- Better support for Python 3.x by running 2to3 within setup (patch by
"foogod", closes #110).
- Added more tests for relationships to forward-declared entities.
- Dropped support for python 2.3, SQLAlchemy 0.4 and deprecated stuff from
- Fixed a few tests to work on SA 0.6.x
- Fixed bad foreign key constraint generated for classes inheriting from a
class with multiple primary keys when using the "multi" inheritance.
Patch from & closes #114.
0.7.1 - 2009-11-16
- Entities can now be declared "abstract" so that they do not create a table,
etc... This allows, among others, an entity to inherit from multiple abstract
classes (patch from Stéphane Klein, closes #89).
- Added a new collection which can resolve entities relative to the current
entity, for example "..other_module.Class" (based on patches from Johannes
Janssen, closes #93).
- Added a new entity option "resolve_root", which allows one to specify the
root module where your entities are defined. The string will be prepended
to all "absolute" entity paths. It can also be used on a per-entity basis.
This feature is based on a patch from Johannes Janssen, see #93.
- using_options_defaults and using_table_options statements can be used several
times within the same class (closes #70).
- Fixed custom base classes and versioned extension when used with zope
interfaces (closes #98, patch from Valentin Lab)
- Fixed having relationships in custom base classes (based on patch
by Stéphane Klein)
0.7.0 - 2009-10-01
Please see http://elixir.ematia.de/trac/wiki/Migrate06to07 for detailed
upgrade notes. If you are upgrading an application with existing data from an
earlier version of Elixir, you are STRONGLY ADVISED to read them!
- Added a new statement 'using_options_defaults' which can be used
on a custom base class to set default values for the options of all its
subclasses. For example, this makes it possible to have all classes
inheriting from your custom base class use some custom options without
having to set it individually on each entity, nor modify
- The local_colname and remote_colname arguments on ManyToMany relationships
can now also be used to set custom names for the ManyToMany table columns.
This effectively replace the column_format on ManyToMany relationships which
is now deprecated. Change based on a patch from Diez B. Roggisch.
- Added (or rather fixed and documented) a "table" argument on ManyToMany
relationships to allow using a manually-defined Table (closes #44).
- Added a "schema" argument on ManyToMany relationships to be able to create
the ManyToMany table in a custom schema and not necessarily the same schema
as the table of the "source" entity (patch from Diez B. Roggisch).
- Added a "table_kwargs" argument on ManyToMany relationships to pass any
extra keyword arguments to the underlying Table object (closes #94).
- Added a new "target_column" argument on ManyToOne relationships so that you
can target unique but non-primary key columns. At the moment, this only works
if the target column is declared before the ManyToOne (see ticket #26).
- Added new "column_names" argument to the acts_as_versioned extension,
allowing to specify custom column names (inspired by a patch by
- Made M2MCOL_NAMEFORMAT option accept a callable, so that more complex naming
generation can be used if so desired.
- Added two new extensions (contributed by Alex Bodnaru)
- "perform_ddl" allows to execute one or several DDL statements upon table
- "preload_data" allows to insert data into the entity table just after it
has been created.
- EntityCollection gained a working extend method
- Moved class instrumentation to a separate function: instrument_class
- Moved all methods of the "Entity" base class, to the "EntityBase" class, so
that people who want to provide their own base class but don't want to loose
all the methods provided by "Entity" can simply inherit from EntityBase
instead of copy-pasting the code for all its methods.
- Renamed remote_side and local_side ManyToMany arguments to remote_colname and
local_colname respectively to not collide with the remote_side argument
provided by SA (it doesn't make much sense on ManyToMany relationships but
- Delete Elixir properties in the setup phase rather than as soon as they are
attached to their class. It makes it possible to access them or refer to them
after the class is defined (but before setup).
- Deprecated act_as_list extension in favor of SQLAlchemy's orderinglist
extension (closes #53).
- Made manually defined mapper options take precedence over Elixir-generated
ones. Not very useful yet since most are expecting Column objects.
- Default table_options (defined in options_defaults['table_options']) are now
also used for ManyToMany tables (closes #94).
- Provide our own Session.mapper equivalent to avoid SQLAlchemy 0.5.5+
deprecation warning. This mapper autosave object instances on __init__ unless
save_on_init=False is passed as a mapper argument (closes #92).
