerjoin() now work with self-referential relationships!
        Using "aliased=True", you can join as many levels deep as desired,
        i.e. query.join(['children', 'children'], aliased=True); filter
        criterion will be against the rightmost joined table

    - Added query.populate_existing(), marks the query to reload all
      attributes and collections of all instances touched in the query,
      including eagerly-loaded entities.  [ticket:660]

    - Added eagerload_all(), allows eagerload_all('x.y.z') to specify eager
      loading of all properties in the given path.

  - Major overhaul for Session:

    - New function which "configures" a session called "sessionmaker()".  Send
      various keyword arguments to this function once, returns a new class
      which creates a Session against that stereotype.

    - SessionTransaction removed from "public" API.  You now can call begin()/
      commit()/rollback() on the Session itself.

    - Session also supports SAVEPOINT transactions; call begin_nested().

    - Session supports two-phase commit behavior when vertically or
      horizontally partitioning (i.e., using more than one engine).  Use
      twophase=True.

    - Session flag "transactional=True" produces a session which always places
      itself into a transaction when first used.  Upon commit(), rollback() or
      close(), the transaction ends; but begins again on the next usage.

    - Session supports "autoflush=True".  This issues a flush() before each
      query.  Use in conjunction with transactional, and you can just
      save()/update() and then query, the new objects will be there.  Use
      commit() at the end (or flush() if non-transactional) to flush remaining
      changes.

    - New scoped_session() function replaces SessionContext and assignmapper.
      Builds onto "sessionmaker()" concept to produce a class whos Session()
      construction returns the thread-local session.  Or, call all Session
      methods as class methods, i.e. Session.save(foo); Session.commit().
      just like the old "objectstore" days.

    - Added new "binds" argument to Session to support configuration of
      multiple binds with sessionmaker() function.

    - A rudimental SessionExtension class has been added, allowing
      user-defined functionality to take place at flush(), commit(), and
      rollback() boundaries.

  - Query-based relation()s available with dynamic_loader().  This is a
    *writable* collection (supporting append() and remove()) which is also a
    live Query object when accessed for reads.  Ideal for dealing with very
    large collections where only partial loading is desired.

  - flush()-embedded inline INSERT/UPDATE expressions.  Assign any SQL
    expression, like "sometable.c.column + 1", to an instance's attribute.
    Upon flush(), the mapper detects the expression and embeds it directly in
    the INSERT or UPDATE statement; the attribute gets deferred on the
    instance so it loads the new value the next time you access it.

  - A rudimental sharding (horizontal scaling) system is introduced.  This
    system uses a modified Session which can distribute read and write
    operations among multiple databases, based on user-defined functions
    defining the "sharding strategy".  Instances and their dependents can be
    distributed and queried among multiple databases based on attribute
    values, round-robin approaches or any other user-defined
    system. [ticket:618]

  - Eager loading has been enhanced to allow even more joins in more places.
    It now functions at any arbitrary depth along self-referential and
    cyclical structures.  When loading cyclical structures, specify
    "join_depth" on relation() indicating how many times you'd like the table
    to join to itself; each level gets a distinct table alias.  The alias
    names themselves are generated at compile time using a simple counting
    scheme now and are a lot easier on the eyes, as well as of course
    completely deterministic. [ticket:659]

  - Added composite column properties.  This allows you to create a type which
    is represented by more than one column, when using the ORM.  Objects of
    the new type are fully functional in query expressions, comparisons,
    query.get() clauses, etc. and act as though they are regular single-column
    scalars... except they're not!  Use the function composite(cls, *columns)
    inside of the mapper's "properties" dict, and instances of cls will be
    created/mapped to a single attribute, comprised of the values correponding
    to *columns. [ticket:211]

  - Improved support for custom column_property() attributes which feature
    correlated subqueries, works better with eager loading now.

  - Primary key "collapse" behavior; the mapper will analyze all columns in
    its given selectable for primary key "equivalence", that is, columns which
    are equivalent via foreign key relationship or via an explicit
    inherit_condition. primarily for joined-table inheritance scenarios where
    different named PK columns in inheriting tables should "collapse" into a
    single-valued (or fewer-valued) primary key.  Fixes things like
    [ticket:611].

  - Joined-table inheritance will now generate the primary key columns of all
    inherited classes against the root table of the join only.  This implies
    that each row in the root table is distinct to a single instance.  If for
    some rare reason this is not desireable, explicit primary_key settings on
    individual mappers will override it.

  - When "polymorphic" flags are used with joined-table or single-table
    inheritance, all identity keys are generated against the root class of the
    inheritance hierarchy; this allows query.get() to work polymorphically
    using the same caching semantics as a non-polymorphic get.  Note that this
    currently does not work with concrete inheritance.

  - Secondary inheritance loading: polymorphic mappers can be constructed
    *without* a select_table argument. inheriting mappers whose tables were
    not represented in the initial load will issue a second SQL query
    immediately, once per instance (i.e. not very efficient for large lists),
    in order to load the remaining columns.

  - Secondary inheritance loading can also move its second query into a
    column-level "deferred" load, via the "polymorphic_fetch" argument, which
    can be set to 'select' or 'deferred'

  - It's now possible to map only a subset of available selectable columns
    onto mapper properties, using include_columns/exclude_columns.
    [ticket:696].

  - Added undefer_group() MapperOption, sets a set of "deferred" columns
    joined by a "group" to load as "undeferred".

  - Rewrite of the "deterministic alias name" logic to be part of the SQL
    layer, produces much simpler alias and label names more in the style of
    Hibernate

- sql

  - Speed!  Clause compilation as well as the mechanics of SQL constructs have
    been streamlined and simplified to a signficant degree, for a 20-30%
    improvement of the statement construction/compilation overhead of 0.3.

  - All "type" keyword arguments, such as those to bindparam(), column(),
    Column(), and func.<something>(), renamed to "type_".  Those objects still
    name their "type" attribute as "type".

  - case_sensitive=(True|False) setting removed from schema items, since
    checking this state added a lot of method call overhead and there was no
    decent reason to ever set it to False.  Table and column names which are
    all lower case will be treated as case-insenstive (yes we adjust for
    Oracle's UPPERCASE style too).

  - Transactions:

    - Added context manager (with statement) support for transactions.
    - Added support for two phase commit, works with mysql and postgres so far.
    - Added a subtransaction implementation that uses savepoints.
    - Added support for savepoints.

  - MetaData:

    - Tables can be reflected from the database en-masse without declaring
      them in advance.  MetaData(engine, reflect=True) will load all tables
      present in the database, or use metadata.reflect() for finer control.
    - DynamicMetaData has been renamed to ThreadLocalMetaData
    - The ThreadLocalMetaData constructor now takes no arguments.
    - BoundMetaData has been removed- regular MetaData is equivalent

  - Numeric and Float types now have an "asdecimal" flag; defaults to True for
    Numeric, False for Float.  When True, values are returned as
    decimal.Decimal objects; when False, values are returned as float().  The
    defaults of True/False are already the behavior for PG and MySQL's DBAPI
    modules. [ticket:646]

  - New SQL operator implementation which removes all hardcoded operators from
    expression structures and moves them into compilation; allows greater
    flexibility of operator compilation; for example, "+" compiles to "||"
    when used in a string context, or "concat(a,b)" on MySQL; whereas in a
    numeric context it compiles to "+".  Fixes [ticket:475].

  - "Anonymous" alias and label names are now generated at SQL compilation
    time in a completely deterministic fashion... no more random hex IDs

  - Significant architectural overhaul to SQL elements (ClauseElement).  All
    elements share a common "mutability" framework which allows a consistent
    approach to in-place modifications of elements as well as generative
    behavior.  Improves stability of the ORM which makes heavy usage of
    mutations to SQL expressions.

  - select() and union()'s now have "generative" behavior.  Methods like
    order_by() and group_by() return a *new* instance - the original instance
    is left unchanged.  Non-generative methods remain as well.

  - The internals of select/union vastly simplified- all decision making
    regarding "is subquery" and "correlation" pushed to SQL generation phase.
    select() elements are now *never* mutated by their enclosing containers or
    by any dialect's compilation process [ticket:52] [ticket:569]

  - select(scalar=True) argument is deprecated; use select(..).as_scalar().
    The resulting object obeys the full "column" interface and plays better
    within expressions.

  - Added select().with_prefix('foo') allowing any set of keywords to be
    placed before the columns clause of the SELECT [ticket:504]

  - Added array slice support to row[<index>] [ticket:686]

  - Result sets make a better attempt at matching the DBAPI types present in
    cursor.description to the TypeEngine objects defined by the dialect, which
    are then used for result-processing. Note this only takes effect for
    textual SQL; constructed SQL statements always have an explicit type map.

  - Result sets from CRUD operations close their underlying cursor immediately
    and will also autoclose the connection if defined for the operation; this
    allows more efficient usage of connections for successive CRUD operations
    with less chance of "dangling connections".

  - Column defaults and onupdate Python functions (i.e. passed to
    ColumnDefault) may take zero or one arguments; the one argument is the
    ExecutionContext, from which you can call "context.parameters[someparam]"
    to access the other bind parameter values affixed to the statement
    [ticket:559].  The connection used for the execution is available as well
    so that you can pre-execute statements.

  - Added "explcit" create/drop/execute support for sequences (i.e. you can
    pass a "connectable" to each of those methods on Sequence).

  - Better quoting of identifiers when manipulating schemas.

  - Standardized the behavior for table reflection where types can't be
    located; NullType is substituted instead, warning is raised.

  - ColumnCollection (i.e. the 'c' attribute on tables) follows dictionary
    semantics for "__contains__" [ticket:606]

- engines

  - Speed! The mechanics of result processing and bind parameter processing
    have been overhauled, streamlined and optimized to issue as little method
    calls as possible.  Bench tests for mass INSERT and mass rowset iteration
    both show 0.4 to be over twice as fast as 0.3, using 68% fewer function
    calls.

  - You can now hook into the pool lifecycle and run SQL statements or other
    logic at new each DBAPI connection, pool check-out and check-in.

  - Connections gain a .properties collection, with contents scoped to the
    lifetime of the underlying DBAPI connection

  - Removed auto_close_cursors and disallow_open_cursors arguments from Pool;
    reduces overhead as cursors are normally closed by ResultProxy and
    Connection.

- extensions

  - proxyengine is temporarily removed, pending an actually working
    replacement.

  - SelectResults has been replaced by Query.  SelectResults /
    SelectResultsExt still exist but just return a slightly modified Query
    object for backwards-compatibility.  join_to() method from SelectResults
    isn't present anymore, need to use join().

- mysql

  - Table and column names loaded via reflection are now Unicode.

  - All standard column types are now supported, including SET.

  - Table reflection can now be performed in as little as one round-trip.

  - ANSI and ANSI_QUOTES sql modes are now supported.

  - Indexes are now reflected.

- postgres

  - Added PGArray datatype for using postgres array datatypes.

- oracle

  - Very rudimental support for OUT parameters added; use sql.outparam(name,
    type) to set up an OUT parameter, just like bindparam(); after execution,
    values are avaiable via result.out_parameters dictionary. [ticket:507]

0.3.11
------

- sql

    - tweak DISTINCT precedence for clauses like
      `func.count(t.c.col.distinct())`

    - Fixed detection of internal '$' characters in :bind$params [ticket:719]

    - [ticket:768] dont assume join criterion consists only of column objects

    - adjusted operator precedence of NOT to match '==' and others, so that
      ~(x==y) produces NOT (x=y), which is compatible with MySQL < 5.0
      (doesn't like "NOT x=y") [ticket:764]

- orm

    - added a check for joining from A->B using join(), along two
      different m2m tables.  this raises an error in 0.3 but is
      possible in 0.4 when aliases are used. [ticket:687]

    - fixed small exception throw bug in Session.merge()

    - fixed bug where mapper, being linked to a join where one table had
      no PK columns, would not detect that the joined table had no PK.

