Skip to content

Latest commit

 

History

History
1416 lines (1132 loc) · 49.6 KB

CHANGELOG.rst

File metadata and controls

1416 lines (1132 loc) · 49.6 KB

Changelog

.. rst-class:: emphasize-children

0.21

0.21.6

Fixed

  • Fix bug in pydantic_model_creator when a foreign key is not included in include param. (#1430)
  • Fix bug in contrib.sanic.register_tortoise causing a deadlock when using asyncpg and > 1 workers (#1696)
  • Open psycopg pool with .open() to remove deprecated warning (#1697)
  • Fix bug in bulk_update when pk field is not id (#1698)
  • Fix mysql uuid compression bug (#1687)
  • Fix comment for fk fields without constraint for mysql (#1679)
  • Removed no_delay option for postgres, as it wasn't doing anything (#1677)

0.21.5 <../0.21.5>`_ - 2024-07-18

Added

  • Propagate _create_db parameter to RegisterTortoise. (#1676)

0.21.4 <../0.21.4>`_ - 2024-07-03

Added

  • Add ObjectDoesNotExistError to show better 404 message. (#759)
  • DoesNotExist and MultipleObjectsReturned support 'Type[Model]' argument. (#742)(#1650)
  • Add argument use_tz and timezone to RegisterTortoise. (#1649)
  • Support await tortoise.contrib.fastapi.RegisterTortoise. (#1662)
  • Add tortoise.contrib.test.init_memory_sqlite. (#1657)

Fixed

  • Fix update_or_create errors when field value changed. (#1584)
  • Fix bandit check error (#1643)
  • Fix potential race condition in ConnectionWrapper (#1656)
  • Fix py312 warning for datetime.utcnow (#1661)
  • Fix reusing values and value_list queries (#780)

Changed

  • Remove obsolete loop._selector from contrib/test. (#659)(#1636)

0.21.3 - 2024-06-01

Fixed

  • Fix bulk_update when using source_field for pk (#1633)

0.21.2 - 2024-05-25

Added

  • Add create_unique_index argument to M2M field and default if it is true (#1620)

0.21.1 - 2024-05-24

Fixed

  • Fix error on using old style pk=True

0.21.0 - 2024-05-23

Added

  • Enhancement for FastAPI lifespan support (#1371)
  • Add __eq__ method to Q to more easily test dynamically-built queries (#1506)
  • Added PlainToTsQuery function for postgres (#1347)
  • Allow field's default keyword to be async function (#1498)
  • Add support for queryset slicing. (#1341)

Fixed

  • Fix DatetimeField use '__year' report 'int' object has no attribute 'utcoffset'. (#1575)
  • Fix bulk_update when using custom fields. (#1564)
  • Fix optional parameter in pydantic_model_creator does not work for pydantic v2. (#1551)
  • Fix get_annotations now evaluates annotations in the default scope instead of the app namespace. (#1552)
  • Fix get_or_create method. (#1404)
  • Use index_name instead of BaseSchemaGenerator._generate_index_name to generate index name.
  • Use subquery for count() and exists() in QuerySet to match count result to QuerySet result. (#1607)

Changed

  • Change utils.chunk from function to return iterables lazily.
  • Removed lower bound of id keys in generated pydantic models. (#1602)
  • Rename Field initial arguments pk/index to primary_key/db_index. (#1621)
  • Renamed Model.check method to Model._check to avoid naming collision issues (#1559) (#1550)

Breaking Changes

  • bulk_create now does not return anything. (#1614)

0.20

0.20.1

Added

  • Add binary compression support for UUIDField in MySQL. (#1458)
  • Only Model, Tortoise, BaseDBAsyncClient, __version__, and connections are now exported from tortoise
  • Add parameter validators to pydantic_model_creator. (#1471)

Fixed

  • Fix order of fields in ValuesListQuery when it has more than 10 fields. (#1492)
  • Fix pydantic v2 pydantic_model_creator nullable field not optional. (#1454)
  • Fix pydantic v2.5 unittest error. (#1535)
  • Fix pydantic_model_creator exclude_readonly parameter not working.
  • Fix annotation propagation for non-filter queries. (#1590)
  • Fix blacksheep example unittest error. (#1534)

0.20.0

Added

  • Allow ForeignKeyField(on_delete=NO_ACTION) (#1393)
  • Support pydantic 2.0. (#1433)

Fixed

  • Fix foreign key constraint not generated on MSSQL Server. (#1400)
  • Fix testcase error with python3.11 (#1308)

Breaking Changes

  • Drop support for pydantic 1.x.
  • Drop support for python 3.7.
  • Param config_class of pydantic_model_creator is renamed to model_config.
  • Attr config_class of PydanticMeta is renamed to model_config.

0.19

0.19.3

Added

  • Added config_class option to pydantic model genator that allows the developer to customize the generated pydantic model's Config class. (#1048)

Fixed

  • Fastapi example test not working. (#1029)
  • Fix create index sql error. (#1202)
  • Fix dependencies resolve error. (#1246)
  • Fix ignoring zero value of limit. (#1270)
  • Fix ForeignKeyField is none when fk is integer 0. (#1274)
  • Fix limit ignore zero. (#1270)
  • Fix min/max value validators for decimal fields. (#1291)

0.19.2

Added

  • Added schema attribute to Model's Meta to specify exact schema to use with the model.

Fixed

  • Mixin does not work. (#1133)
  • using_db wrong position in model shortcut methods. (#1150)
  • Fixed connection to Oracle database by adding database info to DBQ in connection string.
  • Fixed ORA-01435 error while using Oracle database (#1155)
  • Fixed processing of ssl option in MySQL connection string.
  • Fixed type hinting for QuerySetSingle.

