Re: [Fwd: SETOF input parameters (was Re: [HACKERS] proposal:

Tom Lane wrote:
> Joe Conway <mail(at)joeconway(dot)com> writes:
>
>>Tom Lane wrote:
>>
>>>Agreed. The Berkeley code (or more accurately, the descendant code
>>>that's in our source tree) generates the cross product of the rows
>>>output by the SRFs, but I've never understood why that should be a good
>>>approach to take. I could live with just rejecting multiple SRFs in the
>>>same targetlist --- at least till someone comes up with a convincing
>>>semantics for such a thing.
>
>
>>I would like to start spending some time digging in to this. Any pointers or
>>thoughts on the best way to implement it? A little direction might save me
>>days of wheel spinning :-).
>
>
> Implement what exactly?

Well, I want to allow a single table function (or srf if you prefer) in the
target list as discussed above.

Currently, when you try it, record_out() gets called from printtup() when the
srf is encountered, which generates an ERROR. The behavior in 7.2.x is to
return a pointer when the composite type is output. I think that to make this
work as discussed, the target list needs to be "expanded" for the composite
type (similar to expanding a "*" I would think), so I was starting to look at
transformTargetList() and ExpandAllTables().

>
> The code that presently does the dirty work is in ExecTargetList(), if
> that's what you're looking for...
>

OK -- i'll check that out too.

Thanks,

Joe