It turns out, however, that SQLAlchemy has a nice, pluggable database dialect system that is fairly simple to update. One part of this dialect system is a "SchemaDropper". So to cascade the DROP TABLE statements, I just created the following SchemaDropper (derived from the existing PostgreSQL PGSchemaDropper) and installed it as the default PostgreSQL dialect schemadropper. (Most of the code is copied from the base SchemaDropper class in sqlalchemy.sql.compiler)
from sqlalchemy.databases import postgres
def visit_table(self, table):
for column in table.columns:
if column.default is not None:
self.append("\nDROP TABLE " +
postgres.dialect.schemadropper = PGCascadeSchemaDropper
And that's it!