On Friday, October 6, 2023, Tom Lane <
tgl@sss.pgh.pa.us> wrote:
"David G. Johnston" <david.g.johnston@gmail.com> writes:
>> On 10/6/23 08:45, Ron wrote:
>>> Nah. "The programmer -- and DBA -- on the Clapham omnibus" quite
>>> reasonably expects that COPY table_name TO (output)" copies all the
>>> columns listed in "\d table_name".
> Sure, but it doesn't. Mainly since copy's original design was intended to
> solve the dump/restore problem and it doesn't make sense to specify data
> for inbound generated data. So while we do have a POLA violation here the
> desirability to now fix it years later is basically zero. And the current
> behavior is at least defensible and consistent. And there is a very easy
> way to get the desired output making any change that much harder a sell.
Changing the default behavior now is certainly a non-starter.
I don't really see any backwards-compatibility problem with
allowing cases that had been errors, though.
I wouldn’t vote against it but the current simplicity seems sufficient. “Copy table doesn’t recognize generated columns, use copy (select) if you want to include them in the output.”
David J.