On 1/19/15, RSmith <rsm...@rsweb.co.za> wrote: > Might this not be a "reverse_unordered_selects" pragma or compile option > going wrong, or at least the code making it work getting > somehow hooked in the new versions for this query? >
It looks like a bug. If you update to the latest trunk check-in and set: .testctrl never_corrupt 1 prior to running the test query, you get an assertion fault. That new dot-command invokes sqlite3_test_control(SQLITE_TESTCTRL_NEVER_CORRUPT, 1) which tells SQLite that no database it deals with will ever be corrupt. And this, in turn, enables some additional assert() statements which are normally turned off. One of those new asserts finds the incorrect logic. We are working on a simple test case and a fix now. It is a very complex problem. In particular, the sample query works fine as long as the number of columns in the result set is not exactly 60. Adding or removing a single column of result gives the correct answer. -- D. Richard Hipp d...@sqlite.org _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users