Hi Ralf:

Ralf Schlatterbeck added the comment:

>I've now implemented in changeset 6a6b4651be1f.
>Changed tests test equivalence of results with both, client-side
>and server-side cursor in postgres.

In the initial post, you make a case for the connection name as a
static value.  Does that info have to be included as a comment in the
code somewhere so others won't be confused by a static name?

In back_postgresql.py the sql_new_cursor method has an argument
name='default'. But there is no code to make sure that it was called
with a real cursor name. Should there be a check and exception thrown
if name == 'default'?

Do we need code in configuration.py that enforces serverside_cursor =
no if you are not using postgres and if so, is 'yes' the right
default?

I think the call to self.db.sql_new_cursor(name='filter') in
rdbms_common.py is called for all backends: sqlite mysql and postgres
right? Why is there no need for:

   sql_new_cursor(self, name, conn, *args, *(kw))

for other rdbms backends besides postgres?

I see you are testing and comparing both server side and client side
cursors. Is it possible we could get different answers based on the
amount of data returned? That is the server says "hey I only have 4
rows returned, I'll push the 4 rows and kill the serve side
cursor?". So that server side cursor on a small data set is the same
as client side cursor?

I am pushing to CI now. Pleakse keep an eye on
https://travis-ci.org/github/roundup-tracker/roundup in case there are
issues.

Thanks for identify fixing this. Sadly bpo could probably use this but
since they are moving away from roundup ...