Sql Objects And PL/SQL

Editor's Notes

• #3: Did a bit of COBOL and IDMS-XThen Ingres for a couple of yearsFound Oracle around the Forms 3 Era, used a lot of Pro*CobolCame to Australia about 12 years ago from England.Forms work in Perth and Adelaide.Arrived in Sydney in ‘99Contracted with the Office of State Revenue, P&O Nedlloyd, MBF Health Insurance…Consultant with DWS for a year, contracted for a bit more.Been consulting with SMS for about 18 monthsCurrently at SiCorp on a very small data warehouse.
• #4: SQL is very good at set-based operations.When the task isn’t set based, it is worth considering a procedural approach.Processing of large data volumes are more likely to be amenable to set-based solutions. Because PL/SQL runs in the database, and as the same process performing SQL operations, there is a smaller detrimental impact from context switching than you may get from a client language containing SQL (e.g. Perl, Python )In Shared Server, session persistent data is stored in the UGA which is in the SGA, but data that is only required for the duration of the call may still be in the PGA.
• #5: By not requiring a translation between the SQL types and PL/SQL types, any impedance is reduced.
• #6: PL/SQL has its own particular datatypes that are geared towards PROCESSING of data, as opposed to STORAGE. Mostly they are beneficial in terms of performance and/or memory consumption.Unless you do heavy processing in PL/SQL, the benefits are likely to be smallPlus they cannot be used in embedded SQL statements, and have to be converted to a standard datatype.
• #7: You can have type declarations in PL/SQL which are local to the code block.If declared in a package specification, they can be used more widely, but (generally) only within PL/SQL.You can assign variables of the same type to each other. You can assign NULL to a record type.Comparison isn't as simple (and don't forget NULLs in record items when you do comparisons).Sometimes worth creating comparison functions.
• #8: An SQL Object type has attributes (a cross between variables and columns) and optionally member functions/proceduresThese can be used in SQL and PL/SQL
• #9: Objects come with a default constructor method, where you can create an INSTANCE of the object by stating the object name then, in brackets, values for every one of the attributes. You can create your own constructor functions too. They MUST have the same name as the type, but can differ in their parameters. They can have different numbers of parameters, parameter names, datatypes and default values.I like constructors that have parameters that default to null, so I don't have to supply them all
• #10: To implement the method, you need a BODY, in the same way as a package body. Unlike packages, a type body cannot contain functions not declared in the TYPE specification. [But you can declare private functions / procedures in the declaration part of the member functions.CREATE OR REPLACE TYPE body holt is CONSTRUCTOR FUNCTION holt ( i_record_type in varchar2 default null,i_record_column in varchar2 default null) return self as result isv_number number; procedure p_test is beginv_number := 1; end; beginself.record_type := i_record_type;self.record_column := i_record_column; RETURN; end holt;end;
• #11: You can use user-defined types as easily as the built-in types.
• #12: A collection type, in PL/SQL, can be of an ARRAY or TABLE. I find tables more useful as I don’t need to consider size when I declare them. A collection is of a single data type. However that datatype may be a simple SQL datatype [such as NUMBER or VARCHAR2(n)], a user-defined SQL datatype (such as a Spatial co-ordinate)) a PL/SQL type such as PLS_INTEGER or a user-defined PL/SQL record type. You can even define collections of collections, though that would get complicated.
• #13: So you can have a collection like CREATE TYPE tab_varchar_4000 IS TABLE OF VARCHAR2(4000);CREATE TYPE strings_coll IS OBJECT (coll tab_varchar_4000);
• #14: You are limited to a default constructor if you use the type directly.
• #15: There's no public synonym for it, so you need to use the SYS schema declaration
• #16: You need to initialize the collection.Then extend it to the size necessary.Load up the rows and return it.You can then just select from itselect * from table(gen_dates(sysdate, sysdate+10));
• #17: Pipelined functions don't need to assemble the whole dataset into memory before returning it.You MUST have the RETURN at the end, and it returns nothing.
• #18: You've probably seen code like this. Lots of debug statements to report on what the code is doing.Or rather what the code has done, as dbms_output doesn't get reported until the end.Can we use PIPE ROW to report on debugging during processing ?