    - fixed bugs in determining proper sync clauses from custom inherit
      conditions [ticket:769]

    - backref remove object operation doesn't fail if the other-side
      collection doesn't contain the item, supports noload collections
      [ticket:813]

- engine

    - fixed another occasional race condition which could occur
      when using pool with threadlocal setting

- mysql
    - fixed specification of YEAR columns when generating schema

- mssql

    - added support for TIME columns (simulated using DATETIME) [ticket:679]

    - added support for BIGINT, MONEY, SMALLMONEY, UNIQUEIDENTIFIER and
      SQL_VARIANT [ticket:721]

    - index names are now quoted when dropping from reflected tables
      [ticket:684]

    - can now specify a DSN for PyODBC, using a URI like mssql:///?dsn=bob

- postgres

    - when reflecting tables from alternate schemas, the "default" placed upon
      the primary key, i.e. usually a sequence name, has the "schema" name
      unconditionally quoted, so that schema names which need quoting are fine.
      its slightly unnecessary for schema names which don't need quoting
      but not harmful.

- sqlite
   - passthrough for stringified dates

- firebird
    - supports_sane_rowcount() set to False due to ticket #370 (right way).
    - fixed reflection of Column's nullable property.

- oracle
    - removed LONG_STRING, LONG_BINARY from "binary" types, so type objects
      don't try to read their values as LOB [ticket:622], [ticket:751]

0.3.10
- general
    - a new mutex that was added in 0.3.9 causes the pool_timeout
      feature to fail during a race condition; threads would
      raise TimeoutError immediately with no delay if many threads
      push the pool into overflow at the same time.  this issue has been
      fixed.
- sql
  - got connection-bound metadata to work with implicit execution
  - foreign key specs can have any chararcter in their identifiers
   [ticket:667]
  - added commutativity-awareness to binary clause comparisons to
    each other, improves ORM lazy load optimization [ticket:664]
- orm
  - cleanup to connection-bound sessions, SessionTransaction
- postgres
  - fixed max identifier length (63) [ticket:571]


0.3.9
- general
    - better error message for NoSuchColumnError [ticket:607]
    - finally figured out how to get setuptools version in, available
      as sqlalchemy.__version__ [ticket:428]
    - the various "engine" arguments, such as "engine", "connectable",
      "engine_or_url", "bind_to", etc. are all present, but deprecated.
      they all get replaced by the single term "bind".  you also
      set the "bind" of MetaData using
      metadata.bind = <engine or connection>
- ext
    - iteration over dict association proxies is now dict-like, not
      InstrumentedList-like (e.g. over keys instead of values)
    - association proxies no longer bind tightly to source collections
      [ticket:597], and are constructed with a thunk instead
    - added selectone_by() to assignmapper
- orm
    - forwards-compatibility with 0.4: added one(), first(), and
      all() to Query.  almost all Query functionality from 0.4 is
      present in 0.3.9 for forwards-compat purposes.
    - reset_joinpoint() really really works this time, promise ! lets
      you re-join from the root:
      query.join(['a', 'b']).filter(<crit>).reset_joinpoint().\
      join(['a', 'c']).filter(<some other crit>).all()
      in 0.4 all join() calls start from the "root"
    - added synchronization to the mapper() construction step, to avoid
      thread collisions when pre-existing mappers are compiling in a
      different thread [ticket:613]
    - a warning is issued by Mapper when two primary key columns of the
      same name are munged into a single attribute.  this happens frequently
      when mapping to joins (or inheritance).
    - synonym() properties are fully supported by all Query joining/
      with_parent operations [ticket:598]
    - fixed very stupid bug when deleting items with many-to-many
      uselist=False relations
    - remember all that stuff about polymorphic_union ?  for
      joined table inheritance ?  Funny thing...
      You sort of don't need it for joined table inheritance, you
      can just string all the tables together via outerjoin().
      The UNION still applies if concrete tables are involved,
      though (since nothing to join them on).
    - small fix to eager loading to better work with eager loads
      to polymorphic mappers that are using a straight "outerjoin"
      clause
- sql
    - ForeignKey to a table in a schema thats not the default schema
      requires the schema to be explicit; i.e. ForeignKey('alt_schema.users.id')
    - MetaData can now be constructed with an engine or url as the first
      argument, just like BoundMetaData
    - BoundMetaData is now deprecated, and MetaData is a direct substitute.
    - DynamicMetaData has been renamed to ThreadLocalMetaData.  the
      DynamicMetaData name is deprecated and is an alias for ThreadLocalMetaData
      or a regular MetaData if threadlocal=False
    - composite primary key is represented as a non-keyed set to allow for
      composite keys consisting of cols with the same name; occurs within a
      Join.  helps inheritance scenarios formulate correct PK.
    - improved ability to get the "correct" and most minimal set of primary key
      columns from a join, equating foreign keys and otherwise equated columns.
      this is also mostly to help inheritance scenarios formulate the best
      choice of primary key columns.  [ticket:185]
    - added 'bind' argument to Sequence.create()/drop(), ColumnDefault.execute()
    - columns can be overridden in a reflected table with a "key"
      attribute different than the column's name, including for primary key
      columns [ticket:650]
    - fixed "ambiguous column" result detection, when dupe col names exist
      in a result [ticket:657]
    - some enhancements to "column targeting", the ability to match a column
      to a "corresponding" column in another selectable.  this affects mostly
      ORM ability to map to complex joins
    - MetaData and all SchemaItems are safe to use with pickle.  slow
      table reflections can be dumped into a pickled file to be reused later.
      Just reconnect the engine to the metadata after unpickling. [ticket:619]
    - added a mutex to QueuePool's "overflow" calculation to prevent a race
      condition that can bypass max_overflow
    - fixed grouping of compound selects to give correct results. will break
      on sqlite in some cases, but those cases were producing incorrect
      results anyway, sqlite doesn't support grouped compound selects
      [ticket:623]
    - fixed precedence of operators so that parenthesis are correctly applied
      [ticket:620]
    - calling <column>.in_() (i.e. with no arguments) will return
      "CASE WHEN (<column> IS NULL) THEN NULL ELSE 0 END = 1)", so that
      NULL or False is returned in all cases, rather than throwing an error
      [ticket:545]
    - fixed "where"/"from" criterion of select() to accept a unicode string
      in addition to regular string - both convert to text()
    - added standalone distinct() function in addition to column.distinct()
      [ticket:558]
    - result.last_inserted_ids() should return a list that is identically
      sized to the primary key constraint of the table.  values that were
      "passively" created and not available via cursor.lastrowid will be None.
    - long-identifier detection fixed to use > rather than >= for
      max ident length [ticket:589]
    - fixed bug where selectable.corresponding_column(selectable.c.col)
      would not return selectable.c.col, if the selectable is a join
      of a table and another join involving the same table.  messed
      up ORM decision making [ticket:593]
    - added Interval type to types.py [ticket:595]
- mysql
    - fixed catching of some errors that imply a dropped connection [ticket:625]
    - fixed escaping of the modulo operator [ticket:624]
    - added 'fields' to reserved words [ticket:590]
    - various reflection enhancement/fixes
- oracle
    - datetime fixes: got subsecond TIMESTAMP to work [ticket:604],
      added OracleDate which supports types.Date with only year/month/day
    - added dialect flag "auto_convert_lobs", defaults to True; will cause any
      LOB objects detected in a result set to be forced into OracleBinary
      so that the LOB is read() automatically, if no typemap was present
      (i.e., if a textual execute() was issued).
    - mod operator '%' produces MOD [ticket:624]
    - converts cx_oracle datetime objects to Python datetime.datetime when
      Python 2.3 used [ticket:542]
    - fixed unicode conversion in Oracle TEXT type
- postgres
    - fixed escaping of the modulo operator [ticket:624]
    - added support for reflection of domains [ticket:570]
    - types which are missing during reflection resolve to Null type
      instead of raising an error
    - the fix in "schema" above fixes reflection of foreign keys from an
      alt-schema table to a public schema table
- sqlite
    - rearranged dialect initialization so it has time to warn about pysqlite1
      being too old.
    - sqlite better handles datetime/date/time objects mixed and matched
      with various Date/Time/DateTime columns
    - string PK column inserts dont get overwritten with OID [ticket:603]
- mssql
    - fix port option handling for pyodbc [ticket:634]
    - now able to reflect start and increment values for identity columns
    - preliminary support for using scope_identity() with pyodbc

0.3.8
- engines
  - added detach() to Connection, allows underlying DBAPI connection
    to be detached from its pool, closing on dereference/close()
    instead of being reused by the pool.
  - added invalidate() to Connection, immediately invalidates the
    Connection and its underlying DBAPI connection.
- sql
  - _Label class overrides compare_self to return its ultimate
    object. meaning, if you say someexpr.label('foo') == 5, it
    produces the correct "someexpr == 5".
  - _Label propagates "_hide_froms()" so that scalar selects
    behave more properly with regards to FROM clause #574
  - fix to long name generation when using oid_column as an order by
    (oids used heavily in mapper queries)
  - significant speed improvement to ResultProxy, pre-caches
    TypeEngine dialect implementations and saves on function calls
    per column
  - parenthesis are applied to clauses via a new _Grouping
    construct. uses operator precedence to more intelligently apply
    parenthesis to clauses, provides cleaner nesting of clauses
    (doesnt mutate clauses placed in other clauses, i.e. no 'parens'
    flag)
  - added 'modifier' keyword, works like func.<foo> except does not
    add parenthesis.  e.g. select([modifier.DISTINCT(...)]) etc.
  - removed "no group by's in a select thats part of a UNION"
    restriction [ticket:578]
- orm
  - added reset_joinpoint() method to Query, moves the "join point"
    back to the starting mapper. 0.4 will change the behavior of
    join() to reset the "join point" in all cases so this is an
    interim method. for forwards compatibility, ensure joins across
    multiple relations are specified using a single join(), i.e.
    join(['a', 'b', 'c']).
  - fixed bug in query.instances() that wouldnt handle more than
    on additional mapper or one additional column.
  - "delete-orphan" no longer implies "delete". ongoing effort to
    separate the behavior of these two operations.
  - many-to-many relationships properly set the type of bind params
    for delete operations on the association table
  - many-to-many relationships check that the number of rows deleted
    from the association table by a delete operation matches the
    expected results
  - session.get() and session.load() propagate **kwargs through to
    query
  - fix to polymorphic query which allows the original
    polymorphic_union to be embedded into a correlated subquery
    [ticket:577]
  - fix to select_by(<propname>=<object instance>) -style joins in
    conjunction with many-to-many relationships, bug introduced in
    r2556
  - the "primary_key" argument to mapper() is propagated to the
    "polymorphic" mapper. primary key columns in this list get
    normalized to that of the mapper's local table.
  - restored logging of "lazy loading clause" under
    sa.orm.strategies logger, got removed in 0.3.7
  - improved support for eagerloading of properties off of mappers
    that are mapped to select() statements; i.e. eagerloader is
    better at locating the correct selectable with which to attach
    its LEFT OUTER JOIN.
- mysql
  - Nearly all MySQL column types are now supported for declaration
    and reflection. Added NCHAR, NVARCHAR, VARBINARY, TINYBLOB,
    LONGBLOB, YEAR
  - The sqltypes.Binary passthrough now always builds a BLOB,
    avoiding problems with very old database versions
  - support for column-level CHARACTER SET and COLLATE declarations,
    as well as ASCII, UNICODE, NATIONAL and BINARY shorthand.
- firebird
  - set max identifier length to 31
  - supports_sane_rowcount() set to False due to ticket #370.
    versioned_id_col feature wont work in FB.
  - some execution fixes
-extensions
  - new association proxy implementation, implementing complete
    proxies to list, dict and set-based relation collections
  - added orderinglist, a custom list class that synchronizes an
    object attribute with that object's position in the list
  - small fix to SelectResultsExt to not bypass itself during
    select().
  - added filter(), filter_by() to assignmapper