- Fixed Elixir to work with the future SQLAlchemy 0.6 (trunk as of 6377).
- Changed the pattern used by default to generate column names for
self-referencial ManyToMany relationships so that the meaning of
bidirectional self-referential relationships does not depend on the order of
declaration of each side (closes #69). See upgrade notes for details.
- Changed slightly the pattern used to generate the name of the table for
bidirectional self-referential ManyToMany relationships so that it doesn't
depend on the order of declaration of each side (closes #19).
See upgrade notes for details.
- default EntityCollection raise an exception instead of returning None when
trying to resolve an inexisting Entity from outside of another entity (for
example through EntityCollection.__getattr__
- Fixed the case where you specify both "primaryjoin" and "colname" arguments
(useless in this case, but harmless) on a ManyToOne relationship of an
- Fixed bug which broke the "identity" (Entity) option
- Fixed documentation about local_side and remote_side arguments being
required if the entity containing the relationship is autoloaded, when it is
only required if the relationship is self-referencial, and primaryjoin or
secondaryjoin as not been specified manually.
- Added missing documentation for the "filter" argument on OneToMany
- Fixed the act_as_list extension's move_to_bottom method to work on MySQL
- Fixed event methods not being called when they are defined on a parent class.
(introduced in release 0.5.0).
- Added workaround for an odd mod_python behavior (class.__module__ returns a
weird name which is not in sys.modules).
- tablename, remote_colname, local_colname, schema and table_kwargs can now be
defined on either side of a ManyToMany relationship and will propagate to
the other side if that other side doesn't specify anything for that argument.
Also added an assertion to catch the case where the same/mirror
argument is specified on both sides but with different values.
- Fixed filter argument on OneToMany relationship leaking the filter to the
- Fixed encrypted extension to not encrypt several times an instance attributes
when that instance is flushed several times before being expunged from the
- Fixed using to_dict with a ManyToOne relationship in the "deep" set and that
relationship being None in the entity being converted.
0.6.1 - 2008-08-18
- Allow ManyToOne relationships to use manually created fields as their
"supporting column". This means that the columns can be customized without
resorting to using the ugly "column_kwargs" (patch from Jason R. Coombs,
- Extra args and kwargs to Synonym and ColumnProperty are forwarded to their
underlying constructs. This allows for example deferred ColumnProperties.
- Added a more helpful assertion message when inverse relationship types don't
- Removed support for the deprecated "with_fields" syntax
- Entity.__init__ calls Entity.set instead of duplicating its functionality
- Fixed the "Target resolves to several entities" exception message to actually
include the target name.
- Renamed the on_reconstitute method decorator to reconstructor, to track the
corresponding change in SA's trunk.
0.6.0 - 2008-07-18
Please see http://elixir.ematia.de/trac/wiki/Migrate05to06 for detailed
- Fields in a custom base class are added to all their children.
- Added two new methods on the base entity: from_dict and to_dict, which can
be used to create (or output) a whole hierarchy of instances from (to) a
simple JSON-like dictionary notation (patch from Paul Johnston,
closes ticket #40).
- Added experimental (!) support for concrete table inheritance (both
polymorphic or not). Concrete polymorphic inheritance requires SQLAlchemy
0.4.5 or later.
- Moved the "entity to string" mapping and resolving code to the (newly
created) EntityCollection class (which stores lists of entities). This
allows one to provide a custom mapping method if needed. The default class
also overrides the __getattr__ method, providing and handy way to get at your
entities. See http://elixir.ematia.de/trac/browser/elixir/tags/0.6.0/tests/test_collections#L58
- Added new "identity" option which can be used to set a custom polymorphic
identity for an entity. It also accepts a callable so that you can generate
the identity name automatically from the class itself.
- Added __setattr__ method on Metaclass so that you can add properties
slightly more easily after class definition (but *before* setup_all):
A.name = Field(String(32))
- Added add_table_column, add_mapper_property and add_mapper_extension helper
methods in EntityBuilders.
- Added full_tablename property on EntityDescriptor (includes schema name if
- Added on_reconstitute event/method decorator. Only works with SA 0.5.
- Added support for viewonly relationships (OneToMany and OneToOne).
- Added support for filtered OneToMany relationships. Produce viewonly
relations. See http://elixir.ematia.de/trac/browser/elixir/tags/0.6.0/tests/test_o2m.py
for an example.
