Re: [HACKERS] CTE inlining

On 5/2/17 4:44 PM, Andrew Dunstan wrote:
> 
> 
> On 05/02/2017 10:13 AM, Merlin Moncure wrote:
>> On Sun, Apr 30, 2017 at 6:21 PM, Andres Freund <andres@anarazel.de> wrote:
>>> On 2017-04-30 07:19:21 +0200, Pavel Stehule wrote:
>>>> why we cannot to introduce GUC option - enable_cteoptfence ?
>>> Doesn't really solve the issue, and we've generally shied away from GUCs
>>> that influence behaviour after a few bad experiences.  What if you want
>>> one CTE inlined, but another one not?
>> Yeah.  Are we absolutely opposed to SQL syntax against WITH that
>> allows or disallows fencing?   for example,
>>
>> WITH [MATERIALIZED]
>>
>> Pushing people to OFFSET 0 is a giant step backwards IMO, and as in
>> implementation detail is also subject to change.
>>
>>
> 
> Agreed, it's an ugly as sin and completely non-obvious hack.
> 

Isn't OFFSET 0 an implementation detail anyway? Who says the planner 
couldn't get smarter in the future, realize OFFSET 0 is no-op?

In that case replacing CTE optimization fence with "OFFSET 0" would be 
akin to painting yourself into a corner, waiting for the pain to dry, 
walking over to another corner and painting yourself into that one.

cheers

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services