|18:11||• Fixed ticket [f97c4637]: Incorrect ordering with ORDER BY and LIMIT plus 4 other changes (artifact: ac59934a user: drh)|
|03:04||Fix another instance of an incorrect value for KeyInfo.nXField on a sorting index. Ticket [f97c4637102a3ae72b79]. (check-in: 0077f645 user: drh tags: trunk)|
|02:50||Fix another instance of an incorrect value for KeyInfo.nXField on a sorting index. More changes for ticket [f97c4637102a3ae72b79]. (check-in: dc711db4 user: drh tags: branch-3.8.8)|
|21:38||Ensure that the KeyInfo.nXField value for ephemeral tables used to implement ORDER BY or GROUP BY clauses is set correctly, so that the sqlite3VdbeFindCompare() routine can choose the correct comparison function. Add assert() statements to the high-speed comparison functions to detect cases where they are inappropriately chosen. Fix for ticket [f97c4637102a3ae72b7911]. (check-in: 59e592f6 user: drh tags: branch-3.8.8)|
|21:36||Ensure that the KeyInfo.nXField value for ephemeral tables used to implement ORDER BY or GROUP BY clauses is set correctly, so that the sqlite3VdbeFindCompare() routine can choose the correct comparison function. Add assert() statements to the high-speed comparison functions to detect cases where they are inappropriately chosen. Fix for ticket [f97c4637102a3ae72b7911]. (check-in: f7201bb0 user: drh tags: trunk)|
|21:10||There are asserts in place now that will prevent a recurrence of ticket [f97c4637102a3ae7]. Nevertheless, it is good to add some test cases as well. (Closed-Leaf check-in: e02959b9 user: drh tags: tkt-f97c4637)|
|19:21||An alternative way of fixing the key comparison bug of ticket [f97c4637102a3ae72b]. (Closed-Leaf check-in: e41376cf user: drh tags: alt1-tkt-f97c4637)|
|17:28||Add an assert() to verify that the nField+nXField values of a KeyInfo object are never less then the number of columns in a row for a non-corrupt database. This assert() currently fails, which is the root of the problem with ticket [f97c4637102a3ae72b]. (check-in: 083f523d user: drh tags: tkt-f97c4637)|
|15:40||• Ticket [f97c4637] Incorrect ordering with ORDER BY and LIMIT status still Open with 6 other changes (artifact: 0c9d51df user: drh)|
|15:31||• New ticket [f97c4637]. (artifact: fc35a8b2 user: drh)|
|Title:||Incorrect ordering with ORDER BY and LIMIT|
|Last Modified:||2015-01-20 18:11:18|
|Version Found In:||3.8.8|
drh added on 2015-01-19 15:31:16:
The following SQL outputs rows in the incorrect order for SQLite versions 3.8.7 and 3.8.8:
CREATE TABLE t1(x); INSERT INTO t1(x) VALUES(1),(5),(3),(4),(2); SELECT x, 01, 02, 03, 04, 05, 06, 07, 08, 09, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60 FROM t1 ORDER BY x LIMIT -1;
In order to hit this bug, there must be an ORDER BY clause and a LIMIT clause and the number of columns in the query plus the number of sort keys needs to sum to exactly 62. Adding or removing a single column from the result set yields the correct answer.
This bug first appeared in check-in [b1c0f0bc1bd8a] which is when the multi-threaded sorter was introduced.
drh added on 2015-01-19 15:40:42:
Correction: the total number of columns plus the number of sort terms needs to be 62 or greater, not exactly 62. And, if some of the columns contain large text or blob values that use more than one byte in their type header, then numbers less than 62 will provoke the problem as well.
So a better characterization of the problem:
drh added on 2015-01-20 18:11:18:
The problem was introduced by the sqlite3VdbeRecordCompare() optimizations on 2014-03-04 checkin [3325ad5bdc2f81f63b55] (version 3.8.4). The KeyInfo.nXField value was not always accurate for emphemeral tables and so the sqlite3VdbeFindCompare() function might return one of the short-circuit comparison routines when in fact the generalized comparison routine was required. The problem became more acute with the introduction of multi-threaded sorting on on 2014-09-01 checkin [b1c0f0bc1bd8a347] (version 3.8.7) because the new sorter creates tables with lots of extra fields after the key.
The problem is fixed for version 126.96.36.199 by ensuring that the KeyInfo.nXField is set correctly. Two new assert() statements are added to verify correct KeyInfo.nXField values moving forward.