0.3.7
- engines
    - warnings module used for issuing warnings (instead of logging)
    - cleanup of DBAPI import strategies across all engines
      [ticket:480]
    - refactoring of engine internals which reduces complexity,
      number of codepaths; places more state inside of ExecutionContext
      to allow more dialect control of cursor handling, result sets.
      ResultProxy totally refactored and also has two versions of
      "buffered" result sets used for different purposes.
    - server side cursor support fully functional in postgres
      [ticket:514].
    - improved framework for auto-invalidation of connections that have
      lost their underlying database, via dialect-specific detection
      of exceptions corresponding to that database's disconnect
      related error messages.  Additionally, when a "connection no
      longer open" condition is detected, the entire connection pool
      is discarded and replaced with a new instance.  #516
    - the dialects within sqlalchemy.databases become a setuptools
      entry points. loading the built-in database dialects works the
      same as always, but if none found will fall back to trying
      pkg_resources to load an external module [ticket:521]
    - Engine contains a "url" attribute referencing the url.URL object
      used by create_engine().
- sql:
    - keys() of result set columns are not lowercased, come back
      exactly as they're expressed in cursor.description.  note this
      causes colnames to be all caps in oracle.
    - preliminary support for unicode table names, column names and
      SQL statements added, for databases which can support them.
      Works with sqlite and postgres so far.  Mysql *mostly* works
      except the has_table() function does not work.  Reflection
      works too.
    - the Unicode type is now a direct subclass of String, which now
      contains all the "convert_unicode" logic.  This helps the variety
      of unicode situations that occur in db's such as MS-SQL to be
      better handled and allows subclassing of the Unicode datatype.
      [ticket:522]
    - ClauseElements can be used in in_() clauses now, such as bind
      parameters, etc. #476
    - reverse operators implemented for `CompareMixin` elements,
      allows expressions like "5 + somecolumn" etc. #474
    - the "where" criterion of an update() and delete() now correlates
      embedded select() statements against the table being updated or
      deleted.  this works the same as nested select() statement
      correlation, and can be disabled via the correlate=False flag on
      the embedded select().
    - column labels are now generated in the compilation phase, which
      means their lengths are dialect-dependent.  So on oracle a label
      that gets truncated to 30 chars will go out to 63 characters
      on postgres.  Also, the true labelname is always attached as the
      accessor on the parent Selectable so theres no need to be aware
      of the "truncated" label names [ticket:512].
    - column label and bind param "truncation" also generate
      deterministic names now, based on their ordering within the
      full statement being compiled.  this means the same statement
      will produce the same string across application restarts and
      allowing DB query plan caching to work better.
    - the "mini" column labels generated when using subqueries, which
      are to work around glitchy SQLite behavior that doesnt understand
      "foo.id" as equivalent to "id", are now only generated in the case
      that those named columns are selected from (part of [ticket:513])
    - the label() method on ColumnElement will properly propagate the
      TypeEngine of the base element out to the label, including a label()
      created from a scalar=True select() statement.
    - MS-SQL better detects when a query is a subquery and knows not to
      generate ORDER BY phrases for those [ticket:513]
    - fix for fetchmany() "size" argument being positional in most
      dbapis [ticket:505]
    - sending None as an argument to func.<something> will produce
      an argument of NULL
    - query strings in unicode URLs get keys encoded to ascii
      for **kwargs compat
    - slight tweak to raw execute() change to also support tuples
      for positional parameters, not just lists [ticket:523]
    - fix to case() construct to propagate the type of the first
      WHEN condition as the return type of the case statement
- orm:
    - fixed critical issue when, after options(eagerload()) is used,
      the mapper would then always apply query "wrapping" behavior
      for all subsequent LIMIT/OFFSET/DISTINCT queries, even if no
      eager loading was applied on those subsequent queries.
    - added query.with_parent(someinstance) method.  searches for
      target instance using lazy join criterion from parent instance.
      takes optional string "property" to isolate the desired relation.
      also adds static Query.query_from_parent(instance, property)
      version. [ticket:541]
    - improved query.XXX_by(someprop=someinstance) querying to use
      similar methodology to with_parent, i.e. using the "lazy" clause
      which prevents adding the remote instance's table to the SQL,
      thereby making more complex conditions possible [ticket:554]
    - added generative versions of aggregates, i.e. sum(), avg(), etc.
      to query. used via query.apply_max(), apply_sum(), etc.
      #552
    - fix to using distinct() or distinct=True in combination with
      join() and similar
    - corresponding to label/bindparam name generation, eager loaders
      generate deterministic names for the aliases they create using
      md5 hashes.
    - improved/fixed custom collection classes when giving it "set"/
      "sets.Set" classes or subclasses (was still looking for append()
      methods on them during lazy loads)
    - restored old "column_property()" ORM function (used to be called
      "column()") to force any column expression to be added as a property
      on a mapper, particularly those that aren't present in the mapped
      selectable.  this allows "scalar expressions" of any kind to be
      added as relations (though they have issues with eager loads).
    - fix to many-to-many relationships targeting polymorphic mappers
      [ticket:533]
    - making progress with session.merge() as well as combining its
      usage with entity_name [ticket:543]
    - the usual adjustments to relationships between inheriting mappers,
      in this case establishing relation()s to subclass mappers where
      the join conditions come from the superclass' table
- informix:
    - informix support added !  courtesy James Zhang, who put a ton
      of effort in.
- sqlite:
    - removed silly behavior where sqlite would reflect UNIQUE indexes
      as part of the primary key (?!)
- oracle:
    - small fix to allow successive compiles of the same SELECT object
      which features LIMIT/OFFSET.  oracle dialect needs to modify
      the object to have ROW_NUMBER OVER and wasn't performing
      the full series of steps on successive compiles.
- mysql
    - support for SSL arguments given as inline within URL query string,
      prefixed with "ssl_", courtesy terjeros@gmail.com.
    - mysql uses "DESCRIBE [<schemaname>].<tablename>", catching exceptions
      if table doesnt exist, in order to determine if a table exists.
      this supports unicode table names as well as schema names. tested
      with MySQL5 but should work with 4.1 series as well. (#557)
- extensions
    - big fix to AssociationProxy so that multiple AssociationProxy
      objects can be associated with a single association collection.
    - assign_mapper names methods according to their keys (i.e. __name__)
      #551
- mssql
    - pyodbc is now the preferred DB-API for MSSQL, and if no module is
      specifically requested, will be loaded first on a module probe.

    - The @@SCOPE_IDENTITY is now used instead of @@IDENTITY. This
      behavior may be overridden with the engine_connect
      "use_scope_identity" keyword parameter, which may also be specified
      in the dburi.



0.3.6
- sql:
    - bindparam() names are now repeatable!  specify two
      distinct bindparam()s with the same name in a single statement,
      and the key will be shared.  proper positional/named args translate
      at compile time.  for the old behavior of "aliasing" bind parameters
      with conflicting names, specify "unique=True" - this option is
      still used internally for all the auto-genererated (value-based)
      bind parameters.

    - slightly better support for bind params as column clauses, either
      via bindparam() or via literal(), i.e. select([literal('foo')])

    - MetaData can bind to an engine either via "url" or "engine" kwargs
      to constructor, or by using connect() method. BoundMetaData is
      identical to MetaData except engine_or_url param is required.
      DynamicMetaData is the same and provides thread-local connections be
      default.

    - exists() becomes useable as a standalone selectable, not just in a
      WHERE clause, i.e. exists([columns], criterion).select()

    - correlated subqueries work inside of ORDER BY, GROUP BY

    - fixed function execution with explicit connections, i.e.
      conn.execute(func.dosomething())

    - use_labels flag on select() wont auto-create labels for literal text
      column elements, since we can make no assumptions about the text. to
      create labels for literal columns, you can say "somecol AS
      somelabel", or use literal_column("somecol").label("somelabel")

    - quoting wont occur for literal columns when they are "proxied" into
      the column collection for their selectable (is_literal flag is
      propagated). literal columns are specified via
      literal_column("somestring").

    - added "fold_equivalents" boolean argument to Join.select(), which
      removes 'duplicate' columns from the resulting column clause that
      are known to be equivalent based on the join condition. this is of
      great usage when constructing subqueries of joins which Postgres
      complains about if duplicate column names are present.

    - fixed use_alter flag on ForeignKeyConstraint [ticket:503]

    - fixed usage of 2.4-only "reversed" in topological.py [ticket:506]

    - for hackers, refactored the "visitor" system of ClauseElement and
      SchemaItem so that the traversal of items is controlled by the
      ClauseVisitor itself, using the method visitor.traverse(item).
      accept_visitor() methods can still be called directly but will not
      do any traversal of child items. ClauseElement/SchemaItem now have a
      configurable get_children() method to return the collection of child
      elements for each parent object. This allows the full traversal of
      items to be clear and unambiguous (as well as loggable), with an
      easy method of limiting a traversal (just pass flags which are
      picked up by appropriate get_children() methods). [ticket:501]

    - the "else_" parameter to the case statement now properly works when
      set to zero.

- orm:
    - the full featureset of the SelectResults extension has been merged
      into a new set of methods available off of Query.  These methods
      all provide "generative" behavior, whereby the Query is copied
      and a new one returned with additional criterion added.
      The new methods include:

          filter() - applies select criterion to the query
          filter_by() - applies "by"-style criterion to the query
          avg() - return the avg() function on the given column
          join() - join to a property (or across a list of properties)
          outerjoin() - like join() but uses LEFT OUTER JOIN
          limit()/offset() - apply LIMIT/OFFSET
          range-based access which applies limit/offset:
             session.query(Foo)[3:5]
          distinct() - apply DISTINCT
          list() - evaluate the criterion and return results

      no incompatible changes have been made to Query's API and no methods
      have been deprecated.  Existing methods like select(), select_by(),
      get(), get_by() all execute the query at once and return results
      like they always did.  join_to()/join_via() are still there although
      the generative join()/outerjoin() methods are easier to use.

    - the return value for multiple mappers used with instances() now
      returns a cartesian product of the requested list of mappers,
      represented as a list of tuples. this corresponds to the documented
      behavior. So that instances match up properly, the "uniquing" is
      disabled when this feature is used.

    - Query has add_entity() and add_column() generative methods. these
      will add the given mapper/class or ColumnElement to the query at
      compile time, and apply them to the instances() method. the user is
      responsible for constructing reasonable join conditions (otherwise
      you can get full cartesian products). result set is the list of
      tuples, non-uniqued.

    - strings and columns can also be sent to the *args of instances()
      where those exact result columns will be part of the result tuples.

    - a full select() construct can be passed to query.select() (which
      worked anyway), but also query.selectfirst(), query.selectone()
      which will be used as is (i.e. no query is compiled). works
      similarly to sending the results to instances().

    - eager loading will not "aliasize" "order by" clauses that were
      placed in the select statement by something other than the eager
      loader itself, to fix possibility of dupe columns as illustrated in
      [ticket:495]. however, this means you have to be more careful with
      the columns placed in the "order by" of Query.select(), that you
      have explicitly named them in your criterion (i.e. you cant rely on
      the eager loader adding them in for you)

    - added a handy multi-use "identity_key()" method to Session, allowing
      the generation of identity keys for primary key values, instances,
      and rows, courtesy Daniel Miller

    - many-to-many table will be properly handled even for operations that
      occur on the "backref" side of the operation [ticket:249]

    - added "refresh-expire" cascade [ticket:492].  allows refresh() and
      expire() calls to propagate along relationships.

    - more fixes to polymorphic relations, involving proper lazy-clause
      generation on many-to-one relationships to polymorphic mappers
      [ticket:493]. also fixes to detection of "direction", more specific
      targeting of columns that belong to the polymorphic union vs. those
      that dont.