0.19.1

Added

  • Added Postgres/SQLite partial indexes support. (#1103)
  • Added Microsoft SQL Server/Oracle support, powered by asyncodbc, note that which is not fully tested.
  • Added optional parameter to pydantic_model_creator. (#770)
  • Added using_db parameter to Model shortcut methods. (#1109)

Fixed

  • TimeField for MySQL will return datetime.timedelta object instead of datetime.time object.
  • Fix on conflict do nothing. (#1122)
  • Fix _custom_generated_pk attribute not set in Model._init_from_db method. (#633)

0.19.0

Added

  • Added psycopg backend support.
  • Added a new unified and robust connection management interface to access DB connections which includes support for lazy connection creation and much more. For more details, check out this PR
  • Added TimeField. (#1054)
  • Added ArrayField.

Fixed

  • Fix bulk_create doesn't work correctly with more than 1 update_fields. (#1046)
  • Fix bulk_update errors when setting null for a smallint column on postgres. (#1086)

Deprecated

  • Existing connection management interface and related public APIs which are deprecated:
  • Tortoise.get_connection
  • Tortoise.close_connections

Changed

  • Refactored tortoise.transactions.get_connection method to tortoise.transactions._get_connection.
Note that this method has now been marked private to this module and is not part of the public API

0.18.1

Added

  • Add on conflict do update for bulk_create. (#1024)

Fixed

  • Fix bulk_create error. (#1012)
  • Fix unittest invalid.
  • Fix bulk_update in postgres with some type. (#968) (#1022)

0.18.0

Added

  • Add Case-When support. (#943)
  • Add Rand/Random function in contrib. (#944)
  • Add ON CONFLICT support in INSERT statements. (#428)

Fixed

  • Fix bulk_update error when pk is uuid. (#986)
  • Fix mutable default value. (#969)

Changed

  • Move Function, Aggregate from functions.py to expressions.py. (#943)
  • Move Q from query_utils.py to expressions.py.
  • Replace python-rapidjson to orjson.

Removed

  • Remove asynctest and use unittest.IsolatedAsyncioTestCase. (#416)
  • Remove py37 support in tests.
  • Remove green and nose2 test runner.

0.17

0.17.8

Added

  • Add Model.raw method to support the raw sql query.
  • Add QuerySet.bulk_update method. (#924)
  • Add QuerySet.in_bulk method.
  • Add MaxValueValidator and MinValueValidator (#927)

Fixed

  • Fix QuerySet subclass being lost when _clone is run on the instance.
  • Fix bug in .values with source_field. (#844)
  • Fix contrib.blacksheep exception handlers, use builtin json response. (#914)
  • Fix Indexes defined in Meta class do not make use of exists parameter in their template (#928)

Changed

  • Allow negative values with IntEnumField. (#889)
  • Make .values() and .values_list() awaited return more consistent. (#899)

0.17.7

  • Fix select_related behaviour for forward relation. (#825)
  • Fix bug in nested QuerySet and Manager. (#864)
  • Add Concat function for MySQL/PostgreSQL. (#873)
  • Patch for use_index/force_index mutable problem when making query. (#888)
  • Lift annotation field's priority in make query. (#883)
  • Make use/force index available in select type Query. (#893)
  • Fix all logging to use Tortoise's logger instead of root logger. (#879)
  • Rename db_client logger to tortoise.db_client.
  • Add indexes to Model.describe.

0.17.6

  • Add RawSQL expression.
  • Fix columns count with annotations in _make_query. (#776)
  • Make functions nested. (#828)
  • Add db_constraint in field describe.

0.17.5

  • Set field_type of fk and o2o same to which relation field type. (#443)
  • Fix error sql for .sql() call more than once. (#796)
  • Fix incorrect splitting of the import route when using Router (#798)
  • Fix filter error after annotate with F. (#806)
  • Fix select_related for reverse relation. (#808)

0.17.4

  • Fix update_or_create. (#782)
  • Add contains, contained_by and filter to JSONField

0.17.3

  • Fix duplicates when using custom through association class on M2M relations
  • Fix update_or_create and get_or_create. (#721)
  • Fix refresh_from_db without fields pass. (#734)
  • Make update query work with limit and order_by. (#748)
  • Add Subquery expression. (#756) (#9) (#337)
  • Use JSON in JSONField.

0.17.2

  • Add more index types.
  • Add force_index, use_index to queryset.
  • Fix F in update error with update_fields.
  • Make delete query work with limit and order_by. (#697)
  • Filter backward FK fields with IS NULL and NOT IS NULL filters (#700)
  • Add select_for_update in update_or_create. (#702)
  • Add Model.select_for_update.
  • Add __search full text search to queryset.

0.17.1

  • Fix type for modules.
  • Fix select_related when related model specified more than once. (#679)
  • Add __iter__ to model, now can just return model/models in fastapi response.
  • Fix in_transaction bug caused by 'router'. (#677) (#678)

0.17.0

  • Add date part extract filtering.
  • Add Manager support.
  • Add db router support.
  • Add nowait, skip_locked, of parameters to queryset.select_for_update.
  • Add field name to validation exceptions.
  • Compatible with asyncmy.
  • Replace pypika to pypika-tortoise.

0.16

0.16.21

  • Fixed validating JSON before decoding. (#623)
  • Add model method update_or_create.
  • Add batch_size parameter for bulk_create method.
  • Fix save with F expression and field with source_field.

