Re: documentation fix for SET ROLE

Thanks for reviewing.

On 3/8/21 3:49 PM, David G. Johnston wrote:
> On Mon, Mar 8, 2021 at 4:41 PM David G. Johnston <david(dot)g(dot)johnston(at)gmail(dot)com> wrote:
>> You will want to add this to the commitfest if you haven't already.

Here is the commitfest entry: https://commitfest.postgresql.org/32/2993/

>> I would suggest adding a section titled "Identification" and placing these under that.

Good idea.

>> Reading it over it looks good. One point though: SET and SET ROLE are indeed "at run-time" (not 'run time'). ALTER ROLE and ALTER DATABASE should be considered "at connection-time" just like the command-line options.

Makes sense. I've applied these changes.

> Also, as a nearby email just reminded me, the determination of which role name is used to figure out default settings is the presented user name, not the one that would result from a connection-time role change as described here - though this should be tested, and then documented.

Yes, this seems to be correct. I've added a note about this behavior
in the patch.

-- setup
CREATE ROLE test1;
CREATE ROLE test2 WITH LOGIN IN ROLE test1;
ALTER ROLE test1 SET client_min_messages = 'error';
ALTER ROLE test2 SET client_min_messages = 'warning';
ALTER ROLE test2 SET ROLE = 'test1';

-- as test2
postgres=> SELECT CURRENT_USER, setting FROM pg_settings WHERE name = 'client_min_messages';
current_user | setting
--------------+---------
test1 | warning
(1 row)

Nathan