Re: Implement targetlist SRFs using ROWS FROM() (was Changed SRF in targetlist handling)

Kevin Grittner <kgrittn(at)gmail(dot)com> writes:
> On Fri, Sep 2, 2016 at 9:25 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> ORDER BY is not a useful suggestion when there is nothing
>> you could order by to get the old behavior.

> I'm apparently missing something, because I see a column with the
> header "generate_series" in the result set.

You are apparently only thinking about generate_series and not any
other SRF. Other SRFs don't necessarily produce outputs that are
in a nice sortable order. Even for one that does, sorting by it
would destroy the existing behavior:

regression=# select *, generate_series(1,3) from int8_tbl;
q1 | q2 | generate_series
------------------+-------------------+-----------------
123 | 456 | 1
123 | 456 | 2
123 | 456 | 3
123 | 4567890123456789 | 1
123 | 4567890123456789 | 2
123 | 4567890123456789 | 3
4567890123456789 | 123 | 1
4567890123456789 | 123 | 2
4567890123456789 | 123 | 3
4567890123456789 | 4567890123456789 | 1
4567890123456789 | 4567890123456789 | 2
4567890123456789 | 4567890123456789 | 3
4567890123456789 | -4567890123456789 | 1
4567890123456789 | -4567890123456789 | 2
4567890123456789 | -4567890123456789 | 3
(15 rows)

Now you could argue that the ordering of the table rows
themselves is poorly defined, and you'd be right, but that
doesn't change the fact that the generate_series output
has a well-defined repeating sequence. People might be
relying on that property.

regards, tom lane