2010/11/18 Robert Haas <robertmhaas@gmail.com>:
> On Thu, Nov 18, 2010 at 10:24 AM, Merlin Moncure <mmoncure@gmail.com> wrote:
>> On Thu, Nov 18, 2010 at 12:47 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>> Merlin Moncure <mmoncure@gmail.com> writes:
>>>> On Wed, Nov 17, 2010 at 7:08 PM, Jaime Casanova <jaime@2ndquadrant.com> wrote:
>>>>> i will start the review of this one... but before that sorry for
>>>>> suggesting this a bit later but about using UNNEST as part of the
>>>>> sintax?
>>>
>>>> Does for-in-array do what unnset does?
>>>
>>> Yes, which begs the question of why bother at all. AFAICS this patch
>>> simply allows you to replace
>>>
>>> for x in select unnest(array_value) loop
>>>
>>> with
>>>
>>> for x in unnest array_value loop
>>>
>>> (plus or minus a parenthesis or so). I do not think we need to add a
>>> bunch of code and create even more syntactic ambiguity (FOR loops are
>>> already on the hairy edge of unparsability) to save people from writing
>>> "select".
>>
>> Pavel's performance argument is imnsho valid. arrays at present are
>> the best way to pass data around functions and any optimizations here
>> are very welcome. Given that, is there any way to mitigate your
>> concerns on the syntax side?
>
> Can we get the performance benefit any other way? I hate to think
> that it will still be slow for people using the already-supported
> syntax.
>
I afraid so other ways are more difficult with deeper impacts on
plpgsql executor.
what is a slow:
a) repeated detoasting - access with subscripts - maybe detoasted
values can be cached?
b) evaluation of SRF expression - maybe call of SRF function can be
simple expression,
c) faster evaluation ro query
The most important is @a. Only a few people uses a FOR IN SELECT
UNNEST form now. Probably not optimizable on PLpgSQL level is form FOR
IN SELECT * FROM UNNEST.
FOR IN ARRAY doesn't impacts on expression executing - this patch is
just simple and isolated.
Pavel
> --
> Robert Haas
> EnterpriseDB: http://www.enterprisedb.com
> The Enterprise PostgreSQL Company
>