Re: "ERROR: operator is not unique" with Custom Data Type

"David E. Wheeler" <david(at)kineticode(dot)com> writes:
> On Jun 5, 2008, at 11:28, Martijn van Oosterhout wrote:
>> What would you want postgresql to choose in this case.

> I was thinking that the ::text should be cast to ::lctext, as that's
> how `'a'::lctext = 'a'` works, but I keep going back and forth in my
> mind. Maybe 'a'::lctext should not equal 'A'::text.

It seems to me that lctext is sort of like a more-constrained version
of text (like a domain), which suggests that the lctext -> text
direction can be implicit but the other direction should not be.

Moreover, if you don't have lctext -> text be implicit then you
will find that none of the non-comparison text functions work on
lctext except with a cast; which is not the place you want to be.

I concur with Martijn that having both directions implicit is a
Bad Idea.

BTW, I would encourage you to think of this project as citext version 2,
rather than inventing a new name for the datatype. All you'll
accomplish with that is make it hard for users of citext to transition.

regards, tom lane