PL/SQL - Optional conditions in where-clause - without dynamic sql?

Question

I have a query where not all conditions are necessary. Here\'s an example of what it looks like when all conditions are used:

select num
from (select distinct q.         


        

Answer 1

I would just do this

select num
from (select distinct q.num
       from cqqv q
       where q.bcode = '1234567' --this is variable
             and q.lb = 'AXCT' --this is variable
             and q.type = nvl(<variable-type>, q.type)  --this condition ignored because of "type=*" in input
             and q.edate > sysdate - 30 --this is variable
       order by dbms_random.value()) subq
where rownum <= 10; --this is variable

One only has to guarantee that the variable-type is null when the q.TYPE filtering is to be ignored.

Answer 2

where ( columnA = passedValue or passedValue = -1 )

when passed value into sql is -1, columnA can be anything..

Answer 3

The solution I've settled on is one that generates an dynamic SQL query that may look like this:

select num
from (select distinct q.NUM
       from cqqv q 
       where  (q.bcode = :bcode) 
                  and  (1=1 or :lb is null) 
                  and  (1=1 or :type is null) 
                  and  (q.edate> :edate) 
                order by dbms_random.value()) subq 
where rownum <= :numrows

(in this example, the bcode and edate conditions were NOT optional, but the lb and type were)

I think this is (or is very similar to) what Michal Pravda was suggesting, and our DBA here prefers this solution over the context variable solution. Thanks for all that helped and offered advice!

A link our DBA found which details this solution is here:

Ask Tom: On Popularity and Natural Selection

Answer 4

While you could do this...

select num
from (select distinct q.num
       from cqqv q
       where 1=1
             and (:bcode is null or q.bcode = :bcode)
             and (:lb is null or q.lb = :lb)
             and (:type is null or q.type = :type)
             and (:edate is null or q.edate > :edate - 30)
       order by dbms_random.value()) subq
where rownum <= :numrows

... the performance using dynamic SQL will usually be better, as it will generate a more targeted query plan. In the above query, Oracle cannot tell whether to use an index on bcode or lb or type or edate, and will probably perform a full table scan every time.

Of course, you must use bind variables in your dynamic query, not concatenate the literal values into the string, otherwise performance (and scalability, and security) will be very bad.

To be clear, the dynamic version I have in mind would work like this:

declare
    rc sys_refcursor;
    q long;
begin
    q := 'select num
    from (select distinct q.num
           from cqqv q
           where 1=1';

    if p_bcode is not null then
        q := q || 'and q.bcode = :bcode';
    else
        q := q || 'and (1=1 or :bcode is null)';
    end if;

    if p_lb is not null then
        q := q || 'and q.lb = :lb';
    else
        q := q || 'and (1=1 or :lb is null)';
    end if;

    if p_type is not null then
        q := q || 'and q.type = :type';
    else
        q := q || 'and (1=1 or :type is null)';
    end if;

    if p_edate is not null then
        q := q || 'and q.edate = :edate';
    else
        q := q || 'and (1=1 or :edate is null)';
    end if;

    q := q || ' order by dbms_random.value()) subq
    where rownum <= :numrows';

    open rc for q using p_bcode, p_lb, p_type, p_edate, p_numrows;
    return rc;
end;

This means that the result query will be "sargable" (a new word to me I must admit!) since the resulting query run will be (for example):

select num
from (select distinct q.num
       from cqqv q
       where 1=1
             and q.bcode = :bcode
             and q.lb = :lb
             and (1=1 or :type is null)
             and (1=1 or :edate is null)
       order by dbms_random.value()) subq
where rownum <= :numrows

However, I accept that this could require up to 16 hard parses in this example. The "and :bv is null" clauses are required when using native dynamic SQL, but could be avoided by using DBMS_SQL.

Note: the use of (1=1 or :bindvar is null) when the bind variable is null was suggested in a comment by Michal Pravda, as it allows the optimizer to eliminate the clause.

Answer 5

While I agree with Tony that performance of using dynamic SQL is better, context variables is a better approach than using bind variables.

