Re: 9.4 beta - pg_get_viewdef() and WITH CHECK OPTION

On 19 May 2014 02:35, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> writes:
>> On 17 May 2014 13:25, Thomas Kellerer <spam_eater(at)gmx(dot)net> wrote:
>>> when playing with 9.4 beta I noticed that the result of pg_get_viewdef()
>>> will not include the new WITH CHECK OPTION clause when the view was created
>>> using it.
>
>> Yes, that's correct. pg_get_viewdef() only returns the underlying
>> SELECT command for a view. This does not include any of the view's
>> WITH parameters (check option and/or security barrier flag), because
>> they aren't allowed in a SELECT statement.
>
>> The additional parameters are held in pg_class.reloptions, and can be
>> displayed from psql using \d+
>
> I have to concur with the OP that this seems like a pretty darn weird
> design choice. reloptions are for nonstandard PG-specific options, not
> for SQL-spec-mandated syntax. What was the rationale for doing it like
> that?
>

Well I think the question of where to store this option is kind of
independent from the OP's question, which was about what
pg_get_viewdef() should return.

pg_get_viewdef() is currently documented as returning the underlying
SELECT command for the view; it's used in pg_views.definition to show
the "reconstructed SELECT query" and in the view_definition column of
information_schema.views for the same purpose. In that latter case,
there is a separate check_option column to show the value of WITH
CHECK OPTION. So the SQL-spec would appear to mandate that the check
option be kept separate from the view definition, which I think makes
sense, because then the view definition remains a legal SQL SELECT
command.

Thomas Kellerer <spam_eater(at)gmx(dot)net> wrote:
> I do think it would be a good thing to then have something like pg_get_full_viewdef (and a pg_get_full_tabledef() as well)

There was a discussion about adding something like that recently on
-hackers in the context of pg_dump:

http://www.postgresql.org/message-id/flat/CAHyXU0xzs-ow4qyP+Rx8pP_dhtUeReeo3yzB7CmwKF=fv0VDBA(at)mail(dot)gmail(dot)com#CAHyXU0xzs-ow4qyP+Rx8pP_dhtUeReeo3yzB7CmwKF=fv0VDBA@mail.gmail.com

and I agree that there is a strong case for that kind of an API, and
not just for tables and views or for pg_dump, as Merlin points out.
There's still a lot of work to do to get the design right though.

Regards,
Dean