0.16.20

  • Add model field validators.
  • Allow function results in group by. (#608)

0.16.19

  • Replace set TZ environment variable to TIMEZONE to avoid affecting global timezone.
  • Allow passing module objects to models_paths param of Tortoise.init_models(). (#561)
  • Implement PydanticMeta.backward_relations. (#536)
  • Allow overriding PydanticMeta in PydanticModelCreator. (#536)
  • Fixed make_native typo to make_naive in timezone module

0.16.18

  • Support custom function in update. (#537)
  • Add Model.refresh_from_db. (#549)
  • Add timezone support, be careful to upgrade to this version, see docs for details. (#335)
  • Remove aerich in case of cyclic dependency. (#558)

0.16.17

  • Add on_delete in ManyToManyField. (#508)
  • Support F expression in annotate. (#475)
  • Fix QuerySet.select_related in case of join same table twice. (#525)
  • Integrate Aerich into the install. (#530)

0.16.16

  • Fixed inconsistency in integrity error exception of FastAPI
  • add OSError to _get_comments except block

0.16.15

  • Make DateField accept valid date str.
  • Add QuerySet.select_for_update().
  • check default for not None on pydantic model creation
  • propagate default to pydantic model
  • Add QuerySet.select_related().
  • Add custom attribute name for Prefetch instruction.
  • Add db_constraint for RelationalField family.

0.16.14

  • Make F expression work with QuerySet.filter().
  • Include py.typed in source distribution.
  • Added datetime parsing from int for fields.DatetimeField.
  • get_or_create passes the using_db= on if provided.
  • Allow custom loop and connection_class parameters to be passed on to asyncpg.

0.16.13

  • Default install of tortoise-orm now installs with no C-dependencies, if you want to use the C accelerators, please do a pip install tortoise-orm[accel] instead.
  • Added <instance>.clone() method that will create a cloned instance in memory. To persist it you still need to call .save()
  • .clone() will raise a ParamsError if tortoise can't generate a primary key. In that case do a .clone(pk=<newval>)
  • If manually setting the primary key value to None and the primary key can be automatically generated, this will create a new record. We however still recommend the .clone() method instead.
  • .save() can be forced to do a create by setting force_create=True
  • .save() can be forced to do an update by setting force_update=True
  • Setting update_fields for a .save() operation will strongly prefer to do an update if possible

0.16.12

  • Make Field.default effect on db level when generate table
  • Add converters instead of importing from pymysql
  • Fix postgres BooleanField default value convent
  • Fix JSONField typed in pydantic_model_creator
  • Add .sql() method on QuerySet

0.16.11

  • fix: sqlite://:memory: in Windows thrown OSError: [WinError 123]
  • Support bulk_create() insertion of records with overridden primary key when the primary key is DB-generated
  • Add queryset.exists() and Model.exists().
  • Add model subscription lookup, Model[<pkval>] that will return the object or raise KeyError

0.16.10

  • Fix bad import of basestring
  • Better handling of NULL characters in strings. Fixes SQLite, raises better error for PostgreSQL.
  • Support .group_by() with join now

0.16.9

  • Support F expression in .save() now
  • IntEnumField accept valid int value and CharEnumField accept valid str value
  • Pydantic models get created with globally unique identifier
  • Leaf-detection to minimize duplicate Pydantic model creation
  • Pydantic models with a Primary Key that is also a raw field of a relation is now not hidden when exclude_raw_fields=True as it is a critically important field
  • Raise an informative error when a field is set as nullable and primary key at the same time
  • Foreign key id's are now described to have the positive-integer range of the field it is related to
  • Fixed prefetching over OneToOne relations
  • Fixed __contains for non-text fields (e.g. JSONB)

0.16.8

  • Allow Q expression to function with _filter parameter on aggregations
  • Add manual .group_by() support
  • Fixed regression where GROUP BY class is missing for an aggregate with a specified order.

0.16.7

  • Added preliminary support for Python 3.9
  • TruncationTestCase now properly quotes table names when it clears them out.
  • Add model signals support
  • Added app_label to test initializer(...) and TORTOISE_TEST_APP as test environment variable.

0.16.6

Warning

This is a security fix release. We recommend everyone update.

Security fixes

  • Fixed SQL injection issue in MySQL
  • Fixed SQL injection issues in MySQL when using contains, starts_with or ends_with filters (and their case-insensitive counterparts)
  • Fixed malformed SQL for PostgreSQL and SQLite when using contains, starts_with or ends_with filters (and their case-insensitive counterparts)

Other changes

  • Added support for partial models:

    To create a partial model, one can do a .only(<fieldnames-as-strings>) as part of the QuerySet. This will create model instances that only have those values fetched.

    Persisting changes on the model is allowed only when:

    • All the fields you want to update is specified in <model>.save(update_fields=[...])
    • You included the Model primary key in the .only(...)