    - some fixes to relationship calcs when using "viewonly=True" to pull
      in other tables into the join condition which arent parent of the
      relationship's parent/child mappings

    - flush fixes on cyclical-referential relationships that contain
      references to other instances outside of the cyclical chain, when
      some of the objects in the cycle are not actually part of the flush

    - put an aggressive check for "flushing object A with a collection of
      B's, but you put a C in the collection" error condition - **even if
      C is a subclass of B**, unless B's mapper loads polymorphically.
      Otherwise, the collection will later load a "B" which should be a
      "C" (since its not polymorphic) which breaks in bi-directional
      relationships (i.e. C has its A, but A's backref will lazyload it as
      a different instance of type "B") [ticket:500] This check is going
      to bite some of you who do this without issues, so the error message
      will also document a flag "enable_typechecks=False" to disable this
      checking. But be aware that bi-directional relationships in
      particular become fragile without this check.

- extensions:
    - options() method on SelectResults now implemented "generatively"
      like the rest of the SelectResults methods [ticket:472].  But
      you're going to just use Query now anyway.

    - query() method is added by assignmapper.  this helps with
      navigating to all the new generative methods on Query.

- ms-sql:
    - removed seconds input on DATE column types (probably
        should remove the time altogether)

    - null values in float fields no longer raise errors

    - LIMIT with OFFSET now raises an error (MS-SQL has no OFFSET support)

    - added an facility to use the MSSQL type VARCHAR(max) instead of TEXT
      for large unsized string fields. Use the new "text_as_varchar" to
      turn it on. [ticket:509]

    - ORDER BY clauses without a LIMIT are now stripped in subqueries, as
      MS-SQL forbids this usage

    - cleanup of module importing code; specifiable DB-API module; more
      explicit ordering of module preferences. [ticket:480]

- oracle:
    - got binary working for any size input !  cx_oracle works fine,
      it was my fault as BINARY was being passed and not BLOB for
      setinputsizes (also unit tests werent even setting input sizes).

    - also fixed CLOB read/write on a separate changeset.

    - auto_setinputsizes defaults to True for Oracle, fixed cases where
      it improperly propagated bad types.

- mysql:
    - added a catchall **kwargs to MSString, to help reflection of
      obscure types (like "varchar() binary" in MS 4.0)

    - added explicit MSTimeStamp type which takes effect when using
      types.TIMESTAMP.


0.3.5
- sql:
    - the value of "case_sensitive" defaults to True now, regardless of the
      casing of the identifier, unless specifically set to False. this is
      because the object might be label'ed as something else which does
      contain mixed case, and propigating "case_sensitive=False" breaks that.
      Other fixes to quoting when using labels and "fake" column objects
    - added a "supports_execution()" method to ClauseElement, so that
      individual kinds of clauses can express if they are appropriate for
      executing...such as, you can execute a "select", but not a "Table" or a
      "Join".
    - fixed argument passing to straight textual execute() on engine,
      connection. can handle *args or a list instance for positional, **kwargs
      or a dict instance for named args, or a list of list or dicts to invoke
      executemany()
    - small fix to BoundMetaData to accept unicode or string URLs
    - fixed named PrimaryKeyConstraint generation [ticket:466] courtesy
      andrija at gmail
    - fixed generation of CHECK constraints on columns [ticket:464]
    - fixes to tometadata() operation to propagate Constraints at column and
      table level
- oracle:
    - when returning "rowid" as the ORDER BY column or in use with ROW_NUMBER
      OVER, oracle dialect checks the selectable its being applied to and will
      switch to table PK if not applicable, i.e. for a UNION. checking for
      DISTINCT, GROUP BY (other places that rowid is invalid) still a TODO.
      allows polymorphic mappings to function, [ticket:436]
    - sequences on a non-pk column will properly fire off on INSERT
    - added PrefetchingResultProxy support to pre-fetch LOB columns when they
      are known to be present, fixes [ticket:435]
    - implemented reflection of tables based on synonyms, including across
      dblinks [ticket:379]
    - issues a log warning when a related table cant be reflected due to
      certain permission errors [ticket:363]
- mysql:
    - fix to reflection on older DB's that might return array() type for
    "show variables like" statements
- postgres:
    - better reflection of sequences for alternate-schema Tables [ticket:442]
    - sequences on a non-pk column will properly fire off on INSERT
    - added PGInterval type [ticket:460], PGInet type [ticket:444]
- mssql:
    - preliminary support for pyodbc (Yay!) [ticket:419]
    - better support for NVARCHAR types added [ticket:298]
    - fix for commit logic on pymssql
    - fix for query.get() with schema [ticket:456]
    - fix for non-integer relationships [ticket:473]
    - DB-API module now selectable at run-time [ticket:419]
    - now passes many more unit tests [tickets:422, 481, 415]
    - better unittest compatibility with ANSI functions [ticket:479]
    - improved support for implicit sequence PK columns with auto-insert
      [ticket:415]
    - fix for blank password in adodbapi [ticket:371]
    - fixes to get unit tests working with pyodbc [ticket:481]
    - fix to auto_identity_insert on db-url query
    - added query_timeout to db-url query parms. currently works only for
      pymssql
    - tested with pymssql 0.8.0 (which is now LGPL)
- orm bugs:
    - another refactoring to relationship calculation. Allows more accurate
      ORM behavior with relationships from/to/between mappers, particularly
      polymorphic mappers, also their usage with Query, SelectResults. tickets
      include [ticket:439], [ticket:441], [ticket:448].
    - removed deprecated method of specifying custom collections on classes;
      you must now use the "collection_class" option. the old way was
      beginning to produce conflicts when people used assign_mapper(), which
      now patches an "options" method, in conjunction with a relationship
      named "options". (relationships take precedence over monkeypatched
      assign_mapper methods).
    - extension() query option propagates to Mapper._instance() method so that
      all loading-related methods get called [ticket:454]
    - eager relation to an inheriting mapper wont fail if no rows returned for
      the relationship.
    - eager relation loading bug fixed for eager relation on multiple
      descendant classes [ticket:486]
    - fix for very large topological sorts, courtesy ants.aasma at gmail
      [ticket:423]
    - eager loading is slightly more strict about detecting "self-referential"
      relationships, specifically between polymorphic mappers. this results in
      an "eager degrade" to lazy loading.
    - improved support for complex queries embedded into "where" criterion for
      query.select() [ticket:449]
    - mapper options like eagerload(), lazyload(), deferred(), will work for
      "synonym()" relationships [ticket:485]
    - fixed bug where cascade operations incorrectly included deleted
      collection items in the cascade [ticket:445]
    - fixed relationship deletion error when one-to-many child item is moved
      to a new parent in a single unit of work [ticket:478]
    - fixed relationship deletion error where parent/child with a single
      column as PK/FK on the child would raise a "blank out the primary key"
      error, if manually deleted or "delete" cascade without "delete-orphan"
      was used
    - fix to deferred so that load operation doesnt mistakenly occur when only
      PK col attributes are set
- orm enhancements:
    - implemented foreign_keys argument to mapper [ticket:385]. use in
      conjunction with primaryjoin/secondaryjoin arguments to specify/override
      foreign keys defined on the Table instance.
    - contains_eager('foo') automatically implies eagerload('foo')
    - added "alias" argument to contains_eager(). use it to specify the string
      name or Alias instance of an alias used in the query for the eagerly
      loaded child items. easier to use than "decorator"
    - added "contains_alias()" option for result set mapping to an alias of
      the mapped table
    - added support for py2.5 "with" statement with SessionTransaction
      [ticket:468]
- extensions:
    - added distinct() method to SelectResults. generally should only make a
      difference when using count().
    - added options() method to SelectResults, equivalent to query.options()
      [ticket:472]
    - added optional __table_opts__ dictionary to ActiveMapper, will send kw
      options to Table objects [ticket:462]
    - added selectfirst(), selectfirst_by() to assign_mapper [ticket:467]

0.3.4
- general:
  - global "insure"->"ensure" change. in US english "insure" is actually
    largely interchangeable with "ensure" (so says the dictionary), so I'm not
    completely illiterate, but its definitely sub-optimal to "ensure" which is
    non-ambiguous.
- sql:
  - added "fetchmany()" support to ResultProxy
  - added support for column "key" attribute to be useable in
    row[<key>]/row.<key>
  - changed "BooleanExpression" to subclass from "BinaryExpression", so that
    boolean expressions can also follow column-clause behaviors (i.e. label(),
    etc).
  - trailing underscores are trimmed from func.<xxx> calls, such as func.if_()
  - fix to correlation of subqueries when the column list of the select
    statement is constructed with individual calls to append_column(); this
    fixes an ORM bug whereby nested select statements were not getting
    correlated with the main select generated by the Query object.
  - another fix to subquery correlation so that a subquery which has only one
    FROM element will *not* correlate that single element, since at least one
    FROM element is required in a query.
  - default "timezone" setting is now False. this corresponds to Python's
    datetime behavior as well as Postgres' timestamp/time types (which is the
    only timezone-sensitive dialect at the moment) [ticket:414]
  - the "op()" function is now treated as an "operation", rather than a
    "comparison". the difference is, an operation produces a BinaryExpression
    from which further operations can occur whereas comparison produces the
    more restrictive BooleanExpression
  - trying to redefine a reflected primary key column as non-primary key raises
    an error
  - type system slightly modified to support TypeDecorators that can be
    overridden by the dialect (ok, thats not very clear, it allows the mssql
    tweak below to be possible)
- mssql:
  - added an NVarchar type (produces NVARCHAR), also MSUnicode which provides
    Unicode-translation for the NVarchar regardless of dialect convert_unicode
    setting.
- postgres:
  - fix to the initial checkfirst for tables to take current schema into
    account [ticket:424]
  - postgres has an optional "server_side_cursors=True" flag which will utilize
    server side cursors. these are appropriate for fetching only partial
    results and are necessary for working with very large unbounded result
    sets. While we'd like this to be the default behavior, different
    environments seem to have different results and the causes have not been
    isolated so we are leaving the feature off by default for now. Uses an
    apparently undocumented psycopg2 behavior recently discovered on the
    psycopg mailing list.
  - added "BIGSERIAL" support for postgres table with
    PGBigInteger/autoincrement
  - fixes to postgres reflection to better handle when schema names are
    present; thanks to jason (at) ncsmags.com [ticket:402]
- mysql:
  - mysql is inconsistent with what kinds of quotes it uses in foreign keys
    during a SHOW CREATE TABLE, reflection updated to accomodate for all three
    styles [ticket:420]
  - mysql table create options work on a generic passthru now, i.e. Table(...,
    mysql_engine='InnoDB', mysql_collate="latin1_german2_ci",
    mysql_auto_increment="5", mysql_<somearg>...), helps [ticket:418]
- firebird:
  - order of constraint creation puts primary key first before all other
    constraints; required for firebird, not a bad idea for others [ticket:408]
  - Firebird fix to autoload multifield foreign keys [ticket:409]
  - Firebird NUMERIC type properly handles a type without precision
    [ticket:409]
- oracle:
  - *slight* support for binary, but still need to figure out how to insert
    reasonably large values (over 4K). requires auto_setinputsizes=True sent to
    create_engine(), rows must be fully fetched individually, etc.
- orm:
  - poked the first hole in the can of worms: saying
    query.select_by(somerelationname=someinstance) will create the join of the
    primary key columns represented by "somerelationname"'s mapper to the
    actual primary key in "someinstance".
  - reworked how relations interact with "polymorphic" mappers, i.e. mappers
    that have a select_table as well as polymorphic flags. better determination
    of proper join conditions, interaction with user- defined join conditions,
    and support for self-referential polymorphic mappers.
  - related to polymorphic mapping relations, some deeper error checking when
    compiling relations, to detect an ambiguous "primaryjoin" in the case that
    both sides of the relationship have foreign key references in the primary
    join condition. also tightened down conditions used to locate "relation
    direction", associating the "foreignkey" of the relationship with the
    "primaryjoin"
  - a little bit of improvement to the concept of a "concrete" inheritance
    mapping, though that concept is not well fleshed out yet (added test case
    to support concrete mappers on top of a polymorphic base).
  - fix to "proxy=True" behavior on synonym()
  - fixed bug where delete-orphan basically didn't work with many-to-many
    relationships [ticket:427], backref presence generally hid the symptom
  - added a mutex to the mapper compilation step. ive been reluctant to add any
    kind of threading anything to SA but this is one spot that its its really
    needed since mappers are typically "global", and while their state does not
    change during normal operation, the initial compilation step does modify
    internal state significantly, and this step usually occurs not at
    module-level initialization time (unless you call compile()) but at
    first-request time
  - basic idea of "session.merge()" actually implemented.  needs more testing.
  - added "compile_mappers()" function as a shortcut to compiling all mappers
  - fix to MapperExtension create_instance so that entity_name properly
    associated with new instance
  - speed enhancements to ORM object instantiation, eager loading of rows
  - invalid options sent to 'cascade' string will raise an exception
    [ticket:406]
  - fixed bug in mapper refresh/expire whereby eager loaders didnt properly
    re-populate item lists [ticket:407]
  - fix to post_update to ensure rows are updated even for non insert/delete
    scenarios [ticket:413]
  - added an error message if you actually try to modify primary key values on
    an entity and then flush it [ticket:412]
- extensions
  - added "validate=False" argument to assign_mapper, if True will ensure that
    only mapped attributes are named [ticket:426]
  - assign_mapper gets "options", "instances" functions added (i.e.
    MyClass.instances())