- Added support for callables for some arguments on relationships: primaryjoin,
secondaryjoin and remote_side. It means those can be evaluated at setup time
(when tables and their columns already exist) instead of definition time.
- Default "target entity resolving code" changed slightly. It now uses a global
collection keyed on the entity name. This means that entities can refer to
other entities in a different module simply with the target entity name
instead of its full path. The full path is only required when there is an
ambiguity (ie when there are two classes with the same name in two different
modules). Closes #9.
- Added support for SQLAlchemy 0.5, and dropped support for version 0.3 and
- The default session (elixir.session) uses sessionmaker() instead of
create_session(), which means it has now the following characteristics:
* autocommit=False (with SA 0.5 -- equivalent to transactional=True with
* autoexpire=True (with SA 0.5).
- removed objectstore and other SA 0.3 (or older) support code.
- Fixed multi-table inheritance when using a non default schema (closes #38)
- Fixed ManyToOne relationships using 'key' kwarg in their column_kwargs
(patch by Jason R. Coombs)
- Fixed inheritance with autoloaded entities: when using autoload, we
shouldn't try to add columns to the table (closes tickets #41 and #43).
- Fixed acts_as_list extension with autoloaded entities (patch from maqr,
closes ticket #52).
- Fixed ColumnProperty to work with latest version of SQLAlchemy (O.4.5 and
- Fixed ManyToMany relationships when not using the default schema
(patch from Diez B. Roggisch, closes ticket #48)
- Added AUTHORS list. If you are missing from this list, don't hesitate to
0.5.2 - 2008-03-28
- Added an optional `check_concurrency` keyword argument to the versioning
extension, supporting the usage of SQLAlchemy's built-in optimistic
- Made Elixir python 2.3 compatible again (based on patches from
Jason R. Coombs)
- Fixed act_as_list extension to work with DBMS that require subselects to be
aliased (patch by Alice McGregor)
- Fixed the versioning extension so that the history table is updated within
the current transaction (patch from and closes ticket #35).
0.5.1 - 2008-02-07
- Added a new elixir plugin for managing entities as (ordered) lists.
- Added a `column_format` keyword argument to `ManyToMany` which can be used
to specify an alternate format string for column names in the mapping table.
- Added support for custom base classes which inherit from another class (ie
not directly from object).
- Added an alternate (nicer) syntax to define synonym properties. This syntax
has a more limited scope, except that it can refer to properties defined in
a parent entity. This is based on a patch from Alexandre da Silva.
- Added check so that using an inexisting column in an order_by or other
column-name based argument raises an exception.
- The polymorphic_identity kwarg in using_mapper_options is not overriden
anymore by the one generated by Elixir (patch from Ben Bangert).
- Moved the format of the multi-table inheritance column to a constant in
options (so that it can be changed globally).
- The foreign key constraint of the column in a multi-table inheritance is
configured with a cascade rule.
- A child entity doesn't inherit anymore its parent entity statements (such as
options) if it doesn't use any statement itself.
- Made inheritance work for custom base classes (closes #25).
- Fixed the inverse relationship matching when the inverse relationship is
defined in a parent Entity (thanks to Alexandre da Silva).
- Fixed bug in setup_entities (it always used the global entity list and not
the list given as argument).
- Fixed the versioning extension not appropriately handling versioned
entities with onupdate events (patch from Remi Jolin, closes #29).
- Fixed videostore example (patch from Jason R. Coombs, closes #31).
0.5.0 - 2007-12-08
Please see http://elixir.ematia.de/trac/wiki/Migrate04to05 for detailed
- Added set method on base Entity (set attributes using kwargs)
- Autosetup defaults to False ! (please look at those upgrade notes!)
- Polymorphic defaults to True (inheritance is polymorphic by default).
- Removed one of the autosetup triggers altogether: there is no "fake" mapper
registered in SQLAlchemy's mapper_registry anymore, so if you try to
access the class mapper directly (not through the 'mapper' attribute on
the class), before the setup phase happens, it won't work. This was done
because of a change in SQLAlchemy trunk (future SA 0.4.2) which broke that
piece of code (and prevented to use autosetup at all). Since that code
was a hack in the first place, instead of doing some even uglier hackery,
I got rid of it altogether.
