It’s a common misconception; but from a simple philosophical point of view the optimizer can’t “choose” to do anything. If you want to suggest that the optimizer can “choose” to ignore a hint that means that given the same query, data, stats, and hints, it would use one execution path one day and a different execution path another day. It doesn’t.

The optimizer may appear to ignore a hint, but the reason for that will be one of
(a) you don’t know the rules that make the hint relevant, or
(b) you’ve found a bug in the code, or
(c) the access path analysis code took a path through its decision tree that didn’t pass through the code where your hint was relevant (which means you didn’t hint the query properly, of course).

Thanks for another good observation. However, when it comes to “hints”. my understanding has always been that it just a “suggestion” to the optimizer and, of course, it can an choose to ignore it. This is weather you know how to use it or not. There are a lot factors that are involved in the optimizer choosing a path.

The answer to your question is: I don’t know why there are two calculatons for the index fast full scan, but I think (or, rather, I am prepared to guess) that one of them is dumped by the code path that does the original serial cost calculation, and the other is dumped by the code path that does the parallel costing.

In this case, the parallel costing calculation seems to be redundant because the degree stored in the data dictionary for the index is one, and there is no hint to tell Oracle to use parallelism on the index.

There is an anomaly in using parallel_index hints – or declaring indexes parallel, by the way: unless the table has also been declared parallel (default, or a value other than 1) then the cost of the parallel index_ffs is not adjusted by the degree of parallelism specified (even in 11g)

From your 10053 trace above, there are 2 index FFS access paths reported:

 Access Path: index (index (FFS))
   Index: T1_PK
   resc_io: 7.00  resc_cpu: 1342429
   ix_sel: 0.0000e+000  ix_sel_with_filters: 1
 Access Path: index (FFS)
   Cost:  7.23  Resp: 7.23  Degree: 1
     Cost_io: 7.00  Cost_cpu: 1342429
     Resp_io: 7.00  Resp_cpu: 1342429


The one reported as index (index (FFS)) has the index and selectivities
indicated. The one reported as index (FFS) doesn’t. What is the difference
between these 2 access paths? I am also seeing similar index FFS access
paths reported in my 10053 traces.

Many thanks,
Darian