"Note: TKPROF cannot tell the TYPE of the bind variables simply by looking at the text of the SQL statement. It assumes that TYPE is CHARACTER; if this is not the case, you should put appropriate type conversions in the SQL statement."
Additional test results are described by Gerard:
"Michael A. Behr" wrote:
> > According to Mike the customer tried the operation with the ATTRIBUTES_KEY > > index removed and then it *really* ran slow. This also supports the > > contention that the index was being used despite what tkprof says. > > That was actually our own John Bedell, not a customer
Today, we ran this so as to eliminate (as best we could) the effects of caching by restarting Oracle between each test run. The following material shows the result of running tkprof over the trace files. Only the material relevant to the query we were in doubt about is included here. Note that in spite of the clear evidence to the contrary, the "explain plan" indicates a full table scan in all three cases, while only the case where the index was removed were a reasonable number of bffer gets and disk i/o perfromed.
// After we drop the constraint which created the primary index and bounced // the Oracle instance