|User & Date:||drh 2010-08-04 18:43:50|
- Change comment to:
Consider the following SQL:
CREATE TABLE t1(a INTEGER PRIMARY KEY,b,c); CREATE INDEX t1c ON t1(c); INSERT INTO t1 VALUES(1,2,3); CREATE TABLE t2(x, y); ANALYZE; UPDATE sqlite_stat1 SET stat='20000 100'; explain query plan SELECT * FROM t2, t1 WHERE t2.x=5 AND t1.a=t2.y;
The SELECT at the end should ideally do a full table scan over table T2 and then use the PRIMARY KEY to look up entries of T1, resulting in an O(NlogN) execution time. However, beginning with SQLite version 3.7.0 (and specifically check-in [defaf0d99a807] on 2010-04-15 01:04:54, the query planner does two full table scans on T1 first and then T2 as the inner loop, resulting an O(N**2) execution time.
- Change detected to "Application_Fault"
- Change foundin to "3.7.0"
- Change severity to "Important"
- Change status to "Open"
- Change subsystem to "Code_Generator"
- Change title to "Performance regression"
- Change type to "Code_Defect"