Thread: Function: Change data while walking through records
Hi,
I keep having this need to create a function that will change the row data
as I am walking through the data. For example, I process each row in order,
if column1 change from previous row, set column2 to true.
Is this possible? I can run another query to modify the data, but that
doesn't seem that efficient?
Thanks for any help.
--
An example:
--Simple table
create table tmp1 (
id integer,
time timestamp,
sequential boolean);
insert into tmp1 values (1, '2008-01-01 12:00:00', false),(2, '2008-01-02
12:00:00', false),(3, '2008-01-03 12:00:00', false),(4, '2008-01-08
12:00:00', false),(5, '2008-01-09 12:00:00', false),(6, '2008-01-10
12:00:00', false),(7, '2008-01-15 12:00:00', false),(8, '2008-01-22
12:00:00', false),(9, '2008-01-23 12:00:00', false),(10, '2008-01-25
12:00:00', false);
-- I want line: "my_sequential := true;" to effect that table
CREATE OR REPLACE FUNCTION setseq() RETURNS INTEGER AS
$BODY$
DECLARE
my_id integer;
my_time integer;
my_sequential boolean;
old_time integer;
change_count integer;
BEGIN
change_count := 1;
for my_id,my_time,my_sequential in
select id,extract(day from time),sequential from tmp1 order by time
LOOP
if (old_time is not null) then
if (old_time+1 = my_time) then
my_sequential := true; --How do I make this work?
change_count := change_count+1;
end if;
end if;
old_time := my_time;
END LOOP;
return change_count;
END;
$BODY$
LANGUAGE 'plpgsql';
--
View this message in context:
http://www.nabble.com/Function%3A-Change-data-while-walking-through-records-tp23873836p23873836.html
Sent from the PostgreSQL - general mailing list archive at Nabble.com.
stevesub wrote:
> I keep having this need to create a function that will change the row data
> as I am walking through the data. For example, I process each row in order,
> if column1 change from previous row, set column2 to true.
>
> Is this possible? I can run another query to modify the data, but that
> doesn't seem that efficient?
>
> Thanks for any help.
>
> --
>
> An example:
>
> --Simple table
> create table tmp1 (
> id integer,
> time timestamp,
> sequential boolean);
>
> insert into tmp1 values (1, '2008-01-01 12:00:00', false),(2,
> '2008-01-02
> 12:00:00', false),(3, '2008-01-03 12:00:00', false),(4, '2008-01-08
> 12:00:00', false),(5, '2008-01-09 12:00:00', false),(6, '2008-01-10
> 12:00:00', false),(7, '2008-01-15 12:00:00', false),(8, '2008-01-22
> 12:00:00', false),(9, '2008-01-23 12:00:00', false),(10, '2008-01-25
> 12:00:00', false);
>
> -- I want line: "my_sequential := true;" to effect that table
> CREATE OR REPLACE FUNCTION setseq() RETURNS INTEGER AS
> $BODY$
> DECLARE
> my_id integer;
> my_time integer;
> my_sequential boolean;
> old_time integer;
> change_count integer;
> BEGIN
> change_count := 1;
>
> for my_id,my_time,my_sequential in
> select id,extract(day from time),sequential from tmp1 order by time
> LOOP
> if (old_time is not null) then
> if (old_time+1 = my_time) then
> my_sequential := true; --How do I make this work?
> change_count := change_count+1;
> end if;
> end if;
> old_time := my_time;
> END LOOP;
> return change_count;
> END;
> $BODY$
> LANGUAGE 'plpgsql';
You could use a cursor and UPDATE ... WHERE CURRENT OF for that.
Here are some lines of code:
DECLARE
...
mycur CURSOR FOR select id,extract(day from time),sequential from tmp1 order by time FOR UPDATE;
row tmp1%ROWTYPE;
BEGIN
...
OPEN mycur;
LOOP
FETCH mycur INTO row;
EXIT WHEN NOT FOUND;
...
UPDATE tmp1 SET sequential = TRUE WHERE CURRENT OF mycur;
...
END LOOP;
CLOSE mycur;
...
END;
Yours,
Laurenz Albe
Albe Laurenz *EXTERN* wrote: > > stevesub wrote: >> I keep having this need to create a function that will change the row >> data >> as I am walking through the data. For example, I process each row in >> order, >> if column1 change from previous row, set column2 to true. >> >> Is this possible? I can run another query to modify the data, but that >> doesn't seem that efficient? >> >> Thanks for any help. >> >> -- >> >> An example: >> >> --Simple table >> create table tmp1 ( >> id integer, >> time timestamp, >> sequential boolean); >> >> insert into tmp1 values (1, '2008-01-01 12:00:00', false),(2, >> '2008-01-02 >> 12:00:00', false),(3, '2008-01-03 12:00:00', false),(4, '2008-01-08 >> 12:00:00', false),(5, '2008-01-09 12:00:00', false),(6, '2008-01-10 >> 12:00:00', false),(7, '2008-01-15 12:00:00', false),(8, '2008-01-22 >> 12:00:00', false),(9, '2008-01-23 12:00:00', false),(10, '2008-01-25 >> 12:00:00', false); >> >> -- I want line: "my_sequential := true;" to effect that table >> CREATE OR REPLACE FUNCTION setseq() RETURNS INTEGER AS >> $BODY$ >> DECLARE >> my_id integer; >> my_time integer; >> my_sequential boolean; >> old_time integer; >> change_count integer; >> BEGIN >> change_count := 1; >> >> for my_id,my_time,my_sequential in >> select id,extract(day from time),sequential from tmp1 order by time >> LOOP >> if (old_time is not null) then >> if (old_time+1 = my_time) then >> my_sequential := true; --How do I make this work? >> change_count := change_count+1; >> end if; >> end if; >> old_time := my_time; >> END LOOP; >> return change_count; >> END; >> $BODY$ >> LANGUAGE 'plpgsql'; > > You could use a cursor and UPDATE ... WHERE CURRENT OF for that. > > Here are some lines of code: > > DECLARE > ... > mycur CURSOR FOR select id,extract(day from time),sequential from tmp1 > order by time FOR UPDATE; > row tmp1%ROWTYPE; > BEGIN > ... > OPEN mycur; > LOOP > FETCH mycur INTO row; > EXIT WHEN NOT FOUND; > ... > UPDATE tmp1 SET sequential = TRUE WHERE CURRENT OF mycur; > ... > END LOOP; > CLOSE mycur; > ... > END; > > Yours, > Laurenz Albe > > -- > > Thanks. Follow up question: It appears that I cannot modify data using the cursor when there is an 'order by' clause (I get: 'ERROR: cursor "mycur" is not a simply updatable scan of table "tmp1"'). Is there anyway to get around this, via an index or something? Thanks. -- View this message in context: http://www.nabble.com/Function%3A-Change-data-while-walking-through-records-tp23873836p23985514.html Sent from the PostgreSQL - general mailing list archive at Nabble.com.