I have been doing some data transformation/processing in PL/SQL and I want to eliminate duplicate code from my package. This are the relevant code parts.
Input type for the table function:
type t_legs_cur is ref cursor return legs%rowtype;
Procedure that process one record:
procedure discontinuity_rule(p_leg in out nocopy legs%rowtype) as
begin
null; --business logic here
end discontinuity_rule;
Table function that iterates over a cursor, process each row in cursors and pipes the output (if any):
function apply_discontinuity_rule(p_cur t_legs_cur)
return t_legs pipelined
order p_cur by (/* some fields */)
parallel_enable (partition p_cur by range (/* some fields */))
as
v_leg legs%rowtype;
begin
loop
fetch p_cur into v_leg;
exit when p_cur%notfound;
discontinuity_rule(v_leg); --call back
if v_leg.id is not null then
pipe row (v_leg);
end if;
end loop;
end apply_discontinuity_rule;
There are several steps of transformation/processing, e.g. I would run the following select to do the some processing and apply some rules in a given order:
select * from table(trip_rules.generate_trips_from_legs(cursor(
select * from table(trip_rules.apply_5_legs_rule(cursor(
select * from table (trip_rules.apply_previous_city_rule(cursor(
select * from table (trip_rules.apply_backhaul_rule(cursor(
select * from table(trip_rules.apply_connection_time_rule(cursor(
select * from table(trip_rules.apply_discontinuity_rule(cursor(
select * from table(trip_rules.generate_legs_from_input(cursor(
select * from INPUT_DATA
)))
)))
)))
)))
)))
)))
)));
This is all fine and dandy, only problem is, my trip_rule package contains many apply_*_rule functions. They all similar to the example apply_discontinuity_rule. The only difference is the actual procedure (discontinuity_rule) they call back.
SO, my question would be, how can I avoid to copy the code of the apply_* functions. Is there a more elegant way to do this then use a big if: if p_rule_name == 'discontinuity_rule' then
function apply_rule(p_cur t_legs_cur, p_rule_name in varchar2)
return t_legs pipelined
order p_cur by (/* some fields */)
parallel_enable (partition p_cur by range (/* some fields */))
as
v_leg legs%rowtype;
begin
loop
fetch p_cur into v_leg;
exit when p_cur%notfound;
if p_rule_name == 'discontinuity_rule' then
discontinuity_rule(v_leg);
elsif p_rule_name == 'other_rule' then
other_rule(v_leg);
elsif p_rule_name == 'totally_other_rule' then
totally_other_rule(v_leg);
-- and so on...
end if;
if v_leg.id is not null then
pipe row (v_leg);
end if;
end loop;
end apply_rule;
I also understand that it would be possible to create an anonymous PL/SQL block on the fly using the procedure name and execute it as dynamic SQL. I wonder if it can be done properly, without killing my performance. Any idea is appreciated.
Your giant IF statement is not duplicate code.
True, it has parts which resemble each other but this ...
elsif p_rule_name == 'other_rule' then
other_rule(v_leg);
... is most definitely not the same as this ...
elsif p_rule_name == 'totally_other_rule' then
totally_other_rule(v_leg);
Dynamic PL/SQL is something we should avouid unless there really is no alternative. There is no need for it here.
Related
I have a pipeline function. If I catch an exception, I want to return no lines. Even if the exception occurs after the first utilisation of pipe row.
CREATE TYPE a_r IS OBJECT (a1 VARCHAR (1), a2 VARCHAR (1));
CREATE TYPE a_t IS TABLE OF a_r;
CREATE or replace FUNCTION get_a
RETURN a_t
PIPELINED
IS
BEGIN
FOR c IN (SELECT '1' a , '2' b FROM DUAL)
LOOP
PIPE ROW (a_r(c.a,c.b));
END LOOP;
FOR a
IN (SELECT 'a2' a,
'b' b
FROM DUAL)
LOOP
PIPE ROW (a_r(a.a,a.b));
END LOOP;
exception
WHEN VALUE_ERROR
THEN
DBMS_OUTPUT.put_line ('VALUE_ERROR');
WHEN OTHERS
THEN
DBMS_OUTPUT.put_line ('other');
END;
select * from table(get_a())
In this example, Before the error ('ab' is not varchar2(1)) occurs, the first line is already piped. But I want nothing to be returned. Is there a way to write cancel what was piped in the exception block?
code
No because the client may already have consumed the row.
