Function-based indexes problem

Question

I'he entered the lines below:
ALTER SESSION
SET OPTIMIZER_MODE = FIRST_ROWS;
ALTER SESSION
SET QUERY_REWRITE_ENABLED = TRUE;
ALTER SESSION
SET QUERY_REWRITE_INTEGRITY = TRUSTED;
CREATE TABLE qwe(
qwe_id CONSTRAINT pk_qwe_id PRIMARY KEY,
data VARCHAR2(10)
CONSTRAINT nn_qwe_data NOT NULL
CREATE UNIQUE INDEX u_qwe_data_idx
ON qwe(UPPER(data)) COMPUTE STATISTICS;
ANALYZE TABLE qwe COMPUTE STATISTICS;
The reason is to obtain EXECUTION PLAN for
simple queries:
1) EXPLAIN PLAN FOR
SELECT UPPER(data) FROM qwe
WHERE UPPER(data) IS NOT NULL
ORDER BY UPPER(data);
That is OK - Oracle use u_qwe_data_idx INDEX
2) EXPLAIN PLAN FOR
SELECT * FROM qwe
WHERE UPPER(data) IS NOT NULL
ORDER BY UPPER(data)
Oops! TABLE ACCESS - FULL
Jrn-mhasd| lnfer }rn nazqmhr|?
u_qwe_data_idx INDEX is IGNORED
Does anybody know how to make ORACLE to use this INDEX and where is my MISTAKE?

Answer 1

how big is your table? if it's less than
DB_BLOCK_SIZE * DB_FILE MULTIBLOCK_READ_COUNT
( which should sum to 64k or thereabouts, the figure is machine/OS specific) then the whole table can be read in a single sweep whereas an indexed read will always require at least two sweeps.
This is compounded if most of your column data is populated: it is much more efficient to do a full table scan and then produce a result set in memeory by discarding the rows that don't meet your criteria.
In the old days the Rule Based Optimizer would have used your index and performed very badly. the whole point about statistics is they provide the Cost Based Optimizer with an accurate view of the distribution of data in your tables so that it can see whether it is more efficient to use any given index. And it your scenario the answer is no.
I hope that's clear.
HTH, APC