SQLite

Consider the following SQL:

CREATE TABLE t1(m CHAR(2));
INSERT INTO t1 VALUES('ax');
INSERT INTO t1 VALUES('bx');
INSERT INTO t1 VALUES('cy');
SELECT '1', count(*), substr(m,2,1) AS m FROM t1 GROUP BY m;
SELECT '2', count(*), substr(m,2,1) AS mx FROM t1 GROUP BY m;

Both queries should give the same three-row result. But in SQLite, the first query gives a two-row results where one row has a count of 2. The expression of a GROUP BY clause should bind preferentially to the input table, not the output table.

It has come to light that some legacy Android applications are coded with queries that are logically equivalent to this:

CREATE TABLE t1(a,x);
CREATE TABLE t2(a,y);
SELECT t1.a AS a, x, y FROM t1, t2 GROUP BY a;

The original fix for this ticket (see check-in [f2d175f975cd]) caused such queries to give an error: "ambiguous column: a". The query is indeed incorrect. Nevertheless, we need to find a way to get SQLite to run and return the correct result even for these incorrect, legacy queries.