You could code your function so that it fills a local collection with data and only at the end decides whether to iterate through the collection to actually return the rows. That will require more memory since you're materializing the entire result set before returning it and defeats some of the performance benefits of using pipelined table functions.
Something like this would appear to do what you want
CREATE or replace FUNCTION get_a
RETURN a_t
PIPELINED
IS
l_a_t a_t := new a_t();
BEGIN
FOR c IN (SELECT '1' a , '2' b FROM DUAL)
LOOP
l_a_t.extend;
l_a_t( l_a_t.count ) := a_r(c.a,c.b);
END LOOP;
FOR a
IN (SELECT 'a2' a,
'b' b
FROM DUAL)
LOOP
l_a_t.extend;
l_a_t( l_a_t.count ) := a_r(a.a,a.b);
END LOOP;
for i in 1 .. l_a_t.count
loop
pipe row( l_a_t(i) );
end loop;
exception
WHEN VALUE_ERROR
THEN
DBMS_OUTPUT.put_line ('VALUE_ERROR');
WHEN OTHERS
THEN
DBMS_OUTPUT.put_line ('other');
END;
/
See this dbfiddle
Im trying to select some rows from a Table in ORACLE and at the same time update the selected rows state. I found a way to do so with a stored function and Cursors but I cant manage to return the rows after using the cursor to update. This is my code:
CREATE OR REPLACE FUNCTION FUNCTION_NAME
RETURN SYS_REFCURSOR
IS
l_return SYS_REFCURSOR;
CURSOR c_operations IS
SELECT * FROM TABLE1
WHERE STATUS != 'OK'
FOR UPDATE OF TABLE1.STATUS;
BEGIN
FOR r_operation IN c_operations
LOOP
UPDATE
TABLE1
SET
TABLE1.STATUS = 'OK'
WHERE
TABLE1.ID_TABLE1 = r_operation.ID_TABLE1;
END LOOP;
COMMIT;
-- Missing conversion from cursor to sys_refcursor
RETURN l_return;
END;
The update is working but Im still missing how to return the updated rows that are in the cursor (c_operations ).
Thank you.
I'm going to make some assumptions:
id_table1 is the primary key of the table, so your RBAR (*) update affects only one row
id_table1 is numeric
If these assumptions are wrong you will need to tweak the following code.
CREATE OR REPLACE FUNCTION FUNCTION_NAME
RETURN SYS_REFCURSOR
IS
l_return SYS_REFCURSOR;
l_id table1.id_table1%type;
l_upd_ids sys.odcinumberlist := new sys.odcinumberlist();
CURSOR c_operations IS
SELECT * FROM TABLE1
WHERE STATUS != 'OK'
FOR UPDATE OF TABLE1.STATUS;
BEGIN
FOR r_operation IN c_operations LOOP
UPDATE TABLE1
SET TABLE1.STATUS = 'OK'
WHERE TABLE1.ID_TABLE1 = r_operation.ID_TABLE1
returning TABLE1.ID_TABLE1 into l_id;
l_upd_ids.extend();
l_upd_ids(l_upd_ids.count()) := l_id;
END LOOP;
COMMIT;
open l_return for
select * from table(l_upd_ids);
RETURN l_return;
END;
The key points of the solution.
uses Oracle maintained collection (of number) sys.odcinumberlist to store the updated IDs;
uses RETURNING clause to capture the id_table1 value for the updated row;
stores the returned key in the collection;
uses a table() function to casrt the collection into a table which can be queried in the ref cursor.
This last point is why I chose to use sys.odcinumberlist rather than defining a collection in the procedure. It's a SQL type, so we can use it in SELECT statements.
(*) Row-by-agonizing-row. Updating single records in a PL/SQL loop is the slowest way of executing bulk updates, and normally constitutes an anti-pattern. A straightforward set-based UPDATE should suffice. However, you know your own situation so I'm going to leave that as it is.
