about client-side cursors

Hi,

I'd like to discuss about the concept of client-side cursor as defined
in psycopg. This is a concept I've always been quite uncomfortable with;
I'll try to explain why.

First, from a practical point of view:

   cur = connection.cursor()
   cur.execute(...)
   result = cur.fetchxxx()

This typical pattern seems awkward to me. Where do IO operations
actually take place? (on execute()? on fetch()? on both?)

Then, and besides the terminology, the fact that named cursors are very
different from client-side cursors while being constructed by the same
API is confusing. I understand that this may come from the DB-API:

  https://www.python.org/dev/peps/pep-0249/#cursor-objects
  > [Cursor] objects represent a database cursor. [...]

But this does not seem quite right to me since a client-side cursor is
actually not a PostgreSQL cursor. (The documention on psycopg2 is clear
on the difference, though.)

In asyncio context, while playing around with psycopg3, the situation is
even worse because of the additional async/await keywords; for instance:

   conn = await psycopg3.AsyncConnection.connect()
   async with conn:
       cur = await conn.execute("SELECT * FROM test")
       r = await cur.fetchall()

Why do we need two 'await' (i.e. two IO operations) on conn.execute()
and cur.fetchall() if 'cur' is a client-side cursor? (Back to my first
point about where IO happen, this appears to depend on whether the
connection has 'autocommit' or not...)

All in all, my main point is: (why) do we need client-side cursors?

Thanks,
Denis