Bruce Momjian <pgman@candle.pha.pa.us> writes:
>> I still prefer the suggestion I made before: complain only if the
>> implicit FROM entry is for a table already present in the rangelist
>> (under a different alias, obviously). The fact that that choice
>> would not break any existing regression tests seems relevant...
> But it seems mine is going to complain if they forget one in a FROM
> clause, which sort of makes sense to me.
Seems like the real question is what is the goal of having the warning.
Are we (a) trying to nag people into writing their queries in an
SQL-compliant way, or are we (b) trying to warn about probable mistakes
while still considering implicit FROM entries as a fully supported
Postgres feature?
If the goal is (a) then your way is better, but I like mine if the goal
is (b). Seems like some discussion is needed here about just what we
want to accomplish.
regards, tom lane