On Jul 15, 2010, at 11:59 AM, Simon Riggs <simon@2ndQuadrant.com> wrote:
> On Thu, 2010-07-15 at 18:43 +0200, Magnus Hagander wrote:
>> On Thu, Jul 15, 2010 at 18:35, Simon Riggs <simon@2ndquadrant.com> wrote:
>>> On Thu, 2010-07-15 at 17:38 +0200, Magnus Hagander wrote:
>>>
>>>> Is there an actual common use-case for having these commands available
>>>> for *non-psql* interfaces?
>>>
>>> There are many interfaces out there and people writing new ones
>>> everyday. We just wrote an interface for Android, for example.
>>>
>>> It is arguably *more* important to do this from non-psql interfaces.
>>>
>>> There should be one command to "display a list of tables" and it needs
>>> to be easily guessable for those who have forgotten.
>>
>> The downside is that you are then limited to what can be returned as a
>> resultset. A "\d table" in psql returns a hell of a lot more than
>> that. So do we keep two separate formats for this? Or do we remove the
>> current, useful, output format in favor of a much worse formt just to
>> support more clients?
>
> I imagined that we would do something similar to EXPLAIN, a set of text
> rows returned.
That seems rather wretched for machine-parsability, which I think is an important property for anything we do in this
area. We need to think harder about how we could structure this to allow returning more than just a tabular result set
whilestill allowing clients easy programmatic access to the underlying data.
> It should be possible to migrate \d options to using new outputs, when
> everything works in a useful manner. Probably not in this release.
>
> If I get some working solutions ready for Sept 15 we then have 4 months
> for other people to patch away at this.
Sounds good, but we need agreement on a more detailed design first.
...Robert