On 4/12/07, Samuel R. Neff <[EMAIL PROTECTED]> wrote:
Wouldn't implementation dependent mean it's not really standardized? The way I read it the query could still be considered legal in some dbms and not in others (which stinks).
Samuel, That's not what the standard says. It says the name assigned to the result columns are implementation dependent, they could be sqlite_column_1 and sqlite_column_2, or perhaps cnnnnnnn where nnnnnnn is a random number, but they can not be the names of any of the columns in any of the tables in the query. The query should generate a syntax error because it is trying to sort on columns that are not, or at least should not be, present in the result. Besides, the current version of SQLite seems to match on the first tables
names which is consistent with expectations from other databases and not prohibited by the standard (in the way I read it) and backwards compatibility seems to be the most important thing here..
This behavior is prohibited by the standard. Dennis Cote