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 workflow_status IN (7, 12, 17)
       AND workflow_worker = 159

由于日期无效，我预计此查询将失败，但它返回0

此查询的计划表明，在第8步 中 处理了无效子句：

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

如果我们注释掉AND workflow_status IN (7, 12, 17)条件-那么我们有望得到ORA-01858: a non- numeric character was found where a numeric was expected

如果我们注释掉，AND workflow_finished < TO_DATE ('wtf', 'dd.mm.yyyy')那么我们将获得满足该条件的记录数量（> 0）

这怎么可能？

UPD ：

提示/*+no_index(rps2_workflow) */没有任何改变（而在该计划中，我们看到执行了全扫描）

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