    To protect against common mistakes we ensure that errors get raised:

    • If you access a field that is not specified, you will get an AttributeError.
    • If you do a <model>.save() a IncompleteInstanceError will be raised as the model is, as requested, incomplete.
    • If you do a <model>.save(update_fields=[...]) and you didn't include the primary key in the .only(...), then IncompleteInstanceError will be raised indicating that updates can't be done without the primary key being known.
    • If you do a <model>.save(update_fields=[...]) and one of the fields in update_fields was not in the .only(...), then IncompleteInstanceError as that field is not available to be updated.
  • Fixed bad SQL generation when doing a .values() query over a Foreign Key
  • Added <model>.update_from_dict({...}) that will mass update values safely from a dictionary
  • Fixed processing URL encoded password in connection string

0.16.5

  • Moved Tortoise.describe_model(<MODEL>, ...) to <MODEL>.describe(...)
  • Deprecated Tortoise.describe_model()
  • Fix for generate_schemas param being ignored in tortoise.contrib.quart.register_tortoise
  • Fix join query with source_field param

0.16.4

  • More consistent escaping of db columns, fixes using SQL reserved keywords as field names with a function.
  • Fix the aggregates using the wrong side of the join when doing a self-referential aggregation.
  • Fix F functions wrapped forgetting about distinct=True

0.16.3

  • Fixed invalid var IN () SQL generated using __in= and __not_in filters.
  • Fix bug with order_by on nested fields
  • Fix joining with self by reverse-foreign-key for filtering and annotation

0.16.2

  • Default values() & values_list() now includes annotations.
  • Annotations over joins now work correctly with values() & values_list()
  • Ensure GROUP BY precedes HAVING to ensure that filtering by aggregates work correctly.
  • Fix bug with join query with aggregation
  • Cast BooleanField values correctly on SQLite & MySQL

0.16.1

  • QuerySetSingle now has better code completion

  • Created Pydantic models will now have the basic validation elements:

    • required is correctly populated for required fields
    • nullable is added to the schema where nulls are accepted
    • maxLength for CharFields
    • minimum & maximum values for integer fields

    To get Pydantic to handle nullable/default fields correctly one should do a **user.dict(exclude_unset=True) when passing values to a Model class.

  • Added FastAPI helper that is based on the starlette helper but optionally adds helpers to catch and report with proper error DoesNotExist and IntegrityError Tortoise exceptions.

  • Allows a Pydantic model to exclude all read-only fields by setting exclude_readonly=True when calling pydantic_model_creator.

  • a Tortoise PydanticModel now provides two extra helper functions:

    • from_queryset: Returns a List[PydanticModel] which is the format that e.g. FastAPI expects
    • from_queryset_single: allows one to avoid calling await multiple times to get the object and all its related items.

0.16.0

Caution!

This release drops support for Python 3.6:

Tortoise ORM now requires a minimum of CPython 3.7

New features:

  • Model docstrings and #: comments directly preceding Field definitions are now used as docstrings and DDL descriptions.

    This is now cleaned and carried as part of the docstring parameter in describe_model(...)

    If one doesn't explicitly specify a Field description= or Model Meta.table_description= then we default to the first line as the description. This is done because a description is submitted to the DB, and needs to be short (depending on DB, 63 chars) in size.

    Usage example:

    class Something(Model):
        """
        A Docstring.
    
        Some extra info.
        """
    
        # A regular comment
        name = fields.CharField(max_length=50)
        #: A docstring comment
        chars = fields.CharField(max_length=50, description="Some chars")
        #: A docstring comment
        #: Some more detail
        blip = fields.CharField(max_length=50)
    
    # When looking at the describe model:
    {
        "description": "A Docstring.",
        "docstring": "A Docstring.\n\nSome extra info.",
        ...
        "data_fields": [
            {
                "name": "name",
                ...
                "description": null,
                "docstring": null
            },
            {
                "name": "chars",
                ...
                "description": "Some chars",
                "docstring": "A docstring comment"
            },
            {
                "name": "blip",
                ...
                "description": "A docstring comment",
                "docstring": "A docstring comment\nSome more detail"
            }
        ]
    }
  • Early Partial Init of models.

    We now have an early init of models, which can be useful when needing Models that are not bound to a DB, but otherwise complete. e.g. Schema generation without needing to be properly set up.

    Usage example:

    # Lets say you defined your models in "some/models.py", and "other/ddef.py"
    # And you are going to use them in the "model" namespace:
    Tortoise.init_models(["some.models", "other.ddef"], "models")
    
    # Now the models will have relationships built, so introspection of schema will be comprehensive
  • Pydantic serialisation.

    We now include native support for automatically building a Pydantic model from Tortoise ORM models. This will correctly model:

    • Data Fields
    • Relationships (FK/O2O/M2M)
    • Callables

    At this stage we only support serialisation, not deserialisation.

    For mode information, please see :ref:`contrib_pydantic`

  • Allow usage of F expressions to in annotations. (#301)
  • Now negative number with limit(...) and offset(...) raise ParamsError. (#306)
  • Allow usage of Function to queryset.update(). (#308)
  • Add ability to supply distinct flag to Aggregate (#312)

Bugfixes:

  • Fix default type of JSONField

Removals:

  • Removed tortoise.aggregation as this was deprecated since 0.14.0

  • Removed start_transaction as it has been broken since 0.15.0

  • Removed support for Python 3.6 / PyPy-3.6, as it has been broken since 0.15.0

    If you still need Python 3.6 support, you can install tortoise-orm<0.16 as we will still backport critical bugfixes to the 0.15 branch for a while.

.. rst-class:: emphasize-children

0.15

0.15.24

  • Fixed regression where GROUP BY class is missing for an aggregate with a specified order.

