connect() ignores cursor_factory if connection_factory is also specified

[dandanlen]

Hi there,

I tried setting the connection_factory and cursor_factory args through the psycopg2.connect method, but it looks like the cursor factory isn't being used. As a result, rows are returned as tuples instead of dicts.

When you use the cursor_factory in isolation it works fine:

with psycopg2.connect(dsn,
                      cursor_factory=DictCursor) as conn:
    with conn.cursor() as cur:
        print(cur.row_factory)
        # prints: <class 'psycopg2.extras.DictRow'>

However if you also set a connection_factory (Am I using this wrong? 🤷‍♂ ):

with psycopg2.connect(dsn,
                      connection_factory=LoggingConnection,
                      cursor_factory=DictCursor) as conn:
    conn.initialize(logging.Logger('test'))
    with conn.cursor() as cur:
        print(cur.row_factory)
        # prints: None

I can work around it by specifying the cursor_factory when I create the cursor.

with psycopg2.connect(dsn, connection_factory=LoggingConnection) as conn:
    conn.initialize(logging.Logger('test'))
    with conn.cursor(cursor_factory=DictCursor) as cur:
        print(cur.row_factory)
        # prints: <class 'psycopg2.extras.DictRow'>

I'm using version 2.8.4, Python 3.7.4 on mac.

[dvarrazzo]

It's a contrived use case, but a genuine bug, affecting pretty much all the provided connection subclasses.

[dandanlen]

Well, it was easy to work around but I wouldn't call it a contrived use case... Anyway, thanks for the quick response!

[dvarrazzo]

The fact nobody stumbled in it in 15 years the adapter is used in all the world makes it a pretty unlikely use case.

[danielfoguelman]

The fact nobody stumbled in it in 15 years the adapter is used in all the world makes it a pretty unlikely use case.

I stumbled upon it. Do you still think it is an unlikely use case?

[dvarrazzo]

Considering that the bug was closed 3 years ago, yes, pretty unlikely.

[danielfoguelman]

I would say, if you find an unlikely event to happen twice in a finite time interval, it is not as unlikely as you thought. But well, the reporter of this issue was kind enough to provide a way to overcome this issue.

[dvarrazzo]

@danielfoguelman I don't understand: this issue was closed in 2019 in 2.8.5. Are you using an older version than that, and complaining, or have you found the same issue in a more recent version?

[danielfoguelman]

It is true that I'm using an older version and complaining, Danielle. Sorry about that. From this ticket history I thought that it was closed as a not-fix and it amused me to think that two persons could find an unlikely bug. Now I see you merged the fix in 2.9.5 version.
Again, I didn't ment to be rude, I thought wrong about the whole thing. Have a nice week!