How to alter column type from character varying to integer using sqlalchemy-migrate

Question

I'm using sqlalchemy-migrate to alter the type of one of the columns in a table in a Postgre SQL database. The upgrade script I'm using is:

# -*- cofing: utf-8 -*-

from sqlalchemy import MetaData, Table, Column, String, Integer
from migrate import changeset


metadata = MetaData()


def upgrade(migrate_engine):
    # ALTER TABLE courses ALTER COLUMN number SET DATA TYPE character varying;
    metadata.bind = migrate_engine
    courses = Table('courses', metadata, Column("number", Integer), extend_existing=True)
    courses.c.number.alter(type=String)


def downgrade(migrate_engine):
    # ALTER TABLE courses ALTER COLUMN number SET DATA TYPE integer;
    metadata.bind = migrate_engine
    courses = Table('courses', metadata, Column("number", String), extend_existing=True)
    courses.c.number.alter(type=Integer, cast='numeric')

The upgrade part seems to work but the downgrade always fails with the following error:

sqlalchemy.exc.ProgrammingError: (ProgrammingError) column "number" cannot be cast to type integer
'\nALTER TABLE courses ALTER COLUMN number TYPE INTEGER' {}

Now, if I were using plain SQL I could use ALTER TABLE courses ALTER COLUMN number TYPE INTEGER USING number::numeric to alter the column type back from character varying to integer, but I don't know how to achieve that using sqlalchemy-migrate.

Is there a way to force sqlalchemy to include USING number::numeric in the ALTER clause? or is there another way to avoid the error I posted above?

I appreciate your help.

Answer 1

Looks like sqlalchemy.migrate doesn't have support for rendering valid query in case of altering from String to Integer column types for postgresql.

In your case I would implement it as a direct query execution and move on.

def downgrade(migrate_engine):
    # ALTER TABLE courses ALTER COLUMN number SET DATA TYPE integer;
    migrate_engine.execute('ALTER TABLE courses ALTER COLUMN number TYPE INTEGER USING number::numeric')

BTW migrating from String to Integer may fail for different reasons - when column value would contain some value that cannot be converted to number. So I would add some extra validation into application logic to keep downgrade migration possible later on.