Re: ROWS FROM(): A Foolish (In)Consistency?

On Mon, Oct 19, 2015 at 10:24:37AM -0700, David Fetter wrote:
> Folks,
>
> As I was learning how best to add native weighted statistics, coming
> soon, I noticed that our ROWS FROM() constructor takes only
> set-returning functions, gluing the outputs together side by side
> without a join condition of any kind. This is a handy capability,
> which I don't find elsewhere in our code, modulo horrible things
> involving WITH ORDINALITY and FULL JOIN.
>
> Did I miss something?
>
> If not, would it make sense to allow every set-returning construct
> inside ROWS FROM(), rather than just some of them?

Based on off-list feedback, this proposal is not as clear as I would
have liked.

Here's what we get with SRFs:

SELECT * FROM ROWS FROM (generate_series(1,5,2),generate_series(4,1,-1)) AS t(i,j);
i | j
---+---
1 | 4
3 | 3
5 | 2
| 1
(4 rows)

Here's a similar construct, but not really, from LATERAL:

SELECT * FROM generate_series(1,5,2) AS s(i) FULL JOIN LATERAL generate_series(4,1,-1) AS t(j) ON s.i=t.j;
i | j
---+---
1 | 1
| 2
3 | 3
| 4
5 |
(5 rows)

What I'd like to do is lift the restriction on ROWS FROM(), which
currently requires that the stuff inside the parentheses set-returning
functions, so constructs something like the following would actually work:

SELECT *
FROM
ROWS FROM (
(VALUES (...), ..., (...)),
(SELECT ... ),
(INSERT ... RETURNING ... ),
my_srf()
)
AS t(...)

would actually work.

What say?

Cheers,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778 AIM: dfetter666 Yahoo!: dfetter
Skype: davidfetter XMPP: david(dot)fetter(at)gmail(dot)com

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate