merge_result() may throw an exception with joined query results sets

from sqlalchemy/orm/loading.py in 0.8 or sqlalchemy/orm/query.py in 0.7

A query like:

session.query(Table_a, Table_b).outerjoin(Table_a.relation_table_b)

can produce result sets like:

[Table_a instance1>, None), (<Class Table_a instance2>, None), (<Class Table_a instance2>, <Class Table_b instance1>)]((<Class)

The 'None' types cause an exception in merge_result() when it tries to evaluate {{{attributes.instance_state(newrowi)}}} and newrowi is None

Line 129-136 of loading.py:

            for row in iterator:
                newrow = list(row)
                for i in mapped_entities:
                    newrow[i](i) = session._merge(
                            attributes.instance_state(newrow[i](i)),
                            attributes.instance_dict(newrow[i](i)),
                            load=load, _recursive={})
                result.append(util.KeyedTuple(newrow, keys))

To fix this, I added a check:

            for row in iterator:
                newrow = list(row)
                for i in mapped_entities:
                    if newrow[i](i) is not None:
                        newrow[i](i) = session._merge(
                            attributes.instance_state(newrow[i](i)),
                            attributes.instance_dict(newrow[i](i)),
                            load=load, _recursive={})
                result.append(util.KeyedTuple(newrow, keys))

In any case, it seems like a design accident that merge_result() works at all with tuples of class instances instead of a single class instance or list of rows; so, it may require a little more work to generalize merge_results() without any unintended consequences

Comments (5)