It looks to me like you don't need the initial cursor, since you're changing the STATUS of every row which is not 'OK' to 'OK', so you can do this is a simple UPDATE statement. Then use an OPEN...FOR statement to return a cursor of all rows where STATUS is not 'OK', which shouldn't return anything because you've already changed all the status values to 'OK'. I suggest that you rewrite your procedure as:
CREATE OR REPLACE FUNCTION FUNCTION_NAME
RETURN SYS_REFCURSOR
IS
l_return SYS_REFCURSOR;
BEGIN
UPDATE TABLE1
SET STATUS = 'OK'
WHERE STATUS != 'OK';
COMMIT;
OPEN l_return FOR SELECT *
FROM TABLE1
WHERE STATUS != 'OK'
FOR UPDATE OF TABLE1.STATUS;
RETURN l_return;
END;
Instead of a loop to update how about a bulk update collecting the updated ids. Then a table function from those returned ids.
create type t_table1_id is
table of integer;
create or replace function set_table1_status_ok
return sys_refcursor
is
l_results_cursor sys_refcursor;
l_updated_ids t_table1_id;
begin
update table1
set status = 'Ok'
where status != 'Ok'
returning table1.id
bulk collect
into l_updated_ids;
open l_results_cursor for
select *
from table1
where id in (select * from table(l_updated_ids));
return l_results_cursor;
end set_table1_status_ok;
-- test
declare
updated_ids sys_refcursor;
l_this_rec table1%rowtype;
begin
updated_ids := set_table1_status_ok();
loop
fetch updated_ids into l_this_rec;
exit when updated_ids%notfound;
dbms_output.put_line ( l_this_rec.id || ' updated.');
end loop;
close updated_ids;
end ;
I created a function that takes a movie id as input and returns stock information based from the ID. The function mostly works but if I want to retrieve information from a movie that is not in the database(returns no rows) nothing returns. Can't figure out why?
doesn't give me an error when i call an ID that returns no rows so exception handling wouldn't work.
create or replace function stock_info
(p_id IN NUMBER
)
return VARCHAR2
IS
cursor c1 is
select movie_id, movie_title, movie_qty
from mm_movie
where p_id = movie_id;
lv_movie_info VARCHAR2(100);
BEGIN
for i in c1 loop
if p_id = i.movie_id then
lv_movie_info := i.movie_title || ' is available: ' || i.movie_qty || ' on the shelf';
else
lv_movie_info := 'no data found';
end if;
end loop;
return lv_movie_info;
END STOCK_INFO;
/
The reason you don't get anything when there is no data is that the loop doesn't execute. Logically the For expression says "execute the following loop for every row returned in the cursor" but there are no rows in the cursor so it never executes the loop. Further the structure actually indicates you are expecting multiple for a given p_id. If that's not the case you can eliminate the cursor all together. Assuming p_id is the primary key you have either 0 or 1 row so:
create or replace function stock_info (p_id in number)
return text
is
lv_movie_info varchar2(100);
begin
select i.movie_title || ' is available: ' || i.movie_qty || ' on the shelf'
into lv_movie_info
from mm_movie i
where p_id = movie_id;
return lv_movie_info;
exceptions
when no_data_found
then return 'no data found';
end stock_info;
Of course if do expect more that 1 row the cursor is needed, but the IF is not as the were clause guarantees it's true. Still with 0 rows the loop will not be executed so the 'no data found' message needs to go after "End Loop".
Belayer
the cursor statement you used fetches data from the in parameter. i.e., in the cursor select you limiting based on the movie id passed.
on passing a movie id which is not in the data base, the cursor select statement would not fetch any records, and so the flow won't even go inside the for loop.
if you wanted to return no data found - on passing a movie id which is not in the database, two ways to resolve
1. before the loop, have select statement to set a flag to Y or N if exists according and to have your requirement.
2. in if not using for cursor, there is an option to check not found...
sample:
declare
cursor c1 is select * from table_sample; -- empty table
c_rec c1%rowtype;
begin
open c1;
fetch c1 into c_rec;
if c1%notfound then
dbms_output.put_line('not found');
end if;
close c1;
end;
-- output
not found
Really need help regarding Ref Cursor. I have a Stored Procedure GET_PERSONROLES that have parameter type ref cursor. I just wanted to pupulate this ref cursor manually like inserting a row to the refcursor.