- Moved some format strings to constants in options, so that one can change
them if he wants to.
- Allow overriding primary_key columns on autoloaded entities (closes tickets
#20 and #22)
- Columns created by ManyToOne relationships can now optionally (through
column_kwargs) *not* create an index (ie it's not harcoded anymore).
Suggestion by Jason R. Coombs.
- Fixed a nasty bug which prevented inheritance to work correctly when using
the attribute syntax in many cases.
- Fixed associable extension to work with SQLAlchemy trunk (future 0.4.2).
- Fixed an incompatibility with zope.interfaces.
- Tweaked the initialization sequence again (in fact revert an older change)
which prevented to reuse class properties of one class in other (subsequent)
- Fixed our tests to work with SA trunk (future 0.4.2) (unicode data + use of
0.4.0 - 2007-10-29
Please see http://elixir.ematia.de/trac/wiki/Migrate03to04 for detailed
- Implemented a new syntax to declare fields and relationships, much closer to
what is found in other Python ORM's. The with_fields syntax is now
deprecated in favor of a that new syntax. The old statement based (has_field
et al.) syntax is still available though (and will remain so for quite some
time). This was done with help from a patch by Adam Gomaa.
- Implemented polymorphic single-table inheritance as well as polymorphic and
non-polymorphic multi-table (aka joined table) inheritance.
- Added ext sub-package for additional Elixir statements.
- Added associable extension for generating polymorphic associations with
- Added versioning extension to keep track to all changes to your entities by
storing them in a secondary table.
- Added encryption extenstion to encrypt/decrypt some fields data on the fly
when writing to/reading from the database.
- Added support for synonym properties.
- Added shortcut syntax to define column_properties.
- Added a .query attribute on all entities. The old .query() syntax is still
- Added support to add any SQLAlchemy property on your mapper, through the
GenericProperty class (as well as the has_property statement). These can
work even if they rely on the entity columns (an thus need them to be
defined before the property can be declared). See tests/test_properties.py
- Added support for "manual session management" (ie you can now define an
entity with "using_options(session=None)" and it won't use any
SessionContext extension, nor receive the "query" attribute.
- Made the statement system more powerfull.
- The setup time was changed. That is the table and mapper are not created as
soon as the class is defined, but rather when first used, or when explicitly
calling the setup function (recommended). This also allowed us to reorder
the setup process and allows, among others to use a ManyToOne-generated
column as a primary key, to use unique constraints on those columns, to
order by those columns and so on...
- Made Elixir work with both SQLAlchemy 0.4 and 0.3.10 (with help from a patch
by Ants Aasma).
- Moved away from assign_mapper, now all assign_mapper-provided methods are on
the Entity class. Now, if people don't like them, they have the option to
simply provide another base class.
- Default objectstore is now a ScopedSession when working on SQLAlchemy 0.4.
It means that it's not wrapped in an Objectstore object at all. This means,
that depending on the version of SA you are using, you'll get a slightly
- Relationships to other classes can now also be defined using the classes
themselves in addition to the class namees. Obviously, this doesn't work for
- Classes defined inside a function can now have relationships to each other.
- Added default __init__ method on entities so that subclasses can override it
and still have the "set attribute by keyword" behavior by calling super()
- Added "through" and "via" keyword arguments on relationships and has_field
statement, to proxy values through relationships (uses association_proxy)
- Made EntityMeta public, so that people can actually define their own base
- Changed the order of relationship kwargs processing so that computed kwargs
can be overridden by kwargs manually passed to the statement. This should
only be used if you know what you are doing.
- Added onupdate kwarg to BelongsTo relationships for consistency with the
- Added ondelete and onupdate kwargs for use with has_and_belongs_to_many
to apply on delete clauses to foreign key constraints on the m2m table.
- Columns of the intermediary table of an has_and_belongs_to_many relationship
are now marked as primary keys.
- Reworked how entities look for primary keys on related entities. This
enables one "normal" entity (fully defined in Elixir) to refer to an entity
which is autoloaded.
- Added translation (from column name to column object) of the primary_key
mapper option so that it can actually be used. This allows to have entities
without any primary key defined at the table level.