• #19: You don't want to try wrapping it all up in autonomous transactions and ugly stuff like that
• #20: You've probably seen code like this. Lots of debug statements to report on what the code is doing.Or rather what the code has done, as dbms_output doesn't get reported until the end.Can we use PIPE ROW to report on debugging during processing ?The PIPE ROW has to be in the top level (ie can't be delegated to a called subroutine).create function process_records return tab_varchar_4000 pipelined isv_str varchar2(10); procedure put (p_text in varchar2) is begin pipe row (p_text); end;begin for i in 1 .. 10 loop select id into v_str from recs where id = i; put ('update '||i); end loop; return;end;/PLS-00629: PIPE statement cannot be used in non-pipelined functions
• #21: Warning - NO_DATA_FOUND is a PL/SQL exceptionIt is NOT an SQL error.When you select from a function returning a table, if the PL/SQL encounters an unhandled NO_DATA_FOUND exception, it acts like a RETURN. The function ends, and as far as the SQL is concerned, it completed successfully.create function gen_nums return tab_varchar_4000 pipelinedis cursor c_1 is select rownumrn from dual connect by level < 10;v_num number;begin for i in c_1 loop select 1 into v_num from dual where i.rn < 3; pipe row (i.rn); end loop; return;end;/select * from table(gen_nums);Only returns 2 rows.ALWAYS use a NO_DATA_FOUND exception handler and raise it as a different error.drop function gen_nums;create function gen_nums return tab_varchar_4000 pipelinedis cursor c_1 is select rownumrn from dual connect by level < 10;v_num number;begin for i in c_1 loop select 1 into v_num from dual where i.rn < 3; pipe row (i.rn); end loop; return;exception when no_data_found thenraise_application_error(-20001,'NDF',false);end;/select * from table(gen_nums);
• #22: drop function gen_char;create function gen_char return tab_varchar_4000 pipelinedis type tab_val is table of varchar2(10) index by pls_integer;t_valtab_val;begint_val(1) := 'test 1';t_val(10) := 'test 10'; pipe row (t_val(10)); pipe row (t_val(5)); pipe row (t_val(1)); return;--exception-- when no_data_found then-- raise_application_error(-20001,'NDF',false);end;/select * from table(gen_char);
• #23: create or replace package test_pipe_pkg is type rec_pipe is record (id number, val varchar2(20)); type tab_pipe is table of rec_pipe; function ret_pipe return tab_pipe pipelined;end;/create or replace package body test_pipe_pkg is function ret_pipe return tab_pipe pipelined isv_recrec_pipe; begin v_rec.id := 1; v_rec.val := 'test'; pipe row (v_rec); return; end;end;/select * from table(test_pipe_pkg.ret_pipe);select type_name from user_types where type_name like 'SYS_PLSQL%';alter type "GARY"."SYS_PLSQL_13672_9_1" add attribute gid number;
• #24: create table t_sub_postcodes (postcode varchar2(4), suburb_name varchar2(20));insert into t_sub_postcodes values ('2000','Sydney');insert into t_sub_postcodes values ('3000','Melbourne');create or replace procedure demo_cache is type tab_postcode is table of varchar2(4) index by varchar2(40);t_postcodetab_postcode; cursor c_1 is select decode(mod(rownum,2),1,'Sydney','Melbourne') sub from dual connect by level <10; function get_postcode (i_suburb in varchar2) return varchar2 isv_postcode varchar2(4); begin if t_postcode.exists (i_suburb) then return t_postcode(i_suburb); elsedbms_output.put_line('Fetching'); select postcode into v_postcode from t_sub_postcodes where suburb_name = i_suburb;t_postcode(i_suburb) := v_postcode; return t_postcode(i_suburb); end if; end;begin for c_rec in c_1 loopdbms_output.put_line(get_postcode(c_rec.sub)); end loop; end;/show errors procedure demo_cacheexec demo_cache;
• #25: create table test (val varchar2(10));insert into test select sysdate+level from dual connect by level < 10;select cast(collect(val) as tab_varchar_4000) a from test;
• #27: create type mset is table of varchar2(4000);/create type mset_mset is table of mset;/
• #28: create type mset is table of varchar2(4000);/create type mset_mset is table of mset;/
• #30: First one checks if an ELEMENT is in the setSecond/third ones check if a set is contained within a set
• #31: set serveroutput on declare x1 tab_varchar_4000; x2 tab_varchar_4000;begin select cast(collect(to_char(sysdate)) as tab_varchar_4000) into x1 from dual connect by level < 10; -- select cast(collect(to_char(sysdate)) as tab_varchar_4000) into x2 from dual connect by level < 5; -- if x1 = x2 thendbms_output.put_line('Match'); --elsif x1 < x2 then -- dbms_output.put_line('Less'); elsedbms_output.put_line('No Match'); end if;end;./
• #32: Also works with locally defined collectionsdeclare TYPE lset is table of varchar2(30); x1 lset := lset('a','b'); x2 lset := lset('b'); x3 lset;begin x3 := x1 multiset except x2;dbms_output.put_line(x3.count||'.'||x3(1));end;But not complex ones (SQL or PL/SQL)declare TYPE lrc is record (id number, val varchar2(10)); TYPE lset is table of lrc; x1 lset; x2 lset; x3 lset; function x (p_id in number, p_val in varchar2) return lrc isv_xlrc; begin v_x.id := p_id; v_x.val := p_val; return v_x; end;begin x1 := lset(x(1,'a'), x(2,'b')); x2 := lset(x(2,'b')); x3 := x1 multiset except x2;dbms_output.put_line(x3.count);end;
• #33: SQL*Plus is a poor tool for procedural logic (loops and conditional flows)Embedding calls to sqlplus in a shell script is not a good idea.You generally won’t have a persistent connection, so will be continually logging on and off the database. You can’t maintain transactionsThere is poor error/exception handling as you can really only return a generic error/warning to the shell script.Use Perl (which gets installed with the databases, except XE) or Python or….PL/SQL