0.3.3
- string-based FROM clauses fixed, i.e. select(..., from_obj=["sometext"])
- fixes to passive_deletes flag, lazy=None (noload) flag
- added example/docs for dealing with large collections
- added object_session() method to sqlalchemy namespace
- fixed QueuePool bug whereby its better able to reconnect to a database
that was not reachable (thanks to Sébastien Lelong), also fixed dispose()
method
- patch that makes MySQL rowcount work correctly! [ticket:396]
- fix to MySQL catch of 2006/2014 errors to properly re-raise OperationalError
exception

0.3.2
- major connection pool bug fixed.  fixes MySQL out of sync
errors, will also prevent transactions getting rolled back
accidentally in all DBs [ticket:387]
- major speed enhancements vs. 0.3.1, to bring speed
back to 0.2.8 levels
  - made conditional dozens of debug log calls that were
  time-intensive to generate log messages
  - fixed bug in cascade rules whereby the entire object graph
  could be unnecessarily cascaded on the save/update cascade
  - various speedups in attributes module
- identity map in Session is by default *no longer weak referencing*.
to have it be weak referencing, use create_session(weak_identity_map=True)
fixes [ticket:388]
- MySQL detects errors 2006 (server has gone away) and 2014
(commands out of sync) and invalidates the connection on which it occured.
- MySQL bool type fix: [ticket:307]
- postgres reflection fixes: [ticket:349] [ticket:382]
- added keywords for EXCEPT, INTERSECT, EXCEPT ALL, INTERSECT ALL
[ticket:247]
- assign_mapper in assignmapper extension returns the created mapper
[changeset:2110]
- added label() function to Select class, when scalar=True is used
to create a scalar subquery
i.e. "select x, y, (select max(foo) from table) AS foomax from table"
- added onupdate and ondelete keyword arguments to ForeignKey; propagate
to underlying ForeignKeyConstraint if present.  (dont propagate in the
other direction, however)
- fix to session.update() to preserve "dirty" status of incoming object
- sending a selectable to an IN via the in_() function no longer creates
a "union" out of multiple selects; only one selectable to a the in_() function
is allowed now (make a union yourself if union is needed)
- improved support for disabling save-update cascade via cascade="none" etc.
- added "remote_side" argument to relation(), used only with self-referential
mappers to force the direction of the parent/child relationship.  replaces
the usage of the "foreignkey" parameter for "switching" the direction.
"foreignkey" argument is deprecated for all uses and will eventually
be replaced by an argument dedicated to ForeignKey specification on mappers.

0.3.1
- Engine/Pool:
  - some new Pool utility classes, updated docs
  - "use_threadlocal" on Pool defaults to False (same as create_engine)
  - fixed direct execution of Compiled objects
  - create_engine() reworked to be strict about incoming **kwargs.  all keyword
arguments must be consumed by one of the dialect, connection pool, and engine
constructors, else a TypeError is thrown which describes the full set of
invalid kwargs in relation to the selected dialect/pool/engine configuration.
- Databases/Types:
  - MySQL catches exception on "describe" and reports as NoSuchTableError
  - further fixes to sqlite booleans, weren't working as defaults
  - fix to postgres sequence quoting when using schemas
- ORM:
  - the "delete" cascade will load in all child objects, if they were not
loaded already.  this can be turned off (i.e. the old behavior) by setting
passive_deletes=True on a relation().
  - adjustments to reworked eager query generation to not fail on circular
eager-loaded relationships (like backrefs)
  - fixed bug where eagerload() (nor lazyload()) option didn't properly
instruct the Query whether or not to use "nesting" when producing a
LIMIT query.
  - fixed bug in circular dependency sorting at flush time; if object A
contained a cyclical many-to-one relationship to object B, and object B
was just attached to object A, *but* object B itself wasnt changed,
the many-to-one synchronize of B's primary key attribute to A's foreign key
attribute wouldnt occur.  [ticket:360]
  - implemented from_obj argument for query.count, improves count function
on selectresults [ticket:325]
  - added an assertion within the "cascade" step of ORM relationships to check
that the class of object attached to a parent object is appropriate
(i.e. if A.items stores B objects, raise an error if a C is appended to A.items)
  - new extension sqlalchemy.ext.associationproxy, provides transparent
"association object" mappings.  new example
examples/association/proxied_association.py illustrates.
  - improvement to single table inheritance to load full hierarchies beneath
the target class
  - fix to subtle condition in topological sort where a node could appear twice,
for [ticket:362]
  - additional rework to topological sort, refactoring, for [ticket:365]
  - "delete-orphan" for a certain type can be set on more than one parent class;
the instance is an "orphan" only if its not attached to *any* of those parents

0.3.0
- General:
    - logging is now implemented via standard python "logging" module.
    "echo" keyword parameters are still functional but set/unset
    log levels for their respective classes/instances.  all logging
    can be controlled directly through the Python API by setting
    INFO and DEBUG levels for loggers in the "sqlalchemy" namespace.
    class-level logging is under "sqlalchemy.<module>.<classname>",
    instance-level logging under "sqlalchemy.<module>.<classname>.0x..<00-FF>".
    Test suite includes "--log-info" and "--log-debug" arguments
    which work independently of --verbose/--quiet.  Logging added
    to orm to allow tracking of mapper configurations, row iteration.
    - the documentation-generation system has been overhauled to be
    much simpler in design and more integrated with Markdown
- Specific Databases:
    - SQLite:
    - sqlite boolean datatype converts False/True to 0/1 by default
    - fixes to Date/Time (SLDate/SLTime) types; works as good as postgres
    now [ticket:335]
    - MS-SQL:
    - fixes bug 261 (table reflection broken for MS-SQL case-sensitive
    databases)
    - can now specify port for pymssql
    - introduces new "auto_identity_insert" option for auto-switching
    between "SET IDENTITY_INSERT" mode when values specified for IDENTITY columns
    - now supports multi-column foreign keys
    - fix to reflecting date/datetime columns
    - NCHAR and NVARCHAR type support added
    - Oracle:
    - Oracle has experimental support for cx_Oracle.TIMESTAMP, which requires
    a setinputsizes() call on the cursor that is now enabled via the
    'auto_setinputsizes' flag to the oracle dialect.
    - Firebird:
    - aliases do not use "AS"
    - correctly raises NoSuchTableError when reflecting non-existent table
- Schema:
    - a fair amount of cleanup to the schema package, removal of ambiguous
    methods, methods that are no longer needed.  slightly more constrained
    useage, greater emphasis on explicitness
    - the "primary_key" attribute of Table and other selectables becomes
    a setlike ColumnCollection object; is ordered but not numerically
    indexed.  a comparison clause between two pks that are derived from the
    same underlying tables (i.e. such as two Alias objects) can be generated
    via table1.primary_key==table2.primary_key
    - ForeignKey(Constraint) supports "use_alter=True", to create/drop a foreign key
    via ALTER.  this allows circular foreign key relationships to be set up.
    - append_item() methods removed from Table and Column; preferably
    construct Table/Column/related objects inline, but if needed use
    append_column(), append_foreign_key(), append_constraint(), etc.
    - table.create() no longer returns the Table object, instead has no
    return value.  the usual case is that tables are created via metadata,
    which is preferable since it will handle table dependencies.
    - added UniqueConstraint (goes at Table level), CheckConstraint
    (goes at Table or Column level).
    - index=False/unique=True on Column now creates a UniqueConstraint,
    index=True/unique=False creates a plain Index,
    index=True/unique=True on Column creates a unique Index.  'index'
    and 'unique' keyword arguments to column are now boolean only; for
    explcit names and groupings of indexes or unique constraints, use the
    UniqueConstraint/Index constructs explicitly.
    - added autoincrement=True to Column; will disable schema generation
    of SERIAL/AUTO_INCREMENT/identity seq for postgres/mysql/mssql if
    explicitly set to False
    - TypeEngine objects now have methods to deal with copying and comparing
    values of their specific type.  Currently used by the ORM, see below.
    - fixed condition that occurred during reflection when a primary key
    column was explciitly overridden, where the PrimaryKeyConstraint would
    get both the reflected and the programmatic column doubled up
    - the "foreign_key" attribute on Column and ColumnElement in general
    is deprecated, in favor of the "foreign_keys" list/set-based attribute,
    which takes into account multiple foreign keys on one column.
    "foreign_key" will return the first element in the "foreign_keys" list/set
    or None if the list is empty.
- Connections/Pooling/Execution:
    - connection pool tracks open cursors and automatically closes them
    if connection is returned to pool with cursors still opened.  Can be
    affected by options which cause it to raise an error instead, or to
    do nothing.  fixes issues with MySQL, others
    - fixed bug where Connection wouldnt lose its Transaction
    after commit/rollback
    - added scalar() method to ComposedSQLEngine, ResultProxy
    - ResultProxy will close() the underlying cursor when the ResultProxy
    itself is closed.  this will auto-close cursors for ResultProxy objects
    that have had all their rows fetched (or had scalar() called).
    - ResultProxy.fetchall() internally uses DBAPI fetchall() for better efficiency,
    added to mapper iteration as well (courtesy Michael Twomey)
- SQL Construction:
    - changed "for_update" parameter to accept False/True/"nowait"
    and "read", the latter two of which are interpreted only by
    Oracle and Mysql [ticket:292]
    - added extract() function to sql dialect
    (SELECT extract(field FROM expr))
    - BooleanExpression includes new "negate" argument to specify
    the appropriate negation operator if one is available.
    - calling a negation on an "IN" or "IS" clause will result in
    "NOT IN", "IS NOT" (as opposed to NOT (x IN y)).
    - Function objects know what to do in a FROM clause now.  their
    behavior should be the same, except now you can also do things like
    select(['*'], from_obj=[func.my_function()]) to get multiple
    columns from the result, or even use sql.column() constructs to name the
    return columns [ticket:172]
- ORM:
    - attribute tracking modified to be more intelligent about detecting
    changes, particularly with mutable types.  TypeEngine objects now
    take a greater role in defining how to compare two scalar instances,
    including the addition of a MutableType mixin which is implemented by
    PickleType.  unit-of-work now tracks the "dirty" list as an expression
    of all persistent objects where the attribute manager detects changes.
    The basic issue thats fixed is detecting changes on PickleType
    objects, but also generalizes type handling and "modified" object
    checking to be more complete and extensible.
    - a wide refactoring to "attribute loader" and "options" architectures.
    ColumnProperty and PropertyLoader define their loading behaivor via switchable
    "strategies", and MapperOptions no longer use mapper/property copying
    in order to function; they are instead propagated via QueryContext
    and SelectionContext objects at query/instances time.
    All of the internal copying of mappers and properties that was used to handle
    inheritance as well as options() has been removed; the structure
    of mappers and properties is much simpler than before and is clearly laid out
    in the new 'interfaces' module.
    - related to the mapper/property overhaul, internal refactoring to
    mapper instances() method to use a SelectionContext object to track
    state during the operation.
    SLIGHT API BREAKAGE: the append_result() and populate_instances()
    methods on MapperExtension have a slightly different method signature
    now as a result of the change; hoping that these methods are not
    in widespread use as of yet.
    - instances() method moved to Query now, backwards-compatible
    version remains on Mapper.
    - added contains_eager() MapperOption, used in conjunction with
    instances() to specify properties that should be eagerly loaded
    from the result set, using their plain column names by default, or translated
    given an custom row-translation function.
    - more rearrangements of unit-of-work commit scheme to better allow
    dependencies within circular flushes to work properly...updated
    task traversal/logging implementation
    - polymorphic mappers (i.e. using inheritance) now produces INSERT
    statements in order of tables across all inherited classes
    [ticket:321]
    - added an automatic "row switch" feature to mapping, which will
    detect a pending instance/deleted instance pair with the same
    identity key and convert the INSERT/DELETE to a single UPDATE
    - "association" mappings simplified to take advantage of
    automatic "row switch" feature
    - "custom list classes" is now implemented via the "collection_class"
    keyword argument to relation().  the old way still works but is
    deprecated [ticket:212]
    - added "viewonly" flag to relation(), allows construction of
    relations that have no effect on the flush() process.
    - added "lockmode" argument to base Query select/get functions,
    including "with_lockmode" function to get a Query copy that has
    a default locking mode.  Will translate "read"/"update"
    arguments into a for_update argument on the select side.
    [ticket:292]
    - implemented "version check" logic in Query/Mapper, used
    when version_id_col is in effect and query.with_lockmode()
    is used to get() an instance thats already loaded
    - post_update behavior improved; does a better job at not
    updating too many rows, updates only required columns
    [ticket:208]
    - adjustments to eager loading so that its "eager chain" is
    kept separate from the normal mapper setup, thereby
    preventing conflicts with lazy loader operation, fixes
    [ticket:308]
    - fix to deferred group loading
    - session.flush() wont close a connection it opened [ticket:346]
    - added "batch=True" flag to mapper; if False, save_obj
    will fully save one object at a time including calls
    to before_XXXX and after_XXXX
    - added "column_prefix=None" argument to mapper; prepends the
    given string (typically '_') to column-based attributes automatically
    set up from the mapper's Table
    - specifying joins in the from_obj argument of query.select() will
    replace the main table of the query, if the table is somewhere within
    the given from_obj.  this makes it possible to produce custom joins and
    outerjoins in queries without the main table getting added twice.
    [ticket:315]
    - eagerloading is adjusted to more thoughtfully attach its LEFT OUTER JOINs
    to the given query, looking for custom "FROM" clauses that may have
    already been set up.
    - added join_to and outerjoin_to transformative methods to SelectResults,
    to build up join/outerjoin conditions based on property names. also
    added select_from to explicitly set from_obj parameter.
    - removed "is_primary" flag from mapper.

