Re: optimizing immutable vs. stable function calls?

On Mon, Jan 23, 2017 at 9:10 AM, Jim Nasby <Jim(dot)Nasby(at)bluetreble(dot)com> wrote:
> On 1/18/17 6:09 PM, David G. Johnston wrote:
>>
>> That would not be a productive exercise for me, or most people who just
>> want
>> some idea of what to expect in terms of behavior when they write and use a
>> Stable function (Immutable and Volatile seem fairly easy to reason about).
>
>
> Yeah, this isn't an uncommon question for users to have, and "read the code"
> isn't a great answer.
>
> If there's a README or comment block that describes this, that might be a
> reasonable compromise.
>
> It would certainly be useful to document how to push the planner in the
> right direction as well. I didn't realize that SELECT ... (SELECT
> slow_stable_function()) was a thing until reading this thread.

Totally agree.

There are other odd cases as well, mostly relating to SQL inlining
(for example, marking a function IMMUTABLE can cause it to fall out of
inlining you would get by giving no designation). If you documented
all the rules, I think you'd find the rules are a bit odd and could be
simplified.

merlin