unexpected query success

SELECT COUNT (*)
  FROM rps2_workflow
 WHERE     workflow_added > TO_DATE ('01.09.2011', 'dd.mm.yyyy')
       AND workflow_finished < TO_DATE ('wtf', 'dd.mm.yyyy')
       AND workf开发者_运维知识库low_status IN (7, 12, 17)
       AND workflow_worker = 159

I expect this query to fail, because of invalid date, but it returns 0

The plan for this query shows that on 8th step the invalid clause is processed:

8 TABLE ACCESS BY INDEX ROWID TABLE RPS2.RPS2_WORKFLOW Object Instance: 1  Filter Predicates: ("WORKFLOW_STATUS"=7 OR "WORKFLOW_STATUS"=12 OR "WORKFLOW_STATUS"=17) AND SYS_EXTRACT_UTC("WORKFLOW_FINISHED")<SYS_EXTRACT_UTC(TO_DATE('wtf','dd.mm.yyyy'))  Cost: 11  Bytes: 33  Cardinality: 1  CPU Cost: 8 M  IO Cost: 10  Time: 1                     

If we comment out AND workflow_status IN (7, 12, 17) condition - then expectedly we get ORA-01858: a non-numeric character was found where a numeric was expected

If we comment out AND workflow_finished < TO_DATE ('wtf', 'dd.mm.yyyy') then we get amount of records that fit that conditions (> 0)

How is this possible?

UPD:

The hint /*+no_index(rps2_workflow) */ doesn't change anything (whereas in the plan we see that fullscan is performed)

SELECT STATEMENT  ALL_ROWSCost: 254  Bytes: 31  Cardinality: 1  CPU Cost: 34 M  IO Cost: 248  Time: 4       
2 SORT AGGREGATE  Bytes: 31  Cardinality: 1     
    1 TABLE ACCESS FULL TABLE RPS2.RPS2_WORKFLOW Object Instance: 1  Filter Predicates: "WORKFLOW_WORKER"=159 AND ("WORKFLOW_STATUS"=7 OR "WORKFLOW_STATUS"=12 OR "WORKFLOW_STATUS"=17) AND SYS_EXTRACT_UTC("WORKFLOW_ADDED")>SYS_EXTRACT_UTC(TIMESTAMP' 2011-09-01 00:00:00') AND SYS_EXTRACT_UTC("WORKFLOW_FINISHED")<SYS_EXTRACT_UTC(TO_DATE('wtf','dd.mm.yyyy'))  Cost: 254  Bytes: 31  Cardinality: 1  CPU Cost: 34 M  IO Cost: 248  Time: 4  

If the optimizer decides that it doesn't need to evaluate a function, it won't, so the function will never throw exceptions:

select 1 from dual where 1 = 1 OR to_date('asdasdasd','asdasdasdas') > sysdate ;

         1
----------
         1

The function raises an exception only if it actually gets evaluated:

SQL> select 1 from dual where 1 = 1 AND to_date('asdasd','asdas') > sysdate ;
select 1 from dual where 1 = 1 AND to_date('asdasd','asdas') > sysdate
                                                    *
ERROR at line 1:
ORA-01821: date format not recognized

However, if the parser can decide statically that the query is invalid - because the function has the wrong type of arguments or the query has invalid types, then the parser will raise an exception before the optimizer gets a swing at it:

SQL> select 1 from dual where 1 = 1 or to_date('asdasdasd',0) > sysdate ;
select 1 from dual where 1 = 1 or to_date('asdasdasd',0) > sysdate
                                                         *
ERROR at line 1:
ORA-00932: inconsistent datatypes: expected DATE got NUMBER


SQL> select 1 from dual where 1 = 1 or to_date('asdasdasd','asdasdasdas') > 42 ;
select 1 from dual where 1 = 1 or to_date('asdasdasd','asdasdasdas') > 42
                                                                        *
ERROR at line 1:
ORA-00932: inconsistent datatypes: expected DATE got NUMBER

It probably found that every record satisfying all the other conditions have a NULL workflow_finished field.

And anything compared to NULL is unknown so it doesn't need to evaluate the other operand.