0.2.8
- cleanup on connection methods + documentation.  custom DBAPI
arguments specified in query string, 'connect_args' argument
to 'create_engine', or custom creation function via 'creator'
function to 'create_engine'.
- added "recycle" argument to Pool, is "pool_recycle" on create_engine,
defaults to 3600 seconds; connections after this age will be closed and
replaced with a new one, to handle db's that automatically close
stale connections [ticket:274]
- changed "invalidate" semantics with pooled connection; will
instruct the underlying connection record to reconnect the next
time its called.  "invalidate" will also automatically be called
if any error is thrown in the underlying call to connection.cursor().
this will hopefully allow the connection pool to reconnect to a
database that had been stopped and started without restarting
the connecting application [ticket:121]
- eesh !  the tutorial doctest was broken for quite some time.
- add_property() method on mapper does a "compile all mappers"
step in case the given property references a non-compiled mapper
(as it did in the case of the tutorial !)
- [ticket:277] check for pg sequence already existing before create
- if a contextual session is established via MapperExtension.get_session
(as it is using the sessioncontext plugin, etc), a lazy load operation
will use that session by default if the parent object is not
persistent with a session already.
- lazy loads will not fire off for an object that does not have a
database identity (why?
see http://www.sqlalchemy.org/trac/wiki/WhyDontForeignKeysLoadData)
- unit-of-work does a better check for "orphaned" objects that are
part of a "delete-orphan" cascade, for certain conditions where the
parent isnt available to cascade from.
- mappers can tell if one of their objects is an "orphan" based
on interactions with the attribute package. this check is based
on a status flag maintained for each relationship
when objects are attached and detached from each other.
- it is now invalid to declare a self-referential relationship with
"delete-orphan" (as the abovementioned check would make them impossible
to save)
- improved the check for objects being part of a session when the
unit of work seeks to flush() them as part of a relationship..
- [ticket:280] statement execution supports using the same BindParam
object more than once in an expression; simplified handling of positional
parameters.  nice job by Bill Noon figuring out the basic idea.
- postgres reflection moved to use pg_schema tables, can be overridden
with use_information_schema=True argument to create_engine
[ticket:60], [ticket:71]
- added case_sensitive argument to MetaData, Table, Column, determines
itself automatically based on if a parent schemaitem has a non-None
setting for the flag, or if not, then whether the identifier name is all lower
case or not.  when set to True, quoting is applied to identifiers with mixed or
uppercase identifiers.  quoting is also applied automatically in all cases to
identifiers that are known to be reserved words or contain other non-standard
characters. various database dialects can override all of this behavior, but
currently they are all using the default behavior.  tested with postgres, mysql,
sqlite, oracle.  needs more testing with firebird, ms-sql. part of the ongoing
work with [ticket:155]
- unit tests updated to run without any pysqlite installed; pool
test uses a mock DBAPI
- urls support escaped characters in passwords [ticket:281]
- added limit/offset to UNION queries (though not yet in oracle)
- added "timezone=True" flag to DateTime and Time types.  postgres
so far will convert this to "TIME[STAMP] (WITH|WITHOUT) TIME ZONE",
so that control over timezone presence is more controllable (psycopg2
returns datetimes with tzinfo's if available, which can create confusion
against datetimes that dont).
- fix to using query.count() with distinct, **kwargs with SelectResults
count() [ticket:287]
- deregister Table from MetaData when autoload fails; [ticket:289]
- import of py2.5s sqlite3 [ticket:293]
- unicode fix for startswith()/endswith() [ticket:296]

0.2.7
- quoting facilities set up so that database-specific quoting can be
turned on for individual table, schema, and column identifiers when
used in all queries/creates/drops.  Enabled via "quote=True" in
Table or Column, as well as "quote_schema=True" in Table.  Thanks to
Aaron Spike for his excellent efforts.
- assignmapper was setting is_primary=True, causing all sorts of mayhem
by not raising an error when redundant mappers were set up, fixed
- added allow_null_pks option to Mapper, allows rows where some
primary key columns are null (i.e. when mapping to outer joins etc)
- modifcation to unitofwork to not maintain ordering within the
"new" list or within the UOWTask "objects" list; instead, new objects
are tagged with an ordering identifier as they are registered as new
with the session, and the INSERT statements are then sorted within the
mapper save_obj.  the INSERT ordering has basically been pushed all
the way to the end of the flush cycle. that way the various sorts and
organizations occuring within UOWTask (particularly the circular task
sort) dont have to worry about maintaining order (which they werent anyway)
- fixed reflection of foreign keys to autoload the referenced table
if it was not loaded already
- [ticket:256] - pass URL query string arguments to connect() function
- [ticket:257] - oracle boolean type
- custom primary/secondary join conditions in a relation *will* be propagated
to backrefs by default.  specifying a backref() will override this behavior.
- better check for ambiguous join conditions in sql.Join; propagates to a
better error message in PropertyLoader (i.e. relation()/backref()) for when
the join condition can't be reasonably determined.
- sqlite creates ForeignKeyConstraint objects properly upon table
reflection.
- adjustments to pool stemming from changes made for [ticket:224].
overflow counter should only be decremented if the connection actually
succeeded.  added a test script to attempt testing this.
- fixed mysql reflection of default values to be PassiveDefault
- added reflected 'tinyint', 'mediumint' type to MS-SQL [ticket:263],
[ticket:264]
- SingletonThreadPool has a size and does a cleanup pass, so that
only a given number of thread-local connections stay around (needed
for sqlite applications that dispose of threads en masse)
- fixed small pickle bug(s) with lazy loaders [ticket:265] [ticket:267]
- fixed possible error in mysql reflection where certain versions
return an array instead of string for SHOW CREATE TABLE call
- fix to lazy loads when mapping to joins [changeset:1770]
- all create()/drop() calls have a keyword argument of "connectable".
"engine" is deprecated.
- fixed ms-sql connect() to work with adodbapi
- added "nowait" flag to Select()
- inheritance check uses issubclass() instead of direct __mro__ check
to make sure class A inherits from B, allowing mapper inheritance to more
flexibly correspond to class inheritance [ticket:271]
- SelectResults will use a subselect, when calling an aggregate (i.e.
max, min, etc.) on a SelectResults that has an ORDER BY clause
[ticket:252]
- fixes to types so that database-specific types more easily used;
fixes to mysql text types to work with this methodology
[ticket:269]
- some fixes to sqlite date type organization
- added MSTinyInteger to MS-SQL [ticket:263]

0.2.6
- big overhaul to schema to allow truly composite primary and foreign
key constraints, via new ForeignKeyConstraint and PrimaryKeyConstraint
objects.
Existing methods of primary/foreign key creation have not been changed
but use these new objects behind the scenes.  table creation
and reflection is now more table oriented rather than column oriented.
[ticket:76]
- overhaul to MapperExtension calling scheme, wasnt working very well
previously
- tweaks to ActiveMapper, supports self-referential relationships
- slight rearrangement to objectstore (in activemapper/threadlocal)
so that the SessionContext is referenced by '.context' instead
of subclassed directly.
- activemapper will use threadlocal's objectstore if the mod is
activated when activemapper is imported
- small fix to URL regexp to allow filenames with '@' in them
- fixes to Session expunge/update/etc...needs more cleanup.
- select_table mappers *still* werent always compiling
- fixed up Boolean datatype
- added count()/count_by() to list of methods proxied by assignmapper;
this also adds them to activemapper
- connection exceptions wrapped in DBAPIError
- ActiveMapper now supports autoloading column definitions from the
database if you supply a __autoload__ = True attribute in your
mapping inner-class.  Currently this does not support reflecting
any relationships.
- deferred column load could screw up the connection status in
a flush() under some circumstances, this was fixed
- expunge() was not working with cascade, fixed.
- potential endless loop in cascading operations fixed.
- added "synonym()" function, applied to properties to have a
propname the same as another, for the purposes of overriding props
and allowing the original propname to be accessible in select_by().
- fix to typing in clause construction which specifically helps
type issues with polymorphic_union (CAST/ColumnClause propagates
its type to proxy columns)
- mapper compilation work ongoing, someday it'll work....moved
around the initialization of MapperProperty objects to be after
all mappers are created to better handle circular compilations.
do_init() method is called on all properties now which are more
aware of their "inherited" status if so.
- eager loads explicitly disallowed on self-referential relationships, or
relationships to an inheriting mapper (which is also self-referential)
- reduced bind param size in query._get to appease the picky oracle
[ticket:244]
- added 'checkfirst' argument to table.create()/table.drop(), as
well as table.exists() [ticket:234]
- some other ongoing fixes to inheritance [ticket:245]
- attribute/backref/orphan/history-tracking tweaks as usual...

