Re: identity columns

From: Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com>
To: Robert Haas <robertmhaas(at)gmail(dot)com>
Cc: Vitaly Burovoy <vitaly(dot)burovoy(at)gmail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: identity columns
Date: 2017-04-27 19:42:02
Message-ID: 0334bd69-ab72-384a-684d-4595d761beea@2ndquadrant.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 4/27/17 10:03, Robert Haas wrote:
>> So we could make up new syntax
>>
>> ALTER TABLE t1 ALTER COLUMN c1 SET GENERATED ALWAYS AS IDENTITY;
>>
>> and let that be set-or-add, but then the argument for staying within the
>> SQL standard goes out the window.
>
> What does the SQL standard actually mandate here? It's not clear to
> me which parts of this syntax are mandated by the standard and which
> we just made up. I'm gathering (perhaps incorrectly) that you made
> ALTER TABLE ... DROP IDENTITY as per standard, but the reverse
> operation of setting the identity non-standard syntax. If that's so,
> it seems like a questionable choice.

Standard syntax:

1. Add new column with identity:

ALTER TABLE t1 ADD COLUMN c1 int GENERATED ALWAYS AS IDENTITY;

(or equivalent for CREATE TABLE)

2. Change ALWAYS to BY DEFAULT (or vice versa) of existing column:

ALTER TABLE t1 ALTER COLUMN c1 SET GENERATED BY DEFAULT;

3. Change sequence parameters of existing column:

ALTER TABLE t1 ALTER COLUMN c1 SET (START 2)

(the previous two can be combined)

4. Drop identity property of existing column:

ALTER TABLE t1 ALTER COLUMN c1 DROP IDENTITY;

(with IF EXISTS being our extension)

There is no standard syntax for the inverse, that is, adding an identity
property to an existing column. (I have checked DB2 and Oracle. They
don't have anything either. One even documents that explicitly.)
Therefore ...

Nonstandard syntax:

5. Add identity property to existing column:

ALTER TABLE t1 ALTER COLUMN c1 ADD GENERATED ALWAYS AS IDENTITY;

The competing proposal is that we should not have syntax #5 but that
syntax #2 should (also) do that.

My concerns about that, as previously explained, are

- asymmetric idempotency behavior with DROP IDENTITY

- ambiguous/inconsistent behavior when sequence options are specified in
same command

- syntax ambiguity/inconsistency with other SQL standard features

The argument in favor is that syntax #5 is nonstandard. But of course
making #2 doing something nonstandard is also nonstandard.

--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2017-04-27 20:02:52 Re: Partition-wise join for join between (declaratively) partitioned tables
Previous Message Rahila Syed 2017-04-27 19:41:21 Re: Adding support for Default partition in partitioning