0.15.23

  • Fixed SQL injection issue in MySQL
  • Fixed SQL injection issues in MySQL when using contains, starts_with or ends_with filters (and their case-insensitive counterparts)
  • Fixed malformed SQL for PostgreSQL and SQLite when using contains, starts_with or ends_with filters (and their case-insensitive counterparts)

0.15.22

  • Fix the aggregates using the wrong side of the join when doing a self-referential aggregation.
  • Fix for generate_schemas param being ignored in tortoise.contrib.quart.register_tortoise

0.15.21

  • Fixed invalid var IN () SQL generated using __in= and __not_in filters.
  • Fix bug with order_by on nested fields
  • Fix joining with self by reverse-foreign-key for filtering and annotation

0.15.20

  • Default values() & values_list() now includes annotations.
  • Annotations over joins now work correctly with values() & values_list()
  • Ensure GROUP BY precedes HAVING to ensure that filtering by aggregates work correctly.
  • Cast BooleanField values correctly on SQLite & MySQL

0.15.19

  • Fix Function with source_field option. (#311)

0.15.18

  • Install on Windows does not require a C compiler any more.
  • Fix IntegrityError with unique field and get_or_create

0.15.17

  • Now get_or_none(...), classmethod of Model class, works in the same way as queryset

0.15.16

  • get_or_none(...) now raises MultipleObjectsReturned if multiple object fetched. (#298)

0.15.15

  • Add ability to suppply a to_field= parameter for FK/O2O to a non-PK but still uniquely indexed remote field. (#287)

0.15.14

  • add F expression support in queryset.update() - This allows for atomic updates of data in the database. (#294)

0.15.13

  • Applies default ordering on related queries
  • Fix post-ManyToMany related queries not being evaluated correctly
  • Ordering is now preserved on ManyToMany related fetches
  • Fix aggregate function on joined table to use correct primary key
  • Fix filtering by backwards FK to use correct primary key

0.15.12

  • Added range filter to support between and syntax

0.15.11

  • Added ordering option for model Meta class to apply default ordering

0.15.10

  • Bumped requirements to cater for newer feature use (#282)

0.15.9

  • Alias Foreign Key joins as we can have both self-referencing and duplicate joins to the same table. This generates SQL that differentiates between which instance of the table to work with.

0.15.8

  • TextField now recommends usage of CharField if wanting unique indexing instead of just saying "indexing not supported"
  • .count() now honours offset and limit
  • Testing un-awaited ForeignKeyField as a boolean expression will automatically resolve as False if it is None
  • Awaiting a nullable ForeignKeyField won't touch the DB if it is None

0.15.7

  • QuerySet.Update() now returns the count of the no of rows affected. Note, that
  • QuerySet.Delete() now returns the count of the no of rows deleted.
  • Note that internal API of db_connection.execute_query() now returns rows_affected, results. (This is informational only)
  • Added get_or_none(...) as syntactic sugar for filter(...).first()

0.15.6

  • Added BinaryField for storing binary objects (bytes).
  • Changed TextField to use LONGTEXT for MySQL to allow for larger than 64KB of text.
  • De-duplicate index if specified on both index=true and as part of indexes
  • Primary Keyed TextField is marked as deprecated. We can't guarantee that it will be properly indexed or unique in all cases.
  • One can now disable the backwards relation for FK/O2O relations by passing related_name=False
  • One can now pass a PK value to a generated field, and Tortoise ORM will use that as the PK as expected. This allows one to have a model that has a autonumber PK, but setting it explicitly if required.

0.15.5

  • Refactored Fields:

    Fields have been refactored, for better maintenance. There should be no change for most users.

    • More accurate auto-completion.
    • Fields now contain their own SQL schema by dialect, which significantly simplifies adding field types.
    • describe_model() now returns the DB type, and dialect overrides.
  • JSONField will now automatically use python-rapidjson as an accelerator if it is available.
  • DecimalField and aggregations on it, now behaves much more like expected on SQLite (#256)
  • Check whether charset name is valid for the MySQL connection (#261)
  • Default DB driver parameters are now applied consistently, if you use the URI schema or manual.

0.15.4

  • Don't generate a schema if there is no models (#254)
  • Emit a RuntimeWarning when a module has no models to import (#254)
  • Allow passing in a custom SSL context (#255)

0.15.3

  • Added OneToOneField implementation:

    OneToOneField describes a one to one relation between two models. It can be set from the primary side only, but resolved from both sides in the same way.

    describe_model(...) now also reports OneToOne relations in both directions.

    Usage example:

    event: fields.OneToOneRelation[Event] = fields.OneToOneField(
        "models.Event", on_delete=fields.CASCADE, related_name="address"
    )
  • Prefetching is done concurrently now, sending all prefetch requests at the same time instead of in sequence.
  • Enable foreign key enforcement on SQLite for builds where it was optional.

0.15.2

  • The auto_now_add argument of DatetimeField is handled correctly in the SQLite backend.
  • unique_together now creates named constrains, to prevent the DB from auto-assigning a potentially non-unique constraint name.
  • Filtering by an auto_now field doesn't replace the filter value with now() anymore.

0.15.1

  • Handle OR'ing a blank Q() correctly (#240)

0.15.0

New features:

  • Pooling has been implemented, allowing for multiple concurrent databases and all the benefits that comes with it.
    • Enabled by default for databases that support it (mysql and postgres) with a minimum pool size of 1, and a maximum of 5
    • Not supported by sqlite
    • Can be changed by passing the minsize and maxsize connection parameters
  • Many small performance tweaks:
    • Overhead of query generation has been reduced by about 6%
    • Bulk inserts are ensured to be wrapped in a transaction for >50% speedup
    • PostgreSQL prepared queries now use a LRU cache for significant >2x speedup on inserts/updates/deletes
  • DateField & DatetimeField deserializes faster on PostgreSQL & MySQL.
  • Optimized .values() to do less copying, resulting in a slight speedup.
  • One can now pass kwargs and Q() objects as parameters to Q() objects simultaneously.

