Tatsuo Ishii <t-ishii@sra.co.jp> writes:
> You are right. However, SQL92 spec allows:
> select * from t1 where i <> all values(0,1);
No, I still disagree. The ANY/ALL constructs compare a single row on
the left side with all the rows produced by the query on the right.
"values(0,1)" is effectively the same as "SELECT 0,1", ie, it is a
query that happens to produce just one row. The above is illegal in
SQL92 because the lefthand side is not a two-item row.
The semantics you seem to be looking for is "scalar IN/NOT IN list",
which we do have. ANY/ALL is a different animal.
It's true that we don't accept values(...) as a subquery, but that's
only one aspect of this example.
regards, tom lane