Ticket Hash: | 9936b2fa443fec03ff25f9b822528c20a2200a49 | |||
Title: | Infinite loop due to the ORDER BY LIMIT optimization | |||
Status: | Fixed | Type: | Code_Defect | |
Severity: | Critical | Priority: | Immediate | |
Subsystem: | Unknown | Resolution: | Fixed | |
Last Modified: | 2018-09-10 14:49:04 | |||
Version Found In: | 3.24.0 | |||
User Comments: | ||||
drh added on 2018-09-08 03:22:02:
A confluence of multiple defects in the code generator associated with the ORDER BY LIMIT optimization causes the prepared statement to enter an infinite loop while running the final query in the following SQL: CREATE TABLE t1(aa VARCHAR PRIMARY KEY NOT NULL,bb,cc,x VARCHAR(400)); INSERT INTO t1(aa,bb,cc) VALUES('maroon','meal','lecture'); INSERT INTO t1(aa,bb,cc) VALUES('reality','meal','catsear'); CREATE TABLE t2(aa VARCHAR PRIMARY KEY, dd INT DEFAULT 1, ee, x VARCHAR(100)); INSERT INTO t2(aa,dd,ee) VALUES('maroon',0,'travel'),('reality',0,'hour'); CREATE INDEX t2x1 ON t2(dd,ee); ANALYZE; DROP TABLE IF EXISTS sqlite_stat4; DELETE FROM sqlite_stat1; INSERT INTO sqlite_stat1 VALUES ('t2','t2x1','3 3 3'), ('t2','sqlite_autoindex_t2_1','3 1'), ('t1','sqlite_autoindex_t1_1','2 1'); ANALYZE sqlite_master; PRAGMA vdbe_trace=on; SELECT * FROM t1 LEFT JOIN t2 ON t1.aa=t2.aa WHERE t1.bb='meal' ORDER BY t2.dd DESC LIMIT 1; The infinite loop is only entered if the command-line shell is compiled without -DSQLITE_DEBUG or without -DSQLITE_ENABLE_MODULE_COMMENTS. If both of those compile-time options are enabled, then another bug in the code generator causes the ORDER BY LIMIT optimization to be disabled, and so the infinite loop does not occur. This problem was originally reported on the SQLite users mailing list. It took considerable effort to reproduce the problem and then boil it down to the repro script shown above. The script above first began failing with check-in [02ba8a7ba7ba71cd7abd5d] which added the omit-left-join optimization. That optimization does not seem to be the root cause, but merely a changed that allowed it to be expressed with the current script. The ORDER BY LIMIT optimization that probably is the root cause was introduced in version 3.14 (2016-08-08) by check-in [559733b09e9630fac9d]. dan added on 2018-09-08 17:43:12: Different manifestation of one of the same problems causes SQLite to return incorrect data: CREATE TABLE t1(aa, bb); INSERT INTO t1 VALUES('maroon','meal'); CREATE TABLE t2(cc, dd, ee, x VARCHAR(100)); INSERT INTO t2(cc,dd,ee) VALUES('maroon',1,'one'); INSERT INTO t2(cc,dd,ee) VALUES('maroon',2,'two'); INSERT INTO t2(cc,dd,ee) VALUES('maroon',0,'zero'); CREATE INDEX t2ddee ON t2(dd,ee); CREATE INDEX t2cc ON t2(cc); ANALYZE; ---------- returns "maroon|0|zero" (correct) SELECT t2.cc, t2.dd, t2.ee FROM t1 CROSS JOIN t2 ON t1.aa=t2.cc ORDER BY t2.dd LIMIT 1; ---------- returns "maroon|1|one" (incorrect) SELECT t2.cc, t2.dd, t2.ee FROM t1 CROSS JOIN t2 ON t1.aa=t2.cc WHERE t1.bb='meal' ORDER BY t2.dd LIMIT 1; |