[sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Bert Huijben
Hi, As part of the Subversion 1.8.6 release we tried introducing some data in the 'sqlitstat_stat1' table using the recommended approach for Sqlite 3.8.0+ compatibility to tell sqlite about our 'bad indexes': [[ ANALYZE sqlite_master; INSERT INTO "sqlite_stat1" VALUES INSERT INTO

Re: [sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Richard Hipp
On Mon, Feb 10, 2014 at 9:02 PM, Richard Hipp wrote: > > Your work-arounds pending the patch release: > > (1) Do not compile with SQLITE_ENABLE_STAT3 or SQLITE_ENABLE_STAT4 > (2) Change your query so that it says "+moved_to IS NOT NULL" - add a > unary "+" operator before every

Re: [sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Richard Hipp
Thanks for the bug report. This was a real problem. It is now fixed on the SQLite trunk ( http://www.sqlite.org/src/info/c950d6c411). It will be a few days before we can get a patch release (3.8.3.1) together. Your work-arounds pending the patch release: (1) Do not compile with

Re: [sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Richard Hipp
Ticket here: http://www.sqlite.org/src/info/4c86b126f2 The work-around until a fix is available is to avoid compiling with SQLITE_ENABLE_STAT3 or SQLITE_ENABLE_STAT4. On Mon, Feb 10, 2014 at 4:38 PM, Richard Hipp wrote: > STAT3 should never change the answer. It should only

Re: [sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Richard Hipp
STAT3 should never change the answer. It should only help the answer to appear faster. The fact that the queries gives different answers with and without STAT3 clearly indicates a bug. We are working the problem now. Thanks for providing an simplified test case. On Mon, Feb 10, 2014 at 2:05

Re: [sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Petite Abeille
On Feb 10, 2014, at 8:05 PM, Bert Huijben wrote: > As part of the Subversion 1.8.6 release we tried introducing some data in > the 'sqlitstat_stat1' table using the recommended approach for Sqlite 3.8.0+ > compatibility to tell sqlite about our 'bad indexes’: ( Not

[sqlite] Simple SELECT misbehaving in Sqlite 3.8.3 with ENABLE_STAT3

2014-02-10 Thread Bert Huijben
[Retrying with the user I subscribed with to avoid the moderation] Hi, As part of the Subversion 1.8.6 release we tried introducing some data in the 'sqlitstat_stat1' table using the recommended approach for Sqlite 3.8.0+ compatibility to tell sqlite about our 'bad indexes': [[ ANALYZE