SQLite

View Ticket
Login
Ticket Hash: fccbde530a6583bf2748400919f1603d5425995c
Title: DISTINCT thinks a zeroblob() and blob of all zeros are different
Status: Fixed Type: Code_Defect
Severity: Important Priority: Immediate
Subsystem: Unknown Resolution: Fixed
Last Modified: 2014-02-08 19:12:24
Version Found In: 3.8.3
User Comments:
drh added on 2014-02-08 18:47:06:

The logic that computes DISTINCT sometimes thinks that a zeroblob() and a blob of all zeros are different when they should be the same. The following query illustrates the problem:

CREATE TABLE t1(a INTEGER);
INSERT INTO t1 VALUES(3);
INSERT INTO t1 VALUES(2);
INSERT INTO t1 VALUES(1);
INSERT INTO t1 VALUES(2);
INSERT INTO t1 VALUES(3);
CREATE TABLE t2(x);
INSERT INTO t2
  SELECT DISTINCT
    CASE a WHEN 1 THEN x'0000000000'
           WHEN 2 THEN zeroblob(5)
           ELSE 'xyzzy' END
    FROM t1;
SELECT quote(x) FROM t2;

The final output above should only generate two rows: x'0000000000' and 'xyzzy' since x'0000000000' and zeroblob(5) are logically equivalent. But it ends up generating three rows. There are two x'0000000000' values in the output.

Bisecting shows that this problem was introduced by check-in [45e581bff7a75db] on 2011-07-02 for SQLite version 3.7.8.

This problem was discovered by the developers during code analysis and has never been reported in the wild.