Re: Types and SRF's

Hi,

I am trying to get my feet wet in SRF's

I had to define a type in order to get my first attempt
at a srf for an sql language function ie..

create type annual_report_type as
( category text,
jan numeric(9,2),
feb numeric(9,2),
mar numeric(9,2),
apr numeric(9,2),
may numeric(9,2),
jun numeric(9,2),
jul numeric(9,2),
aug numeric(9,2),
sep numeric(9,2),
oct numeric(9,2),
nov numeric(9,2),
dec numeric(9,2) ,
total numeric(9,2) )

and then use this type as

create or replace function annual_report(integer) returns setof
annual_report_type
as '
select a.category, (select sum(amount) from all_accounts where
category=a.category and
extract (month from date) = 1 and extract (year from date) = $1) as
jan,
(select sum(amount) from all_accounts where category=a.category and
extract (month from date) = 2 and extract (year from date) = $1) as
feb,
...
...
(select sum(amount) from all_accounts where category=a.category and
extract (year from date) = $1) as total
from all_accounts a
group by category
order by category
' language sql

The above seems to be working fine...
I would feel a bit more comfortable if I could recover the definition
of the type at a later time, I cannot seem to find the definition of
the type in pg_type (there is an entry but the definition does not seem
to
be visible).

It does not seem possible to replace "annual_report_type" in the
function
definition with just the type...All of the placements fail for me in
any case.

Any suggestions as to how I can remember the rowtype? or (embed the
definiton
of the type in the definition of the function without having to create
an explicit type?

Jerry

Jerry LeVan <jerry(dot)levan(at)eku(dot)edu> writes:
> I would feel a bit more comfortable if I could recover the definition
> of the type at a later time,

Try "\d annual_report_type" in psql.

regards, tom lane

Doh, I was using \dT and \dT+....

Thanks

Jerry
On Aug 31, 2004, at 3:14 PM, Tom Lane wrote:

> Jerry LeVan <jerry(dot)levan(at)eku(dot)edu> writes:
>> I would feel a bit more comfortable if I could recover the definition
>> of the type at a later time,
>
> Try "\d annual_report_type" in psql.
>
> regards, tom lane
>

Your query looks suspiciously complicated...
Why not process all 12 months in one shot with something like this :
- only one subquery
- no join
- date between can make an index scan on date

select category, sum(amount) as sum_amount, extract (month from date) as
month
from all_accounts where (date between beginning of the year and end of
the year)
group by category,month order by category,month )

Not what you wanted but probably massively faster.

Or you can do this (in approximate SQL):

create type annual_report_type as
( sums numeric(9,2)[12] );

create type my_type as ( month integer, amount numeric );

CREATE AGGREGATE my_sum
takes one input which is my_type and sums the amount into the month column
of annual_report_type

Then :
select category, my_sum( my_type(month,amount) as report, extract (month
from date) as month
from all_accounts where (date between beginning of the year and end of
the year)
group by category,month order by category,month )

Dunno if this would work, it would be nice I think.

Thank you for the response Pierre,

select category, sum(amount) as sum_amount, extract (month from date)
as month
from all_accounts where (extract(year from date)=2003)
group by category,month order by category,month

is certainly much faster than what I am doing but as you pointed out,
I want the table to have a column for each month ( and a grand total
as the last column).

I have not used arrays and aggregates, I will take a look....

Jerry

On Sep 1, 2004, at 3:03 AM, Pierre-Frédéric Caillaud wrote:

>
> Your query looks suspiciously complicated...
> Why not process all 12 months in one shot with something like this :
> - only one subquery
> - no join
> - date between can make an index scan on date
>
> select category, sum(amount) as sum_amount, extract (month from date)
> as month
> from all_accounts where (date between beginning of the year and end
> of the year)
> group by category,month order by category,month )
>
> Not what you wanted but probably massively faster.
>
> Or you can do this (in approximate SQL):
>
> create type annual_report_type as
> ( sums numeric(9,2)[12] );
>
> create type my_type as ( month integer, amount numeric );
>
> CREATE AGGREGATE my_sum
> takes one input which is my_type and sums the amount into the month
> column of annual_report_type
>
> Then :
> select category, my_sum( my_type(month,amount) as report, extract
> (month from date) as month
> from all_accounts where (date between beginning of the year and end
> of the year)
> group by category,month order by category,month )
>
> Dunno if this would work, it would be nice I think.
>