Re: Query to match location transitions

sub3 napisal 14.11.2008 20:12:
> Hi,
>
> I was hoping someone could help me build a better query. I have a table of
> time/locations. Occasionally, we have multiple timestamps for the same
> location. I would like to remove those extra timestamps and only show the
> transition from one location to another. So...
>
> create table time_locations (
> id integer,
> timestamp double precision,
> location integer
> )

Try to not use data type name for column name.

> Data:
> 1,1197605841,1
> 2,1197608001,2
> 3,1197609802,2
> 4,1197611951,2
> 5,1199145360,2
> 6,1199145480,3
> 7,1199147280,3
> 8,1199149140,3
> 9,1199151300,1
> 10,1199152000,3
>
> I would like to return a table like:
> 1197605841,1,1197608001,2
> 1199145360,2,1199145480,3
> 1199149140,3,1199151300,1
> 1199151300,1,1199152000,3
>
> The only way I can think of to do this would be a procedure which would do a
> large loop over the
> table (sorted by time) returning a row when last.location <> this.location.
> However, when I try this on a
> large table, it seems like the 'select into' doesn't order & ruins the whole
> solution.

Select into is used for fetching single row result. You need rather
for..in loop

> Is there a query approach?

Example below gives the same result as described:

CREATE OR REPLACE FUNCTION location_changes(
last_time OUT double precision, last_location OUT integer,
new_time OUT double precision, new_location OUT integer
) RETURNS setof RECORD AS $$
BEGIN
for new_location, new_time in select location,timestamp
from time_locations order by timestamp
loop
if last_location<>new_location then
return next;
end if;
last_location=new_location;
last_time=new_time;
end loop;
END;
$$ language 'plpgsql';

select * from location_changes();

--
Regards,
Tomasz Myrta