Is this really the implementation detail that we want to hide from the user? User can just run "select pg_my_temp_schema();" and see that there is no temp schema in the current session.
No ordinary user uses that function; it serves no everyday usage need.
Don't get me wrong - I can agree with that, but for now it seems odd to me... Steven Niu also mentioned this issue, but IMO we must give the most accurate description of the problem - tell "relation not found" only if we have temp namespace, but not specified relation in it.
"I want to give a better error message" is not a good enough reason to change this long-standing behavior in a back-patchable bug fix.
IOW, you don't get to change:
postgres=# select * from pg_temp.temp_table; ERROR: relation "pg_temp.temp_table" does not exist LINE 1: select * from pg_temp.temp_table;
to
postgres=# select * from pg_temp.tmptable; ERROR: pg_temp was specified but it contains no relations LINE 1: select * from pg_temp.tmptable;
In a released branch; and I do not agree that it is an improvement worth making in HEAD.
