On Mon, Nov 8, 2010 at 3:21 PM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
> 2010/11/8 Merlin Moncure <mmoncure@gmail.com>:
>> On Mon, Nov 8, 2010 at 3:00 PM, Pavel Stehule <pavel.stehule@gmail.com> wrote:
>>>> Most cases of this feature are for dealing with new/old from trigger
>>>> function right? Why not build a complete new plan for each specific
>>>> trigger that invokes the function, along with some magic values like
>>>> (TG_FIELDNAMES -> text[]) that could be iterated for the mojo. Not
>>>> sure how you get direct type assignment to variable but it could
>>>> probably be worked out.
>>>
>>> if I understand well - it's not too far to my idea - just you create
>>> instance on function level? It is possible too. As disadvantages I
>>> see:
>>> a) you need some special syntax too
>>> b) there is overhead with multiple function call
>>> c) you have to manage some space for temporary values
>>
>> yes. If you need to deal with plan instance it should be at function
>> level IMO. There are other cases for this, search_path for example.
>> What overhead?
>
> you call a trigger body more times then once. The call of plpgsql
> isn't cheep. Main problem is missing a some working memory. Task:
> ensure so sum of fields must be less than some constant?
>
> What is solution in your design?
nothing specific. My main interest is in the plan management portion
(only because of side interest in search_path which is complete mess
at present). if plpgsql_compile knows that it is trigger function and
which table fired the trigger, you can generate specific hash key, and
take advantage of new/old being well defined inside function
execution. This seems less fiddly than trying to deal with plan
mechanics inside the function. My knowledge stops there -- I don't
have a real good understanding of how plpgsql works internally. If
this idea passes smell test maybe it merits more research.
merlin