Can I insert a row into a refcursor though a loop?
Thank you in advance.
The procedure depends on this publicly declared type:
create or replace package types
as
type cursorTypePersonRole is ref cursor;
end;
Here is my pseudo-codeL
create or replace PROCEDURE GET_PERSONROLES
(
P_CURSOR IN OUT types.cursorTypePersonRole
) AS
REFCUR SYS_REFCURSOR;
TYPE REFTABLETYPE IS RECORD (
IS_MANAGER_LEVEL1 VARCHAR2(1),
IS_MANAGER_LEVEL2 VARCHAR2(1)
);
TYPE TABLETYPE IS TABLE OF REFTABLETYPE;
PERSONROLES_TABLETYPE TABLETYPE;
BEGIN
--calls another stored proc to populate REFCUR with data without problem
MY_STOREDPROC('12345', REFCUR);
LOOP
FETCH REFCUR BULK COLLECT INTO PERSONROLES_TABLETYPE;
EXIT WHEN PERSONROLES_TABLETYPE.COUNT = 0;
FOR indx IN 1 .. PERSONROLES_TABLETYPE.COUNT
LOOP
-- I'm able to query perfectly the values of IS_MANAGER_LEVEL1 and IS_MANAGER_LEVEL 2
-- I'm aware that the below codes are wrong
-- However this means I wanted to insert these values to a row of the cursor if possible
-- Do some logic to know what data will be assigned in the row.
if PERSONROLES_TABLETYPE(indx).IS_MANAGER_LEVEL1 = 'Y' then
P_CURSOR := <DO SOME LOGIC AND ASSIGN THE VALUE TO THE ROW>
end if;
if PERSONROLES_TABLETYPE(indx).IS_MANAGER_LEVEL2 = 'Y' then
P_CURSOR := <DO SOME LOGIC AND ASSIGN THE VALUE TO THE ROW>
end if;
END LOOP;
END LOOP;
CLOSE REFCUR;
END GET_PERSONROLES;
A ref cursor is not a variable: it is a pointer to a result set which is consumed by the act of reading it. The result set itself is immutable.
Immutability makes sense, because it reflects Oracle's emphasis on read consistency.
The simplest way to produce the output you appear to want is to create a SQL Type
open P_CURSOR for
select IS_MANAGER_LEVEL1,
IS_MANAGER_LEVEL2
from table ( PERSONROLES_TABLETYPE );
This will work in 12c; in earlier versions to use the table() call like this you may need to declare REFTABLETYPE and TABLETYPE as SQL types( rather than in PL/SQL).
"Ok edited it now"
Alas your requirements are still not clear. You haven't given us the structure of the output ref cursor or shown what other processing you want to undertake.
However, given the title of your question, let's have a guess. So:
create or replace PROCEDURE GET_PERSONROLES ( P_CURSOR IN OUT types.cursorTypePersonRole) AS
REFCUR SYS_REFCURSOR;
TYPE REFTABLETYPE IS RECORD (IS_MANAGER_LEVEL1 VARCHAR2(1),
IS_MANAGER_LEVEL2 VARCHAR2(1));
TYPE TABLETYPE IS TABLE OF REFTABLETYPE;
PERSONROLES_TABLETYPE TABLETYPE;
personrole_rec PersonRole%rowtype;
type personrole_nt is table of PersonRole%rowtype;
personroles_recs personrole_nt := new personrole_nt() ;
BEGIN
MY_STOREDPROC('12345', REFCUR);
FETCH REFCUR BULK COLLECT INTO PERSONROLES_TABLETYPE;
FOR indx IN 1 .. PERSONROLES_TABLETYPE.COUNT LOOP
/* in the absence of requirements I'm just making some stuff up */
if PERSONROLES_TABLETYPE(indx).IS_MANAGER_LEVEL1 = 'Y' then
personrole_rec.whatever1 := 'something';
else
personrole_recc.whatever1 := null;
end if;
if PERSONROLES_TABLETYPE(indx).IS_MANAGER_LEVEL2 = 'Y' then
personrole_rec.whatever2 := 'something else';
else
personrole_recc.whatever2 := null;
end if;
if personrole_rec.whatever1 is not null
or personrole_rec.whatever2 is mot null then
personroles_recs.exend();
personroles_recs(personroles_recs.count()) := personroles_rec;
end if;
END LOOP;
CLOSE REFCUR;
open p_cursor for
select * from table ( personroles_recs );
END GET_PERSONROLES;
This code uses a second collection to store the desired output. Like your code it reads the populated collection and evaluates the attributes of each row. If a value which means the criteria it sets an attribute in a rowtype variable. If one or both attributes are set it populates a new row in a second collection. At the end of the procedure it opens the ref cursor using a table() function call on the second collection.
Note that you do not need the nested loop: you're not using the LIMIT clause so your coder reads the entire cursor into the collection in one swoop.
The implemented rules may not be exactly what you want (because you haven't explained exactly what you want) but this should give you the general idea.
Note that, depending on exactly what processing is masked by <DO SOME LOGIC AND ASSIGN THE VALUE TO THE ROW>, the simpler approach could still be feasible:
open P_CURSOR for
select case when IS_MANAGER_LEVEL1 = 'Y' then 'YES' else 'NO' end,
case when IS_MANAGER_LEVEL2 = 'Y' then 'YES' else 'NO' end
from table ( PERSONROLES_TABLETYPE );
Thought I had followed creation pattern, but the body will not compile. What I am trying to accomplish is to develop a package to run a procedrure periodically to determine at what time and date more than 15 are in use.. Oracle 11g.
The only other data that needs to go into the table beingg inserted into the sysdate.
CREATE OR REPLACE
PACKAGE TAPES_USED AS
function TAPESCOUNT(count number) return number;
procedure INSERT_TAPES_COUNT(sysdate date, count NUMBER);
END TAPES_USED;
/
-----------------------------------------
CREATE OR REPLACE
PACKAGE body TAPES_USED AS
function TAPESCOUNT(count number) return number as count number;
begin
select count(*)
into
count
from DEV.TAPES_IN USE where count(*) > 15;
procedure INSERT_TAPES_COUNT(sysdate date, count NUMBER)as
begin
INSERT INTO DEV.TAPES_USED VALUES
(sysdate, count);
end INSERT_TAPES_COUNT;
END TAPES_USED;
/
Any help or suggestion anyone can offer will be appreciated.
CREATE OR REPLACE
PACKAGE BODY tapes_used AS
FUNCTION tapescount(in_ct NUMBER) RETURN NUMBER IS
ct NUMBER;
BEGIN
SELECT COUNT(*)
INTO ct
FROM dev.tapes_in_use;
IF ct > in_ct THEN
RETURN ct;
ELSE
RETURN NULL;
END IF;
END tapescount;
PROCEDURE insert_tapes_count(sysdt date, ct NUMBER) IS
BEGIN
INSERT INTO dev.tapes_used VALUES (sysdt, ct);
END insert_tapes_count;
END tapes_used;
/
You should refrain from using reserved words such as COUNT and SYSDATE for variable names (I don't know but that could be some of your compilation issues), so I've renamed them. Also, you forgot to END your function. I think you were missing an underscore in your table name in the FROM clause of the SELECT in your function, and you didn't have a RETURN statement in your function, which you must have.
Generally speaking, a function should accept one or more input parameters and return a single value. You're not making use of the input parameter in your function. I've implemented a suggested parameter.
As Egor notes, this isn't a realistic function, and I'm not certain about your intent here. What is the function supposed to do?
Maybe you want your function to return the Date/Time your count was exceeded? You could also combine everything into a single procedure:
PROCEDURE ck_tape_ct(min_tape_ct NUMBER) IS
ct NUMBER;
BEGIN
SELECT COUNT(*)
INTO ct
FROM dev.tapes_in_use;
IF ct > min_tape_ct THEN
INSERT INTO dev.tapes_used VALUES(SYSDATE, ct);
END IF;
END;