On Jun 7, 2010, at 7:53 AM, Tom Lane wrote:
> Andy Balholm <andy@balholm.com> writes:
>> Is there a way to make values of "undefined" type pass through the
>> SELECT DISTINCT filter (getting checked for uniqueness) but remain
>> "undefined"
>=20
> No. What is your criterion for deciding that two values are distinct?
> It's not possible to do that without imputing a data type to them.
> (In particular, comparing the character strings amounts to deciding
> that they are of a textual data type --- failing to acknowledge that
> isn't a workaround but merely sloppy thinking.)
>=20
> regards, tom lane
>=20
I see your point about the fuzziness of deciding what constitutes a distinc=
t value before the type is determined. My proposal was so general that it w=
ould still open a can of worms.
In Farid's particular use case, it's easy to see that the values aren't dis=
tinct, because they're all textually identical. In that case, SELECT DISTIN=
CT could simply ignore that column while deciding which rows are distinct, =
and then tack it back on when it returns its result. That would be a specia=
l-case hack, but I suspect that it would actually cover most cases where un=
defined types are used in SELECT DISTINCT, since undefined types come from =
literals in the SQL, which would generally be the same for all rows. Data t=
hat varies by row would usually come from real tables, where types are alre=
ady defined.