0.2.5
- fixed endless loop bug in select_by(), if the traversal hit
two mappers that referenced each other
- upgraded all unittests to insert './lib/' into sys.path,
working around new setuptools PYTHONPATH-killing behavior
- further fixes with attributes/dependencies/etc....
- improved error handling for when DynamicMetaData is not connected
- MS-SQL support largely working (tested with pymssql)
- ordering of UPDATE and DELETE statements within groups is now
in order of primary key values, for more deterministic ordering
- after_insert/delete/update mapper extensions now called per object,
not per-object-per-table
- further fixes/refactorings to mapper compilation

0.2.4
- try/except when the mapper sets init.__name__ on a mapped class,
supports python 2.3
- fixed bug where threadlocal engine would still autocommit
despite a transaction in progress
- lazy load and deferred load operations require the parent object
to be in a Session to do the operation; whereas before the operation
would just return a blank list or None, it now raises an exception.
- Session.update() is slightly more lenient if the session to which
the given object was formerly attached to was garbage collected;
otherwise still requires you explicitly remove the instance from
the previous Session.
- fixes to mapper compilation, checking for more error conditions
- small fix to eager loading combined with ordering/limit/offset
- utterly remarkable:  added a single space between 'CREATE TABLE'
and '(<the rest of it>' since *thats how MySQL indicates a non-
reserved word tablename.....* [ticket:206]
- more fixes to inheritance, related to many-to-many relations
properly saving
- fixed bug when specifying explicit module to mysql dialect
- when QueuePool times out it raises a TimeoutError instead of
erroneously making another connection
- Queue.Queue usage in pool has been replaced with a locally
modified version (works in py2.3/2.4!) that uses a threading.RLock
for a mutex.  this is to fix a reported case where a ConnectionFairy's
__del__() method got called within the Queue's get() method, which
then returns its connection to the Queue via the the put() method,
causing a reentrant hang unless threading.RLock is used.
- postgres will not place SERIAL keyword on a primary key column
if it has a foreign key constraint
- cursor() method on ConnectionFairy allows db-specific extension
arguments to be propagated [ticket:221]
- lazy load bind params properly propagate column type [ticket:225]
- new MySQL types: MSEnum, MSTinyText, MSMediumText, MSLongText, etc.
more support for MS-specific length/precision params in numeric types
patch courtesy Mike Bernson
- some fixes to connection pool invalidate() [ticket:224]

0.2.3
- overhaul to mapper compilation to be deferred.  this allows mappers
to be constructed in any order, and their relationships to each
other are compiled when the mappers are first used.
- fixed a pretty big speed bottleneck in cascading behavior particularly
when backrefs were in use
- the attribute instrumentation module has been completely rewritten; its
now a large degree simpler and clearer, slightly faster.  the "history"
of an attribute is no longer micromanaged with each change and is
instead part of a "CommittedState" object created when the
instance is first loaded.  HistoryArraySet is gone, the behavior of
list attributes is now more open ended (i.e. theyre not sets anymore).
- py2.4 "set" construct used internally, falls back to sets.Set when
"set" not available/ordering is needed.
- fix to transaction control, so that repeated rollback() calls
dont fail (was failing pretty badly when flush() would raise
an exception in a larger try/except transaction block)
- "foreignkey" argument to relation() can also be a list.  fixed
auto-foreignkey detection [ticket:151]
- fixed bug where tables with schema names werent getting indexed in
the MetaData object properly
- fixed bug where Column with redefined "key" property wasnt getting
type conversion happening in the ResultProxy [ticket:207]
- fixed 'port' attribute of URL to be an integer if present
- fixed old bug where if a many-to-many table mapped as "secondary"
had extra columns, delete operations didnt work
- bugfixes for mapping against UNION queries
- fixed incorrect exception class thrown when no DB driver present
- added NonExistentTable exception thrown when reflecting a table
that doesnt exist [ticket:138]
- small fix to ActiveMapper regarding one-to-one backrefs, other
refactorings
- overridden constructor in mapped classes gets __name__ and
__doc__ from the original class
- fixed small bug in selectresult.py regarding mapper extension
[ticket:200]
- small tweak to cascade_mappers, not very strongly supported
function at the moment
- some fixes to between(), column.between() to propagate typing
information better [ticket:202]
- if an object fails to be constructed, is not added to the
session [ticket:203]
- CAST function has been made into its own clause object with
its own compilation function in ansicompiler; allows MySQL
to silently ignore most CAST calls since MySQL
seems to only support the standard CAST syntax with Date types.
MySQL-compatible CAST support for strings, ints, etc. a TODO

0.2.2
- big improvements to polymorphic inheritance behavior, enabling it
to work with adjacency list table structures [ticket:190]
- major fixes and refactorings to inheritance relationships overall,
more unit tests
- fixed "echo_pool" flag on create_engine()
- fix to docs, removed incorrect info that close() is unsafe to use
with threadlocal strategy (its totally safe !)
- create_engine() can take URLs as string or unicode [ticket:188]
- firebird support partially completed;
thanks to James Ralston and Brad Clements for their efforts.
- Oracle url translation was broken, fixed, will feed host/port/sid
into cx_oracle makedsn() if 'database' field is present, else uses
straight TNS name from the 'host' field
- fix to using unicode criterion for query.get()/query.load()
- count() function on selectables now uses table primary key or
first column instead of "1" for criterion, also uses label "rowcount"
instead of "count".
- got rudimental "mapping to multiple tables" functionality cleaned up,
more correctly documented
- restored global_connect() function, attaches to a DynamicMetaData
instance called "default_metadata".  leaving MetaData arg to Table
out will use the default metadata.
- fixes to session cascade behavior, entity_name propigation
- reorganized unittests into subdirectories
- more fixes to threadlocal connection nesting patterns

0.2.1
- "pool" argument to create_engine() properly propagates
- fixes to URL, raises exception if not parsed, does not pass blank
fields along to the DB connect string (a string such as
user:host@/db was breaking on postgres)
- small fixes to Mapper when it inserts and tries to get
new primary key values back
- rewrote half of TLEngine, the ComposedSQLEngine used with
'strategy="threadlocal"'.  it now properly implements engine.begin()/
engine.commit(), which nest fully with connection.begin()/trans.commit().
added about six unittests.
- major "duh" in pool.Pool, forgot to put back the WeakValueDictionary.
unittest which was supposed to check for this was also silently missing
it.  fixed unittest to ensure that ConnectionFairy properly falls out
of scope.
- placeholder dispose() method added to SingletonThreadPool, doesnt
do anything yet
- rollback() is automatically called when an exception is raised,
but only if theres no transaction in process (i.e. works more like
autocommit).
- fixed exception raise in sqlite if no sqlite module present
- added extra example detail for association object doc
- Connection adds checks for already being closed

0.2.0
- overhaul to Engine system so that what was formerly the SQLEngine
is now a ComposedSQLEngine which consists of a variety of components,
including a Dialect, ConnectionProvider, etc. This impacted all the
db modules as well as Session and Mapper.
- create_engine now takes only RFC-1738-style strings:
driver://user:password@host:port/database
- total rewrite of connection-scoping methodology, Connection objects
can now execute clause elements directly, added explicit "close" as
well as support throughout Engine/ORM to handle closing properly,
no longer relying upon __del__ internally to return connections
to the pool [ticket:152].
- overhaul to Session interface and scoping.  uses hibernate-style
methods, including query(class), save(), save_or_update(), etc.
no threadlocal scope is installed by default.  Provides a binding
interface to specific Engines and/or Connections so that underlying
Schema objects do not need to be bound to an Engine.  Added a basic
SessionTransaction object that can simplistically aggregate transactions
across multiple engines.
- overhaul to mapper's dependency and "cascade" behavior; dependency logic
factored out of properties.py into a separate module "dependency.py".
"cascade" behavior is now explicitly controllable, proper implementation
of "delete", "delete-orphan", etc.  dependency system can now determine at
flush time if a child object has a parent or not so that it makes better
decisions on how that child should be updated in the DB with regards to deletes.
- overhaul to Schema to build upon MetaData object instead of an Engine.
Entire SQL/Schema system can be used with no Engines whatsoever, executed
solely by an explicit Connection object.  the "bound" methodlogy exists via the
BoundMetaData for schema objects.  ProxyEngine is generally not needed
anymore and is replaced by DynamicMetaData.
- true polymorphic behavior implemented, fixes [ticket:167]
- "oid" system has been totally moved into compile-time behavior;
if they are used in an order_by where they are not available, the order_by
doesnt get compiled, fixes [ticket:147]
- overhaul to packaging; "mapping" is now "orm", "objectstore" is now
"session", the old "objectstore" namespace gets loaded in via the
"threadlocal" mod if used
- mods now called in via "import <modname>".  extensions favored over
mods as mods are globally-monkeypatching
- fix to add_property so that it propagates properties to inheriting
mappers [ticket:154]
- backrefs create themselves against primary mapper of its originating
property, priamry/secondary join arguments can be specified to override.
helps their usage with polymorphic mappers
- "table exists" function has been implemented [ticket:31]
- "create_all/drop_all" added to MetaData object [ticket:98]
- improvements and fixes to topological sort algorithm, as well as more
unit tests
- tutorial page added to docs which also can be run with a custom doctest
runner to ensure its properly working.  docs generally overhauled to
deal with new code patterns
- many more fixes, refactorings.
- migration guide is available on the Wiki at
http://www.sqlalchemy.org/trac/wiki/02Migration

0.1.7
- some fixes to topological sort algorithm
- added DISTINCT ON support to Postgres (just supply distinct=[col1,col2..])
- added __mod__ (% operator) to sql expressions
- "order_by" mapper property inherited from inheriting mapper
- fix to column type used when mapper UPDATES/DELETEs
- with convert_unicode=True, reflection was failing, has been fixed
- types types types!  still werent working....have to use TypeDecorator again :(
- mysql binary type converts array output to buffer, fixes PickleType
- fixed the attributes.py memory leak once and for all
- unittests are qualified based on the databases that support each one
- fixed bug where column defaults would clobber VALUES clause of insert objects
- fixed bug where table def w/ schema name would force engine connection
- fix for parenthesis to work correctly with subqueries in INSERT/UPDATE
- HistoryArraySet gets extend() method
- fixed lazyload support for other comparison operators besides =
- lazyload fix where two comparisons in the join condition point to the
samem column
- added "construct_new" flag to mapper, will use __new__ to create instances
instead of __init__ (standard in 0.2)
- added selectresults.py to SVN, missed it last time
- tweak to allow a many-to-many relationship from a table to itself via
an association table
- small fix to "translate_row" function used by polymorphic example
- create_engine uses cgi.parse_qsl to read query string (out the window in 0.2)
- tweaks to CAST operator
- fixed function names LOCAL_TIME/LOCAL_TIMESTAMP -> LOCALTIME/LOCALTIMESTAMP
- fixed order of ORDER BY/HAVING in compile

0.1.6
- support for MS-SQL added courtesy Rick Morrison, Runar Petursson
- the latest SQLSoup from J. Ellis
- ActiveMapper has preliminary support for inheritance (Jeff Watkins)
- added a "mods" system which allows pluggable modules that modify/augment
core functionality, using the function "install_mods(*modnames)".
- added the first "mod", SelectResults, which modifies mapper selects to
return generators that turn ranges into LIMIT/OFFSET queries 
(Jonas Borgstr?
- factored out querying capabilities of Mapper into a separate Query object
which is Session-centric.  this improves the performance of mapper.using(session)
and makes other things possible.
- objectstore/Session refactored, the official way to save objects is now
via the flush() method.  The begin/commit functionality of Session is factored
into LegacySession which is still established as the default behavior, until
the 0.2 series.
- types system is bound to an engine at query compile time, not schema
construction time.  this simplifies the types system as well as the ProxyEngine.
- added 'version_id' keyword argument to mapper. this keyword should reference a
Column object with type Integer, preferably non-nullable, which will be used on
the mapped table to track version numbers. this number is incremented on each
save operation and is specifed in the UPDATE/DELETE conditions so that it
factors into the returned row count, which results in a ConcurrencyError if the
value received is not the expected count.
- added 'entity_name' keyword argument to mapper. a mapper is now associated
with a class via the class object as well as an optional entity_name parameter,
which is a string defaulting to None. any number of primary mappers can be
created for a class, qualified by the entity name. instances of those classes
will issue all of their load and save operations through their
entity_name-qualified mapper, and maintain separate a identity in the identity
map for an otherwise equilvalent object.
- overhaul to the attributes system. code has been clarified, and also fixed to
support proper polymorphic behavior on object attributes.
- added "for_update" flag to Select objects
- some fixes for backrefs
- fix for postgres1 DateTime type
- documentation pages mostly switched over to Markdown syntax

