Re: Proposal: RETURNING primary_key()

On Mon, Mar 7, 2016 at 11:18 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Craig Ringer <craig@2ndquadrant.com> writes:
>> On 8 March 2016 at 08:56, Igal @ Lucee.org <igal@lucee.org> wrote:
>>> I'm not sure why it was not accepted at the end?
>
>> The biggest issue, though it might not be clear from that thread, is that
>> what exactly it means to "return generated keys" is poorly defined by JDBC,
>> and not necessarily the same thing as "return the PRIMARY KEY".
>>
>> Should we return the DEFAULT on a UNIQUE column, for example?
>>
>> IMO other vendors' drivers should be tested for behaviour in a variety of
>> cases.
>
> Yeah.  It was asserted in the earlier thread that other vendors implement
> this feature as "return the pkey", but that seems to conflict with the
> plain language of the JDBC spec: generated columns are an entirely
> different thing than primary key columns.  So really what I'd like to see
> is some work on surveying other implementations to confirm exactly what
> behavior they implement.  If we're to go against what the spec seems to
> say, I want to see a whole lot of evidence that other people do it
> consistently in a different way.

I agree that some research should be done on how this works in other
systems, but I think we have a general problem with the server lacking
certain capabilities that make it easy to implement a high-quality
JDBC driver.  And I think it would be good to work on figuring out how
to fix that.  I feel that some of the replies on this thread were
rather hostile considering that the goal -- good connectors for the
database server -- is extremely important.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company