Bugfixes:

  • indexes will correctly map the foreign key if referenced by name.
  • Setting DB generated PK in constructor/create generates exception instead of silently being ignored.

Deprecations:

  • start_transaction is deprecated, please use @atomic() or async with in_transaction(): instead.

  • This release brings with it, deprecation of Python 3.6 / PyPy-3.6:

    This is due to small differences with how the backported aiocontextvars behaves in comparison to the built-in in Python 3.7+.

    There is a known context confusion, specifically regarding nested transactions.

.. rst-class:: emphasize-children

0.14

0.14.2

  • A Field name of alias is now no longer reserved.
  • Restored support for inheriting from Abstract classes. Order is now also deterministic, with the inherited classes' fields being placed before the current.

0.14.1

  • ManyToManyField is now a function that has the type of the relation for autocomplete, this allows for better type hinting at less effort.
  • Added section on adding better autocomplete for relations in editors.

0.14.0

Caution!

This release drops support of Python 3.5:

Tortoise ORM now requires a minimum of CPython 3.6 or PyPy3.6-7.1

Enhancements:

  • Models, Fields & QuerySets have significant type annotation improvements, leading to better IDE integration and more comprehensive static analysis.
  • Fetching records from the DB is now up to 25% faster.
  • Database functions Trim(), Length(), Coalesce(), Lower(), Upper() added to tortoise.functions module.
  • Annotations can be selected inside Queryset.values() and Queryset.values_list() expressions.
  • Added support for Python 3.8
  • The Foreign Key property is now await-able as long as one didn't populate it via .prefetch_related()
  • One can now specify compound indexes in the Meta: class using indexes. It works just like unique_together.

Bugfixes:

  • The generated index name now has significantly lower chance of collision.
  • The compiled SQL query contains HAVING and GROUP BY only for aggregation functions.
  • Fields for FK relations are quoted properly.
  • Fields are quoted properly in UNIQUE statements.
  • Fields are quoted properly in KEY statements.
  • Comment Fields are quoted properly in PostgreSQL dialect.
  • unique_together will correctly map the foreign key if referenced by name.

Deprecations:

  • import from tortoise.aggregation is deprecated, please do import from tortoise.functions instead.

Breaking Changes:

  • The hash used to make generated indexes unique has changed. The old algorithm had a very high chance of collisions, the new hash algorithm is much better in this regard.
  • Dropped support for Python 3.5
.. rst-class:: emphasize-children

0.13

0.13.12

  • Reverted "The Field class now calls super().__init__, so mixins are properly initialised." as it was causing issues on Python 3.6.

0.13.11

  • Fixed the _FieldMeta class not to checking if the 1st base class was Field, so would break with mixins.
  • The Field class now calls super().__init__, so mixins are properly initialised.

0.13.10

  • Names ForeignKey constraints in a consistent way

0.13.9

  • Fields can have 2nd base class which makes IDEs know python type (str, int, datetime...) of the field.
  • The type parameter of Field.__init__ is removed, instead we use the 2nd base class
  • Foreign keys and indexes are now defined correctly in MySQL so that they take effect as expected
  • MySQL now doesn't warn of unsafe index creation anymore

0.13.8

  • Fixed bug in schema creation for MySQL where non-int PK did not get declared properly (#195)

0.13.7

  • iexact filter modifier was implemented. Queries like «queryset».filter(name__iexact=...) will perform case-insensitive search.

0.13.6

  • Fix minor bug in Model.__init__ where we raise the wrong error on setting RFK/M2M values directly.
  • Fields in Queryset.values_list() is now in the defined Model order.
  • Fields in Queryset.values() is now in the defined Model order.

0.13.5

  • Sample Starlette integration
  • Relational fields are now lazily constructed via properties instead of in the constructor, this results in a significant overhead reduction for Model instantiation with many relationships.

0.13.4

  • Assigning to the FK field will correctly set the associated db-field
  • Reading a nullalble FK field can now be None
  • Nullalble FK fields reverse-FK is now also nullable
  • Deleting a nullable FK field sets it to None

0.13.3

  • Fixed installing Tortoise-ORM in non-unicode systems. (#180)
  • «queryset».update(…) now correctly uses the DB-specific to_db_value()
  • fetch_related(…) now correctly encodes non-integer keys.
  • ForeignKey fields of type UUIDField are now escaped consistently.
  • Pre-generated ForeignKey fields (e.g. UUIDField) is now checked for persistence correctly.
  • Duplicate M2M .add(…) now checks using consistent field encoding.
  • source_field Fields are now handled correctly for ordering.
  • source_field Fields are now handled correctly for updating.

0.13.2

  • Security fixes for «model».save() & «model».delete():

    This is now fully parametrized, and these operations are no longer susceptible to escaping issues.

  • Performance improvements:

    • Simple update is now ~3-6× faster
    • Partial update is now ~3× faster
    • Delete is now ~2.7x faster
  • Fix generated Schema Primary Key for BigIntField for MySQL and PostgreSQL.
  • Added support for using a SmallIntField as a auto-gen Primary Key.
  • Ensure that default PK is added to the top of the attrs.

0.13.1

  • Fix in generating comments for Foreign Keys in MySQL
  • Added schema support for PostgreSQL. Either set "schema": "custom" var in credentials or as a query parameter ?schema=custom
  • Default MySQL charset to utf8mb4. If a charset is provided it will also force the TABLE charset to the same.

0.13.0

Warning

This release brings with it, deprecation of Python 3.5:

We will increase the minimum supported version of Python to 3.6, as 3.5 is reaching end-of-life, and is missing many useful features for async applications.