Using IN_VARIABLE IS NULL OR table.fieldx = IN_VARIABLE is not ideal for handling optional values. Each time a query is submitted, Oracle first checks in its shared pool to see if the statement has been submitted before. If it has, the execution plan for the query is retrieved and the SQL is executed. If the statement can not be found in the shared pool, Oracle has to go through the process of parsing the statement, working out various execution paths and coming up with the optimal access plan (AKA “best path”) before it can be executed. This process is known as a “hard parse”, and can take longer than the query itself. Read more about the hard/soft parse in Oracle here, and AskTom here.

In short - this:

and (:bcode is null or q.bcode = :bcode)

...will execute the same, dynamic or otherwise. There's no benefit to using bind variables in dynamic SQL for optional parameters. The setup still destroys SARGability...

Context parameters are a feature that was introduced in Oracle 9i. They are tied to a package, and can be used to set attribute values (only for users with EXECUTE permission on the package, and you'll have to grant CREATE CONTEXT to the schema). Context variables can be used to tailor dynamic SQL so it includes only what is necessary for the query based on the filter/search criteria. In comparison, Bind variables (also supported in dynamic SQL) require that a value is specified which can result in IN_VARIABLE IS NULL OR table.fieldx = IN_VARIABLE tests in the search query. In practice, a separate context variable should be used for each procedure or function to eliminate the risk of value contamination.

Here's your query using context variables:

L_CURSOR SYS_REFCURSOR;
L_QUERY  VARCHAR2(5000) DEFAULT 'SELECT num
                                   FROM (SELECT DISTINCT q.num
                                           FROM CQQV q
                                          WHERE 1 = 1 ';
BEGIN

    IF IN_BCODE IS NOT NULL THEN
      DBMS_SESSION.SET_CONTEXT('THE_CTX',
                               'BCODE',
                               IN_BCODE);
      L_QUERY := L_QUERY || ' AND q.bcode = SYS_CONTEXT(''THE_CTX'', ''BCODE'') ';
    END IF;

    IF IN_LB IS NOT NULL THEN
      DBMS_SESSION.SET_CONTEXT('THE_CTX',
                               'LB',
                               IN_LB);
      L_QUERY := L_QUERY || ' AND q.lb = SYS_CONTEXT(''THE_CTX'', ''LB'') ';
    END IF;

    IF IN_TYPE IS NOT NULL THEN
      DBMS_SESSION.SET_CONTEXT('THE_CTX',
                               'TYPE',
                               IN_TYPE);
      L_QUERY := L_QUERY || ' AND q.type = SYS_CONTEXT(''THE_CTX'', ''TYPE'') ';
    END IF;

    IF IN_EDATE IS NOT NULL THEN
      DBMS_SESSION.SET_CONTEXT('THE_CTX',
                               'EDATE',
                               IN_EDATE);
      L_QUERY := L_QUERY || ' AND q.edate = SYS_CONTEXT(''THE_CTX'', ''EDATE'') - 30 ';
    END IF;

    L_QUERY := L_QUERY || ' ORDER BY dbms_random.value()) subq
           WHERE rownum <= :numrows ';

    FOR I IN 0 .. (TRUNC(LENGTH(L_QUERY) / 255)) LOOP
      DBMS_OUTPUT.PUT_LINE(SUBSTR(L_QUERY, I * 255 + 1, 255));
    END LOOP;

    OPEN L_CURSOR FOR L_QUERY USING IN_ROWNUM;
    RETURN L_CURSOR;

END;

The example still uses a bind variable for the rownum, because the value is not optional.

DBMS_SESSION.SET_CONTEXT('THE_CTX', 'LB', IN_LB);

The SET_CONTEXT parameters are as follows:

1. The context variable name. There's no instance creation involved
2. The variable within the context variable. A context variable is like a session variable, assuming familiarity with web applications & session objects.
3. The value for the variable defined in parameter #2.

Bind vs Context

Bind variables means Oracle expects a variable reference to populate - it's an ORA error otherwise. For example:

... L_QUERY USING IN_EXAMPLE_VALUE

...expects that there is a single bind variable reference to be populated. If IN_EXAMPLE_VALUE is null, there has to be :variable in the query. IE: AND :variable IS NULL

Using a context variable means not having to include the extraneous/redundant logic, checking if a value is null.

IMPORTANT: Bind variables are processed in order of occurrence (known as ordinal), NOT by name. You'll notice there's no datatype declaration in the USING clause. Ordinals aren't ideal - if you change them in the query without updating the USING clause, it will break the query until it's fixed.