Creating self-referential tables with polymorphism in SQLALchemy

Question

I'm trying to create a db structure in which I have many types of content entities, of which one, a Comment, can be attached to any other.

Consider the following:

from datetime import datetime
from sqlalchemy import create_engine
from sqlalchemy import Column, ForeignKey
from sqlalchemy import Unicode, Integer, DateTime
from sqlalchemy.orm import relation, backref
from sqlalchemy.ext.declarative import declarative_base

Base = declarative_base()

class Entity(Base):
    __tablename__ = 'entities'
    id = Column(Integer, primary_key=True)
    created_at = Column(DateTime, default=datetime.utcnow, nullable=False)
    edited_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow, nullable=False)
    type = Column(Unicode(20), nullable=False)
    __mapper_args__ = {'polymorphic_on': type}

# <...insert some models based on Entity...>

class Comment(Entity):
    __tablename__ = 'comments'
    __mapper_args__ = {'polymorphic_identity': u'comment'}
    id = Column(None, ForeignKey('entities.id'), primary_key=True)
    _idref = relation(Entity, foreign_keys=id, primaryjoin=id == Entity.id)
    attached_to_id = Column(Integer, ForeignKey('entities.id'), nullable=False)
    #attached_to = relation(Entity, remote_side=[Entity.id])
    attached_to = relation(Entity, foreign_keys=attached_to_id,
                           primaryjoin=attached_to_id == Entity.id,
                           backref=backref('comments', cascade="all, delete-orphan"))

    text = Column(Unicode(255), nullable=False)

engine = create_engine('sqlite://', echo=True)
Base.metadata.bind = engine
Base.metadata.create_all(engine)

This seems about right, except SQLAlchemy doesn't like having two foreign keys pointing to the same parent. It says ArgumentError: Can't determine join between 'entities' and 'comments'; tables have more than one foreign key constraint relationship between them. Please specify the 'onclause' of this join explicitly. ArgumentError: Can't determine join between 'entities' and 'comments'; tables have more than one foreign key constraint relationship between them. Please specify the 'onclause' of this join explicitly.

How do I specify onclause ?

Answer 1

Try to supplement your Comment.__mapper_args__ to:

__mapper_args__ = {
    'polymorphic_identity': 'comment',
    'inherit_condition': (id == Entity.id),
}

PS I haven't understand what you need _idref relationship for? If you want to use a comment somewhere where Entity is expected you could just pass the Comment instance as is.

Answer 2

To supplement @nailxx's answer: Repeating all that boilerplate is tedious if you have many dependent tables. Solution: move it all to a metaclass.

class EntityMeta(type(Entity)):
    def __init__(cls, name, bases, dct):
        ident = dct.get('_identity',None)
        if '__abstract__' not in dct:
            xid = Column(None, ForeignKey(Entity.id), primary_key=True)
            setattr(cls,'id',xid)
            setattr(cls,'__mapper_args__', { 'polymorphic_identity': dct['_identity'], 'inherit_condition': (xid == Entity.id,), 'primary_key':(xid,) })
            setattr(cls,'__tablename__',name.lower())
        super(EntityMeta, cls).__init__(name, bases, dct)

class EntityRef(Entity):
    __abstract__ = True
    __metaclass__ = EntityMeta

class Comment(EntityRef):
    _identity = 'comment'
    [...]

Variations, as an exercise for the reader: You can omit the _identity=… statement and use the class name, as in the setattr(cls,'__tablename__',…) line. Or not overwrite an existing __tablename__ or __mapper_args__ attribute.

Be sure to test dct and not cls for existence: the latter will find attributes in the parent class, which you obviously do not want at this point.