On Sun, Jan 19, 2020 at 4:38 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> True for casts involving concrete types, mainly because we'd like
> the identity "value::typename == typename(value)" to hold without
> too much worry about whether the latter is a plain function call
> or a special case. Not sure whether it makes as much sense for
> polymorphics, since casting to a polymorphic type is pretty silly:
> we do seem to allow you to do that, but it's a no-op.
>
> ...
>
> Alternatively, consider this: a cast from some concrete multirange type
> to anymultirange is a no-op, while any other sort of cast probably ought
> to be casting to some particular concrete multirange type. That would
> line up with the existing operations for plain ranges.
I agree you wouldn't actually cast by saying x::anymultirange, and the
casts we define are already concrete, so instead you'd say
x::int4multirange. But I think having a polymorphic function to
convert from an anyrange to an anymultirange is useful so you can
write generic functions. I can see how calling it "anymultirange" may
be preferring the implementor perspective over the user perspective
though, and how simply "multirange" would be more empathetic. I don't
mind taking that approach.
Yours,
Paul