- Added the possibility to give a custom name for ManyToOne constraints
(patch from and closes ticket #16)
- Dropped support for the old threadlocal SA extension (which doesn't even exist
anymore in SA 0.4)
- Reworked/cleaned tests so that they don't leak stuff to other tests (both at
the method level and module level) anymore. Uses nosetest's module level
- Fixed relationships to entities whose primary_key field has been defined
with a "key" argument (based on a patch by Paul Johnston).
- Fixed some buggy tests.
- Fixed relationships to tables using a schema (patch by Neil Blakey-Milner)
- Made inverse relationships use backrefs. This fixes the "bidirectional
coherency" problem some people had before doing a flush. (based on a patch
from Remi Jolin).
0.3.0 - 2007-03-27
- Made the provided metadata not threadlocal. This could break things for you
in some rare case. Please see the (newly created) FAQ file for details about
- Added support for autoloading/reflecting databases with
has_and_belongs_to_many relationships. The tablename argument is now
optional, but still recommended, otherwise you'll have to use the same exact
name for your intermediary table than the one generated. You also _have to_
specify at least one of either local_side or remote_side argument.
- Added support for the "version_id_col" option on entities. This option adds
a column to the table which will be used to prevent concurrent modifications
on any row of the entity's table (i.e. it will raise an error if it happens).
- Made the colname argument optional for belongs_to relationships in
autoloaded entities. It is only required to specify it if you have several
belongs_to relationships between two entities/tables.
- Applied patch from "Wavy" so that columns of a table are in the same order
as they were declared (this only works for the has_field statement).
- Applied patch from Isaac Csandl to add an "ondelete" argument to
belongs_to relationships. The content of that argument is forwarded to the
foreign key constraint.
- Foreign key names generated by belongs_to relationships use column names
instead of relation names in case we have a relation with the same name
defined in several entities inheriting from the same entity using single-
table inheritance (and we set a custom column name in one of them to avoid
a column-name conflict).
- Using invalid options on entities will now raise an exception
- Added __version__
- Use an explicit metaclass for entities, so that people can define their own
- Changed the approach to reflecting/autoloading belongs_to relationships.
This shouldn't change anything to how it's used but allowed me to factor
some code with has_and_belongs_to_many relationships.
- The tablename option can now be given a callable so that people can provide
their own function to get the table name for an entity. The tablename option
can now also be set globally (using the options_defaults dictionary). Of
course, this only makes sense for the callable usecase.
- Fixed bug preventing having entities without any statement.
- Fixed documentation for belongs_to relationships (the arguemnt is "required",
- Fixed typo which broke the use_alter argument on belongs_to relationships.
- Fixed inheritance unit test to pass SQLAlchemy type check on relations
(introduced in SA 0.3.6)
- Fixed wrong field length in autoload test (it was not noticeable with sqlite).
- Actually make the code python 2.3 compatible (Robin's patch was based on
0.1.0 while I had introduced more decorators in the trunk in the mean time).
- Made some PEP8 tweaks in many places. Used the pep8 script provided with
- Some cleanup/useless code removal
0.2.0 - 2007-02-28
- Applied patch from Robin Munn to make the code python 2.3 compatible
- Per a suggestion on the mailing list, look at the calling stack frame to
ensure that we apply statements to the proper class. We now attach the
statement list to the class itself, rather than attaching it to a global
list that is neither threadsafe, nor safe when doing nested class
definition. Also added a test to validate that this works.
- implemented singletable non-polymorphic inheritance
- added support to pass non-keyword arguments to tables. You just pass
them to the using_table_options statement and they will be forwarded to the
table along with the keyword arguments. This can be used to set table
- added support for deferred columns (use the "deferred" keyword argument on
- added a "required" keyword argument on fields and BelongsTo
relationships. This is the opposite of the "nullable" SA argument.
- added a "column_kwargs" keyword argument to BelongsTo relationships
to forward any keyword argument directly to the SA Column.
- added support for the use_alter and constraint_kwargs kwargs on BelongsTo
relationships (forwarded to SA ForeignKeyConstraint).
-> removed the systematic use_alter on BelongsTo relations since it
can now be specified only when needed.
-> removed it from HasAndBelongsToMany relations, since I think a
circular foreign key dependency can't happen with those relations.
- fixed foreign key names on MySQL (and possibly other) databases by
making sure the generated name is unique for the whole database, and not
only for the table on which it applies.
- corrected some docstrings
0.1.0 - 2007-02-12