Tom> That would eliminate the backward-compatibility problem
Tom> pretty well (since the constraints aren't documented and
Tom> hence aren't being used now), while not posing a big problem
Tom> for ISO cases (since if there's no constraint there are no
Tom> ambiguous cases, I believe --- the ISO syntax would require
Tom> all fields to be present).
The 'constraint' (interval type descriptor or whatever it's really
called) is mandatory in standard SQL, I think, so there's no ambiguity
anyway, unless anyone is using this undocumented syntax at the
moment...