Re: generic pseudotype IO functions?

Peter Eisentraut <peter_e@gmx.net> writes:
> On Mon, 2014-01-06 at 17:36 +0100, Andres Freund wrote:
>> FWIW, I am perfectly fine with duplicating the functions for now - I
>> just thought that that might not be the best way but I didn't (and
>> still don't) have a strong opinion.

> Could we just put 0 in for the functions' OID and have code elsewhere
> that errors "there is no input function for this type"?

That doesn't seem like much of an improvement to me: that would be
taking a catalog corruption condition and blessing it as a legitimate
state of affairs, thereby reducing our ability to detect problems.

One instance where it would create issues is that I'm pretty sure
pg_dump would get confused by such a type.  Admittedly, pg_dump will
never try to dump the built-in pseudotypes, but do we really want them
handled so differently from user-definable types?
        regards, tom lane