We will discontinue Python 3.5 support on the next major release (Likely 0.14.0)

New Features:

  • Example Sanic integration along with register_tortoise hook in contrib (#163)
  • .values() and .values_list() now default to all fields if none are specified.
  • generate_schema() now generates well-formatted DDL SQL statements.
  • Added TruncationTestCase testing class that truncates tables to allow faster testing of transactions.
  • Partial saves are now supported (#157): obj.save(update_fields=['model','field','names'])

Bugfixes:

  • Fixed state leak between database drivers which could cause incorrect DDL generation.
  • Fixed missing table/column comment generation for ForeignKeyField and ManyToManyField
  • Fixed comment generation to escape properly for SQLite
  • Fixed comment generation for PostgreSQL to not duplicate comments
  • Fixed generation of schema for fields that defined custom source_field values defined
  • Fixed working with Models that have fields with custom source_field values defined
  • Fixed safe creation of M2M tables for MySQL dialect (#168)

Docs/examples:

  • Examples have been reworked:
    • Simplified init of many examples
    • Re-did generate_schema.py example
    • A new relations_recirsive.py example (turned into test case)
  • Lots of small documentation cleanups
.. rst-class:: emphasize-children

0.12

0.12.7 (retracted)

  • Support connecting to PostgreSQL via Unix domain socket (simple case).
  • Self-referential Foreign and Many-to-Many keys are now allowed

0.12.6 / 0.12.8

  • Handle a __models__ variable within modules to override the model discovery mechanism.

    If you define the __models__ variable in yourapp.models (or wherever you specify to load your models from), generate_schema() will use that list, rather than automatically finding all models for you.

  • Split model constructor into from-Python and from-DB paths, leading to 15-25% speedup for large fetch operations.
  • More efficient queryset manipulation, 5-30% speedup for small fetches.

0.12.5

  • Using non registered models or wrong references causes an ConfigurationError with a helpful message.

0.12.4

  • Inherit fields from Mixins, together with abstract model classes.

0.12.3

  • Added description attribute to Field class. (#124)
  • Added the ability to leverage field description from (#124) to generate table column comments and ability to add table level comments

0.12.2

  • Fix accidental double order-by for .values() based queries. (#143)

0.12.1

  • Bulk insert operation:

    Note

    The bulk insert operation will do the minimum to ensure that the object created in the DB has all the defaults and generated fields set, this may result in incomplete references in Python.

    e.g. IntField primary keys will not be populated.

    This is recommend only for throw away inserts where you want to ensure optimal insert performance.

    User.bulk_create([
        User(name="...", email="..."),
        User(name="...", email="...")
    ])
  • Notable efficiency improvement for regular inserts

0.12.0

  • Tortoise ORM now supports non-autonumber primary keys.

    Note

    This is a big feature change. It should not break any existing implementations.

    That primary key will be accessible through a reserved field pk which will be an alias of whichever field has been nominated as a primary key. That alias field can be used as a field name when doing filtering e.g. .filter(pk=...) etc…

    We currently support single (non-composite) primary keys of any indexable field type, but only these field types are recommended:

    IntField
    BigIntField
    CharField
    UUIDField

    One must define a primary key by setting a pk parameter to True.

    If you don't define a primary key, we will create a primary key of type IntField with name of id for you.

    Any of these are valid primary key definitions in a Model:

    id = fields.IntField(pk=True)
    
    checksum = fields.CharField(pk=True)
    
    guid = fields.UUIDField(pk=True)
.. rst-class:: emphasize-children

0.11

0.11.13

  • Fixed connection retry to work with transactions
  • Added broader PostgreSQL connection failure detection

0.11.12

  • Added automatic PostgreSQL connection retry

0.11.11

  • Extra parameters now get passed through to the MySQL & PostgreSQL drivers

0.11.10

  • Fixed SQLite handling of DatetimeField

0.11.9

  • Code has been reformatted using black, and minor code cleanups (#120 #123)
  • Sample Quart integration (#121)
  • Better isolation of connection handling — Allows more dynamic connections so we can do pooling & reconnections.
  • Added automatic MySQL connection retry

0.11.8

  • Fixed .count() when a join happens (#109)

0.11.7

  • Fixed unique_together for foreign keys (#114)
  • Fixed Field.to_db_value method to handle Enum (#113 #115 #116)

0.11.6

  • Added ability to use unique_together meta Model option

0.11.5

  • Fixed concurrency isolation when attempting to do multiple concurrent operations on a single connection.

0.11.4

  • Fixed several convenience issues with foreign relations:
    • FIXED: .all() actually returns the _query property as was documented.
    • New models with FK don't automatically fail to resolve any data. They can now be evaluated lazily.
  • Some DB's don't support OFFSET without Limit, added caps to signal workaround, which is to automatically add limit of 1000000
  • Pylint plugin to know about default related_name for ForeignKey fields.
  • Simplified capabilities to be static, and defined at class level.

0.11.3

  • Added basic DB driver Capabilities.

    Test runner now has the ability to skip tests conditionally, based on the DB driver Capabilities:

    @requireCapability(dialect='sqlite')
    async def test_run_sqlite_only(self):
        ...
  • Added per-field indexes.

    When setting index=True on a field, Tortoise will now generate an index for it.

    Note

    Due to MySQL limitation of not supporting conditional index creation, if safe=True (the default) is set, it won't create the index and emit a warning about it.

    We plan to work around this limitation in a future release.

  • Performance fix with PyPika for small fetch queries
  • Remove parameter hack now that PyPika support Parametrized queries
  • Fix typos in JSONField docstring
  • Added .explain() method on QuerySet.
  • Add required read-only property to fields

0.11.2

  • Added "safe" schema generation
  • Correctly convert values to their db representation when using the "in" filter
  • Added some common missing field types:
    • BigIntField
    • TimeDeltaField
  • BigIntField can also be used as a primary key field.

0.11.1

  • Test class isolation fixes & contextvars update
  • Turned on autocommit for MySQL
  • db_url now supports defaults and casting parameters to the right types

0.11.0

  • Added .exclude() method for QuerySet
  • Q objects can now be negated for NOT query (~Q(...))
  • Support subclassing on existing fields
  • Numerous bug fixes
  • Removed known broken connection pooling
.. rst-class:: emphasize-children

0.10

0.10.11

  • Pre-build some query & filters statically, 15-30% speed up for smaller queries.
  • Required field params are now positional, so Python and IDE linters will pick up on it easier.
  • Filtering also applies DB-specific transforms, Fixes #62
  • Fixed recursion error on m2m management with big lists

0.10.10

  • Refactor Tortoise.init() and test runner to not re-create connections per test, so now tests pass when using an SQLite in-memory database
  • Can pass event loop to test initializer function: initializer(loop=loop)
  • Fix relative URI for SQLite
  • Better error message for invalid filter param.
  • Better error messages for missing/bad field params.
  • nose2 plugin
  • Test utilities compatible with py.test

0.10.9

  • Uses macros on SQLite driver to minimise syncronisation. aiosqlite>=0.7.0
  • Uses prepared statements for insert, large insert performance increase.
  • Pre-generate base pypika query object per model, providing general purpose speedup.

0.10.8

  • Performance fixes from pypika>=0.15.6
  • Significant reduction in object creation time

0.10.7

  • Fixed SQLite relative db path and :memory: now also works
  • Removed confusing error message for missing db driver dependency
  • Added aiosqlite as a required dependency
  • execute_script() now annotates errors just like execute_query(), to reduce confusion
  • Bumped aiosqlite>=0.6.0 for performance fix
  • Added tortoise.run_async() helper function to make smaller scripts easier to run. It cleans up connections automatically.
  • SQLite does autocommit by default.

0.10.6

  • Fixed atomic decorator to get connection only on function call

0.10.5

  • Fixed pre-init queryset objects creation

0.10.4

  • Added support for running separate transactions in multidb config

0.10.3

  • Changed default app label from 'models' to None
  • Fixed ConfigurationError message for wrong connection name

0.10.2

  • Set single_connection to True by default, as there is known issues with connection pooling
  • Updated documentation

0.10.1

  • Fixed M2M manager methods to correctly work with transactions
  • Fixed mutating of queryset on select queries

0.10.0

  • Refactored Tortoise.init() to init all connections and discover models from config passed as argument.

    Caution!

    This is a breaking change.

    You no longer need to import the models module for discovery, instead you need to provide an app ⇒ modules map with the init call:

    async def init():
        # Here we create a SQLite DB using file "db.sqlite3"
        #  also specify the app name of "models"
        #  which contain models from "app.models"
        await Tortoise.init(
            db_url='sqlite://db.sqlite3',
            modules={'models': ['app.models']}
        )
        # Generate the schema
        await Tortoise.generate_schemas()

    For more info, please have a look at :ref:`init_app`

  • New transactions module for implicit working with transactions
  • Test frameworks overhauled: - Better performance for test runner, using transactions to keep tests isolated. - Now depends on an initializer() and finalizer() to set up and tear down DB state.
  • Exceptions have been further clarified
  • Support for CPython 3.7
  • Added support for MySQL/MariaDB
.. rst-class:: emphasize-children

0.9 & older

0.9.4

  • No more asserts, only Tortoise Exceptions
  • Fixed PyLint plugin to work with pylint>=2.0.0
  • Formalised unittest classes & documented them.
  • __slots__ where it was easy to do. (Changes class instances from dicts into tuples, memory savings)

0.9.3

  • Fixed backward incompatibility for Python 3.7

0.9.2

  • JSONField is now promoted to a standard field.
  • Fixed DecimalField and BooleanField to work as expected on SQLite.
  • Added FloatField.
  • Minimum supported version of PostgreSQL is 9.4
  • Added .get(...) shortcut on query set.
  • values() and values_list() now converts field values to python types

0.9.1

  • Fixed through parameter honouring for ManyToManyField

0.9.0

  • Added support for nested queries for values and values_list:

    result = await Event.filter(id=event.id).values('id', 'name', tournament='tournament__name')
    result = await Event.filter(id=event.id).values_list('id', 'participants__name')
  • Fixed DatetimeField and DateField to work as expected on SQLite.
  • Added PyLint plugin.
  • Added test class to mange DB state for testing isolation.

0.8.0

  • Added PostgreSQL JSONField

0.7.0

  • Added .annotate() method and basic aggregation funcs

0.6.0

  • Added Prefetch object

0.5.0

  • Added contains and other filter modifiers.
  • Field kwarg default now accepts functions.

0.4.0

  • Immutable QuerySet. unique flag for fields

0.3.0

  • Added schema generation and more options for fields:

    from tortoise import Tortoise
    from tortoise.backends.sqlite.client import SqliteClient
    from tortoise.utils import generate_schema
    
    client = SqliteClient(db_name)
    await client.create_connection()
    Tortoise.init(client)
    await generate_schema(client)

0.2.0

  • Added filtering and ordering by related models fields:

    await Tournament.filter(
        events__name__in=['1', '3']
    ).order_by('-events__participants__name').distinct()