sqlalchemy-json

SQLAlchemy-JSON provides mutation-tracked JSON types to SQLAlchemy:

• MutableJson is a straightforward implementation for keeping track of top-level changes to JSON objects;
• NestedMutableJson is an extension of this which tracks changes even when these happen in nested objects or arrays (Python dicts and lists).

Examples

Basic change tracking

This is essentially the SQLAlchemy mutable JSON recipe. We define a simple author model which list the author's name and a property handles for various social media handles used:

class Author(Base):
    name = Column(Text)
    handles = Column(MutableJson)

Or, using the declarative mapping style:

class Category(Base):
    __tablename__ = "categories"

    id = mapped_column(Integer, primary_key=True)
    created_at: Mapped[DateTime] = mapped_column(DateTime, default=datetime.now)
    updated_at: Mapped[DateTime] = mapped_column(
        DateTime, default=datetime.now, onupdate=datetime.now
    )
    keywords: Mapped[list[str]] = mapped_column(MutableJson)

The example below loads one of the existing authors and retrieves the mapping of social media handles. The error in the twitter handle is then corrected and committed. The change is detected by SQLAlchemy and the appropriate UPDATE statement is generated.

>>> author = session.query(Author).first()
>>> author.handles
{'twitter': '@JohnDoe', 'facebook': 'JohnDoe'}
>>> author.handles['twitter'] = '@JDoe'
>>> session.commit()
>>> author.handles
{'twitter': '@JDoe', 'facebook': 'JohnDoe'}

Nested change tracking

The example below defines a simple model for articles. One of the properties on this model is a mutable JSON structure called references which includes a count of links that the article contains, grouped by domain:

class Article(Base):
    author = Column(ForeignKey('author.name'))
    content = Column(Text)
    references = Column(NestedMutableJson)

With this in place, an existing article is loaded and its current references inspected. Following that, the count for one of these is increased by ten, and the session is committed:

>>> article = session.query(Article).first()
>>> article.references
{'github.com': {'edelooff/sqlalchemy-json': 4, 'zzzeek/sqlalchemy': 7}}
>>> article.references['github.com']['edelooff/sqlalchemy-json'] += 10
>>> session.commit()
>>> article.references
{'github.com': {'edelooff/sqlalchemy-json': 14, 'zzzeek/sqlalchemy': 7}}

Had the articles model used MutableJson like in the previous example this code would have failed. This is because the top level dictionary is never altered directly. The nested mutable ensures the change happening at the lower level bubbles up to the outermost container.

Non-native JSON / other serialization types

By default, sqlalchemy-json uses the JSON column type provided by SQLAlchemy (specifically sqlalchemy.types.JSON.) If you wish to use another type (e.g. PostgreSQL's JSONB), your database does not natively support JSON (e.g. versions of SQLite before 3.37.2/), or you wish to serialize to a format other than JSON, you'll need to provide a different backing type.

This is done by using the utility function mutable_json_type. This type creator function accepts two parameters:

• dbtype controls the database type used. This can be an existing type provided by SQLAlchemy or SQLALchemy-utils, or an augmented type to provide serialization to any other format;
• nested controls whether the created type is made mutable based on MutableDict or NestedMutable (defaults to False for MutableDict).

import json

from sqlalchemy import JSON, String, TypeDecorator
from sqlalchemy.dialects.postgresql import JSONB
from sqlalchemy_json import mutable_json_type

class JsonString(TypeDecorator):
    """Enables JSON storage by encoding and decoding on the fly."""

    impl = String

    def process_bind_param(self, value, dialect):
        return json.dumps(value)

    def process_result_value(self, value, dialect):
        return json.loads(value)


postgres_jsonb_mutable = mutable_json_type(dbtype=JSONB)
string_backed_nested_mutable = mutable_json_type(dbtype=JsonString, nested=True)

Dependencies

• sqlalchemy

Development

Here's how to setup your development environment:

python -m venv .venv
. .venv/bin/activate
pip install -e ".[dev]"
# run tests
pytest

Changelog

0.7.0

• Adds support for top-level list for MutableJson, rather than having that support only be available in the nested variant (edelooff#51)
• Adds pytest as development dependency

0.6.0

• Fixes pickling support (edelooff#36)
• Drops python 2.x support (previously claimed, but already broken for some time)
• Removes test runners for CPython 3.6 since Github actions support has been dropped

0.5.0

• Fixes a lingering Python 3 compatibility issue (cmp parameter for TrackedList.sort)
• Adds pickling and unpickling support (edelooff#28)
• Adds tracking for dictionary in-place updates (edelooff#33)

0.4.0

• Adds a type creation function to allow for custom or alternate serialization types. This allows for a way around the regression in SQLite compatibility introduced by v0.3.0.

0.3.0

• Switches JSON base type to sqlalchemy.types.JSON from deprecated JSON type provided by SQLAlchemy-utils.

0.2.2

• Fixes a bug where assigning None to the column resulted in an error (edelooff#10)

0.2.1

• Fixes a typo in the README found after uploading 0.2.0 to PyPI.

0.2.0 (unreleased)

• Now uses JSONType provided by SQLAlchemy-utils to handle backend storage;
• Backwards incompatible: Changed class name JsonObject to MutableJson and NestedJsonObject to NestedMutableJson
• Outermost container for NestedMutableJson can now be an array (Python list)

0.1.0 (unreleased)

Initial version. This initially carried a 1.0.0 version number but has never been released on PyPI.