0.1.5
- added SQLSession concept to SQLEngine. this object keeps track of retrieving a
connection from the connection pool as well as an in-progress transaction.
methods push_session() and pop_session() added to SQLEngine which push/pop a new
SQLSession onto the engine, allowing operation upon a second connection "nested"
within the previous one, allowing nested transactions. Other tricks are sure to
come later regarding SQLSession.
- added nest_on argument to objectstore.Session. This is a single SQLEngine or
list of engines for which push_session()/pop_session() will be called each time
this Session becomes the active session (via objectstore.push_session() or
equivalent). This allows a unit of work Session to take advantage of the nested
transaction feature without explicitly calling push_session/pop_session on the
engine.
- factored apart objectstore/unitofwork to separate "Session scoping" from
"uow commit heavy lifting"
- added populate_instance() method to MapperExtension. allows an extension to
modify the population of object attributes. this method can call the
populate_instance() method on another mapper to proxy the attribute population
from one mapper to another; some row translation logic is also built in to help
with this.
- fixed Oracle8-compatibility "use_ansi" flag which converts JOINs to
comparisons with the = and (+) operators, passes basic unittests
- tweaks to Oracle LIMIT/OFFSET support
- Oracle reflection uses ALL_** views instead of USER_** to get larger
list of stuff to reflect from
- fixes to Oracle foreign key reflection [ticket:105]
- objectstore.commit(obj1, obj2,...) adds an extra step to seek out private
relations on properties and delete child objects, even though its not a global
commit
- lots and lots of fixes to mappers which use inheritance, strengthened the
concept of relations on a mapper being made towards the "local" table for that
mapper, not the tables it inherits.  allows more complex compositional patterns
to work with lazy/eager loading.
- added support for mappers to inherit from others based on the same table,
just specify the same table as that of both parent/child mapper.
- some minor speed improvements to the attributes system with regards to
instantiating and populating new objects.
- fixed MySQL binary unit test
- INSERTs can receive clause elements as VALUES arguments, not just literal
values
- support for calling multi-tokened functions, i.e. schema.mypkg.func()
- added J. Ellis' SQLSoup module to extensions package
- added "polymorphic" examples illustrating methods to load multiple object types
from one mapper, the second of which uses the new populate_instance() method.
small improvements to mapper, UNION construct to help the examples along
- improvements/fixes to session.refresh()/session.expire() (which may have
been called "invalidate" earlier..)
- added session.expunge() which totally removes an object from the current
session
- added *args, **kwargs pass-thru to engine.transaction(func) allowing easier
creation of transactionalizing decorator functions
- added iterator interface to ResultProxy:  "for row in result:..."
- added assertion to tx = session.begin(); tx.rollback(); tx.begin(), i.e. cant
use it after a rollback()
- added date conversion on bind parameter fix to SQLite enabling dates to
work with pysqlite1
- improvements to subqueries to more intelligently construct their FROM
clauses [ticket:116]
- added PickleType to types.
- fixed two bugs with column labels with regards to bind parameters: bind param
keynames they are now generated from a column "label" in all relevant cases to
take advantage of excess-name-length rules, and checks for a peculiar collision
against a column named the same as "tablename_colname" added
- major overhaul to unit of work documentation, other documentation sections.
- fixed attributes bug where if an object is committed, its lazy-loaded list got
blown away if it hadnt been loaded
- added unique_connection() method to engine, connection pool to return a
connection that is not part of the thread-local context or any current
transaction
- added invalidate() function to pooled connection.  will remove the connection
from the pool.  still need work for engines to auto-reconnect to a stale DB
though.
- added distinct() function to column elements so you can do
func.count(mycol.distinct())
- added "always_refresh" flag to Mapper, creates a mapper that will always
refresh the attributes of objects it gets/selects from the DB, overwriting any
changes made.

0.1.4
- create_engine() now uses genericized parameters; host/hostname,
db/dbname/database, password/passwd, etc. for all engine connections. makes
 engine URIs much more "universal"
- added support for SELECT statements embedded into a column clause, using the
flag "scalar=True"
- another overhaul to EagerLoading when used in conjunction with mappers that
inherit; improvements to eager loads figuring out their aliased queries
correctly, also relations set up against a mapper with inherited mappers will
create joins against the table that is specific to the mapper itself (i.e. and
not any tables that are inherited/are further down the inheritance chain),
this can be overridden by using custom primary/secondary joins.
- added J.Ellis patch to mapper.py so that selectone() throws an exception
if query returns more than one object row, selectfirst() to not throw the
exception. also adds selectfirst_by (synonymous with get_by) and selectone_by
- added onupdate parameter to Column, will exec SQL/python upon an update
statement.Also adds "for_update=True" to all DefaultGenerator subclasses
- added support for Oracle table reflection contributed by Andrija Zaric;
still some bugs to work out regarding composite primary keys/dictionary selection
- checked in an initial Firebird module, awaiting testing.
- added sql.ClauseParameters dictionary object as the result for
compiled.get_params(), does late-typeprocessing of bind parameters so
that the original values are easier to access
- more docs for indexes, column defaults, connection pooling, engine construction
- overhaul to the construction of the types system. uses a simpler inheritance
pattern so that any of the generic types can be easily subclassed, with no need
for TypeDecorator.
- added "convert_unicode=False" parameter to SQLEngine, will cause all String
types to perform unicode encoding/decoding (makes Strings act like Unicodes)
- added 'encoding="utf8"' parameter to engine.  the given encoding will be
used for all encode/decode calls within Unicode types as well as Strings
when convert_unicode=True.
- improved support for mapping against UNIONs, added polymorph.py example
to illustrate multi-class mapping against a UNION
- fix to SQLite LIMIT/OFFSET syntax
- fix to Oracle LIMIT syntax
- added backref() function, allows backreferences to have keyword arguments
that will be passed to the backref.
- Sequences and ColumnDefault objects can do execute()/scalar() standalone
- SQL functions (i.e. func.foo()) can do execute()/scalar() standalone
- fix to SQL functions so that the ANSI-standard functions, i.e. current_timestamp
etc., do not specify parenthesis.  all other functions do.
- added settattr_clean and append_clean to SmartProperty, which set
attributes without triggering a "dirty" event or any history. used as:
myclass.prop1.setattr_clean(myobject, 'hi')
- improved support to column defaults when used by mappers; mappers will pull
pre-executed defaults from statement's executed bind parameters
(pre-conversion) to populate them into a saved object's attributes; if any
PassiveDefaults have fired off, will instead post-fetch the row from the DB to
populate the object.
- added 'get_session().invalidate(*obj)' method to objectstore, instances will
refresh() themselves upon the next attribute access.
- improvements to SQL func calls including an "engine" keyword argument so
they can be execute()d or scalar()ed standalone, also added func accessor to
SQLEngine
- fix to MySQL4 custom table engines, i.e. TYPE instead of ENGINE
- slightly enhanced logging, includes timestamps and a somewhat configurable
formatting system, in lieu of a full-blown logging system
- improvements to the ActiveMapper class from the TG gang, including
many-to-many relationships
- added Double and TinyInt support to mysql

0.1.3
- completed "post_update" feature, will add a second update statement before
inserts and after deletes in order to reconcile a relationship without any
dependencies being created; used when persisting two rows that are dependent
on each other
- completed mapper.using(session) function, localized per-object Session
functionality; objects can be declared and manipulated as local to any
user-defined Session
- fix to Oracle "row_number over" clause with multiple tables
- mapper.get() was not selecting multiple-keyed objects if the mapper's table was a join,
such as in an inheritance relationship, this is fixed.
- overhaul to sql/schema packages so that the sql package can run all on its own,
producing selects, inserts, etc. without any engine dependencies.  builds upon
new TableClause/ColumnClause lexical objects.  Schema's Table/Column objects
are the "physical" subclasses of them.  simplifies schema/sql relationship,
extensions (like proxyengine), and speeds overall performance by a large margin.
removes the entire getattr() behavior that plagued 0.1.1.
- refactoring of how the mapper "synchronizes" data between two objects into a
separate module, works better with properties attached to a mapper that has an
additional inheritance relationship to one of the related tables, also the same
methodology used to synchronize parent/child objects now used by mapper to
synchronize between inherited and inheriting mappers.
- made objectstore "check for out-of-identitymap" more aggressive, will perform the
check when object attributes are modified or the object is deleted
- Index object fully implemented, can be constructed standalone, or via
"index" and "unique" arguments on Columns.
- added "convert_unicode" flag to SQLEngine, will treat all String/CHAR types
as Unicode types, with raw-byte/utf-8 translation on the bind parameter and
result set side.
- postgres maintains a list of ANSI functions that must have no parenthesis so
function calls with no arguments work consistently
- tables can be created with no engine specified.  this will default their engine
to a module-scoped "default engine" which is a ProxyEngine.  this engine can
be connected via the function "global_connect".
- added "refresh(*obj)" method to objectstore / Session to reload the attributes of
any set of objects from the database unconditionally

0.1.2
- fixed a recursive call in schema that was somehow running 994 times then returning
normally.  broke nothing, slowed down everything.  thanks to jpellerin for finding this.

0.1.1
- small fix to Function class so that expressions with a func.foo() use the type of the
Function object (i.e. the left side) as the type of the boolean expression, not the
other side which is more of a moving target (changeset 1020).
- creating self-referring mappers with backrefs slightly easier (but still not that easy -
changeset 1019)
- fixes to one-to-one mappings (changeset 1015)
- psycopg1 date/time issue with None fixed (changeset 1005)
- two issues related to postgres, which doesnt want to give you the "lastrowid"
since oids are deprecated:
   * postgres database-side defaults that are on primary key cols *do* execute
explicitly beforehand, even though thats not the idea of a PassiveDefault.  this is
because sequences on columns get reflected as PassiveDefaults, but need to be explicitly
executed on a primary key col so we know what we just inserted.
   * if you did add a row that has a bunch of database-side defaults on it,
and the PassiveDefault thing was working the old way, i.e. they just execute on
the DB side, the "cant get the row back without an OID" exception that occurred
also will not happen unless someone (usually the ORM) explicitly asks for it.
- fixed a glitch with engine.execute_compiled where it was making a second
ResultProxy that just got thrown away.
- began to implement newer logic in object properities.  you can now say
myclass.attr.property, which will give you the PropertyLoader corresponding to that
attribute, i.e. myclass.mapper.props['attr']
- eager loading has been internally overhauled to use aliases at all times.  more
complicated chains of eager loads can now be created without any need for explicit
"use aliases"-type instructions.  EagerLoader code is also much simpler now.
- a new somewhat experimental flag "use_update" added to relations, indicates that
this relationship should be handled by a second UPDATE statement, either after a
primary INSERT or before a primary DELETE.  handles circular row dependencies.
- added exceptions module, all raised exceptions (except for some
KeyError/AttributeError exceptions) descend from these classes.
- fix to date types with MySQL, returned timedelta converted to datetime.time
- two-phase objectstore.commit operations (i.e. begin/commit) now return a
transactional object (SessionTrans), to more clearly indicate transaction boundaries.
- Index object with create/drop support added to schema
- fix to postgres, where it will explicitly pre-execute a PassiveDefault on a table
if it is a primary key column, pursuant to the ongoing "we cant get inserted rows
back from postgres" issue
- change to information_schema query that gets back postgres table defs, now
uses explicit JOIN keyword, since one user had faster performance with 8.1
- fix to engine.process_defaults so it works correctly with a table that has
different column name/column keys (changset 982)
- a column can only be attached to one table - this is now asserted
- postgres time types descend from Time type
- fix to alltests so that it runs types test (now named testtypes)
- fix to Join object so that it correctly exports its foreign keys (cs 973)
- creating relationships against mappers that use inheritance fixed (cs 973)

0.1.0
initial release