On Tue, Jan 22, 2013 at 11:02:18PM -0500, Tom Lane wrote:
> David Fetter <david@fetter.org> writes:
> > I've been working with Andrew Gierth (well, mostly he's been doing
> > the work, as usual) to add WITH ORDINALITY as an option for
> > set-returning functions. In the process, he found a minor
> > opportunity to clean up the interface for $SUBJECT, reducing the
> > call to a Single Point of Truth for lateral-ness, very likely
> > improving the efficiency of calls to that function.
>
> As I mentioned in our off-list discussion, I think this is going in
> the wrong direction. It'd make more sense to me to get rid of the
> RangeFunction parameter, instead passing the two fields that
> addRangeTableEntryForFunction actually uses out of that.
With utmost respect, of the four fields currently in RangeFunction:
type (tag), lateral, funccallnode, alias, and coldeflist, the function
needs three (all but funccallnode, which has already been transformed
into a funcexpr). The patch for ordinality makes that 4/5 with the
ordinality field added.
> If we do what you have here, we'll be welding together the alias and
> lateral settings for the new RTE; which conceivably some other
> caller would want to specify in a different way. As a comparison
> point, you might want to look at the various calls to
> addRangeTableEntryForSubquery: some of those pass multiple fields
> out of the same RangeSubselect, and some do not.
As to addRangeTableEntryForSubquery, I'm not seeing the connection to
the case at hand. Could you please spell it out?
Cheers,
David.
--
David Fetter <david@fetter.org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david.fetter@gmail.com
iCal: webcal://www.tripit.com/feed/ical/people/david74/tripit.ics
Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate