[Sqlalchemy-tickets] Issue #3427: SQLAlchemy 1.0 behavior difference regarding column default handling (zzzeek/sqlalchemy)

[Raphaël S. <iss...@bi...>]

New issue 3427: SQLAlchemy 1.0 behavior difference regarding column default handling
https://bitbucket.org/zzzeek/sqlalchemy/issue/3427/sqlalchemy-10-behavior-difference

Raphaël Slinckx:

Here is a test case yielding different results between sqlalchemy 0.9.9 and 1.0+

```python
import json

from sqlalchemy import create_engine, Column
from sqlalchemy.types import TypeDecorator, UnicodeText, Integer
from sqlalchemy.ext.mutable import MutableDict
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import Session

class JSONDictType(TypeDecorator):
	impl = UnicodeText

	def process_bind_param(self, value, dialect):
		if value is not None:
			value = unicode(json.dumps(value))
		return value

	def process_result_value(self, value, dialect):
		if value is not None:
			value = json.loads(value)
		return value

def JSON():
	return MutableDict.as_mutable(JSONDictType)

Base = declarative_base()

class MyDataClass(Base):
	__tablename__ = 'my_data'

	id   = Column(Integer, primary_key=True)
	data = Column(JSON(), default={}, nullable=False)

session = Session(create_engine("sqlite:///:memory:", echo=True))
Base.metadata.create_all(session.bind)

m1 = MyDataClass()
session.add(m1)
session.flush() # Note: without this, the m1.data is None on sqlalchemy 1.0
m1.data["test"] = 1
session.commit()

m2 = MyDataClass()
session.add(m2)
session.commit()

assert m1.data == {'test': 1}
assert m2.data == {}
# With sqlalchemy 1.0 this results in
# m1.data == {}
# m2.data == {'test': 1}
```

As you can see the default={} seemed to work previously (also the m1.data['test'] = 1 worked without the session.flush() just above)

With the latest version, the dict provided as default seems to be used and re-used resulting in a process-wide globally shared default dict which can lead to catastrophic leaks in a webapp for example. Also the object needs to be flushed before m1.data is populated.

Maybe this usage was not expected or even discouraged, but I didn't find anything in the docs about this. The closest thing I identified is the change discussed here: http://docs.sqlalchemy.org/en/latest/changelog/migration_10.html#changes-to-attribute-events-and-other-operations-regarding-attributes-that-have-no-pre-existing-value and possibly the new default handling.

What is the expected behavior as of 1.0 ? And how do I get the old behavior using the latest version ?

Responsible: zzzeek