SQLite

    ** that do not use indices.  But this nRow reduction only happens if the
    ** table really is the innermost join.  
    **
    ** The second loop iteration is only performed if no optimal scan
    ** strategies were found by the first iteration. This second iteration
    ** is used to search for the lowest cost scan overall.
    **
    ** Previous versions of SQLite performed only the second iteration -
    ** Without the optimal scan step (the first iteration) a suboptimal
    ** the next outermost loop was always that with the lowest overall
    ** cost. However, this meant that SQLite could select the wrong plan
    ** for scripts such as the following:
    ** plan might be chosen for queries like this:
    **   
    **   CREATE TABLE t1(a, b); 
    **   CREATE TABLE t2(c, d);
    **   SELECT * FROM t2, t1 WHERE t2.rowid = t1.a;
    **
    ** The best strategy is to iterate through table t1 first. However it
    ** is not possible to determine this with a simple greedy algorithm.