Обсуждение: unnest multirange, returned order

The following documentation comment has been logged on the website:

Page: https://www.postgresql.org/docs/16/functions-range.html
Description:

The doc says:
* unnest ( anymultirange ) → setof anyrange
* Expands a multirange into a set of ranges. The ranges are read out in
storage order (ascending).

What is storage order ? 

At first I thought that it was the order in which the different ranges are
inserted in the internal data structure. However, the following sort of
shows that it is not:
```
postgres=# select unnest('{[1,4), [8,10)}'::int4multirange + '{[-5,-3)}' -
'{[2,3)}') ;
 unnest
---------
 [-5,-3)
 [1,2)
 [3,4)
 [8,10)
(4 lignes)
```
Whatever I try, it always return in range order instead of "storage order".


Some context: I ask because we have some seemingly random (and impossible to
repro in tests up to now) errors in our code. The code assumes that this
returns things in range order and as the doc is unclear to me on this point,
I cannot exclude this to be our culprit.

Thank you

On Mon, 2023-10-02 at 18:42 +0000, PG Doc comments form wrote:
> Page: https://www.postgresql.org/docs/16/functions-range.html
>
> The doc says:
> * unnest ( anymultirange ) → setof anyrange
> * Expands a multirange into a set of ranges. The ranges are read out in
> storage order (ascending).
>
> What is storage order ?
>
> At first I thought that it was the order in which the different ranges are
> inserted in the internal data structure. However, the following sort of
> shows that it is not:
> ```
> postgres=# select unnest('{[1,4), [8,10)}'::int4multirange + '{[-5,-3)}' -
> '{[2,3)}') ;
>  unnest
> ---------
>  [-5,-3)
>  [1,2)
>  [3,4)
>  [8,10)
> (4 lignes)
> ```
> Whatever I try, it always return in range order instead of "storage order".

I'd say that the storag order is the order in which PostgreSQL stores
multiranges internally:

SELECT '{[100,200),[-100,-50),[-1,2)}'::int4multirange;

        int4multirange
═══════════════════════════════
 {[-100,-50),[-1,2),[100,200)}
(1 row)

Yours,
Laurenz Albe

On Tue, 2023-10-03 at 20:40 -0400, Daniel Fredouille wrote:
> > I'd say that the storag order is the order in which PostgreSQL stores
> > multiranges internally:
>
> Right, I believe that you are right but then this information is not useful for the developer. 
> If storage order is always ascending by range order then let's make it clear,
> if order cannot be counted upon as it may evolve from postgres version to version,
> then let's make it clear as well. WDYT ?

I personally think that it is clear as it is written now.

If you have a good suggestion for an improvement, you could send it;
perhaps someone will pick it up.

Yours,
Laurenz Albe

On Wed, 2023-10-04 at 20:12 -0400, Daniel Fredouille wrote:
> unnest ( anymultirange ) → setof anyrange
> Expands a multirange into a set of ranges. The ranges are read out in storage order (ascending) and therefore order
cannot berelied upon. 

That's not true.  The order is deterministic and can be relied on.

How about the attached patch, which does away with the confusing
mention of "storage order"?

Yours,
Laurenz Albe