Tom Lane wrote:
> Joe Conway <mail@joeconway.com> writes:
>
>>So array[] should produce '{}' of (an array) type determined by the
>>context? OK -- seems easy enough.
>
> Is it? I think we'd decided that this could only reasonably be handled
> by creating a datatype representing array-of-UNKNOWN. I'm afraid to do
> that because I think it might allow the parser's type resolution
> algorithms to follow paths we will not like. Perhaps it can be made to
> work, but I think it will require some careful study.
I took a closer look -- yeah, without array-of-UNKNOWN I don't think we
can make this work.
I got something working by forcing the element type to UNKNOWN when the
elements list is empty in transformExpr(), but then select_common_type()
turns around and turns UNKNOWN into TEXT, so you wind up with an empty
text[].
I won't bother sending that patch in because I *know* it will get
rejected ;-)
I guess we should put array-of-UNKNOWN on the list of things to look at
for 7.5.
Joe