Re: One-Shot Plans

On Tue, Jun 14, 2011 at 1:25 PM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
>
> We can work out the various paths through the traffic cop to see when
> a plan will be a "one-shot" - planned and then executed immediately,
> then discarded.
>
> In those cases we can take advantage of better optimisations. Most
> interestingly, we can evaluate stable functions at plan time, to allow
> us to handle partitioning and partial indexes better.
>
> Patch attached. Works...
>

this breaks test guc.c for me... specifically the test at
src/test/regress/sql/guc.sql circa line 226:
"""
set work_mem = '3MB';

-- but SET isn't
create or replace function myfunc(int) returns text as $$
begin
set work_mem = '2MB';
return current_setting('work_mem');
end $$
language plpgsql
set work_mem = '1MB';

select myfunc(0), current_setting('work_mem');
"""

regressions.diff
"""
*** 656,662 ****
select myfunc(0), current_setting('work_mem');
myfunc | current_setting
--------+-----------------
! 2MB | 2MB
(1 row)

set work_mem = '3MB';
--- 656,662 ----
select myfunc(0), current_setting('work_mem');
myfunc | current_setting
--------+-----------------
! 2MB | 3MB
(1 row)

set work_mem = '3MB';
"""

it seems that the effect of SET is being discarded

--
Jaime Casanova         www.2ndQuadrant.com
Professional PostgreSQL: Soporte 24x7 y capacitación