Thanks. Naturally I had experimented with several versions of the program, and saw that any ordering makes the new feature/error disappear. You can see it if you read the comment at the bottom of my original script. But my purpose was the opposite: demonstrate the regression. I am not interested in this thing. I do not have applications based on SQLite. Simply I think that this feature/error cannot be left in its current state, because this is in contradiction with the axiom, that the result of a query must not depend from the existence of an index. I wanted to help you with this bug report, but I cannot do more for that.
> As Dan already observed, the problem results because you are modifying > an index in the middle of a scan of that index, thereby messing up the > scan. Don't do that. The safest approach is to run the query to > completion, then go back and start the loop over UPDATEs. > > If you add "ORDER BY +rowid" to the query, that forces the query to > run to completion first and then go through the sorter, before you get > any results back, which solves the problem. -- Vermes Mátyás <vermes.mat...@comfirm.hu> _______________________________________________ sqlite-users mailing list sqlite-users@mailinglists.sqlite.org http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users
