Re: Documentation of return values of range functions lower and upper

On Wed, 2023-11-01 at 16:28 -0400, Bruce Momjian wrote:
> On Wed, Nov 18, 2020 at 05:28:44PM +0100, Laurenz Albe wrote:
> > On Wed, 2020-11-18 at 22:49 +0900, Fujii Masao wrote:
> > > On 2020/11/12 17:14, Laurenz Albe wrote:
> > > > On Wed, 2020-11-11 at 18:19 +0100, Laurenz Albe wrote:
> > > > > > Table 9.54 in page
> > > > > > https://www.postgresql.org/docs/current/functions-range.html states that the
> > > > > > functions lower and upper return NULL if the requested bound is infinite. If
> > > > > > the element type of the range contains the special values infinity and
> > > > > > -infinity, this is not correct, as those values are returned if explicitly
> > > > > > used as either bound.
> > > > > +1
> > > > > Perhaps it would be better to say
> > > > > NULL if the range is empty or has no lower/upper bound
> > >
> > > I agree this description looks a bit confusing. But according to the section
> > > "Infinite (Unbounded) Ranges" (*1), we already call "lower/upper bound
> > > omitted" just infinite. So I don't think the current description is incorrect.
> > >
> > > (*1)
> > > https://www.postgresql.org/docs/devel/rangetypes.html#RANGETYPES-INFINITE
> >
> > That is correct, but I'd argue that it would be better to clarify the paragraph too,
> > in particular:
> >
> > The functions lower_inf and upper_inf test for infinite lower and upper bounds of a range, respectively.
> >
> > should better read
> >
> > The functions lower_inf and upper_inf test for omitted lower and upper bounds of a range, respectively.
> >
> > The rest of the paragraph is pretty unambiguous.
> >
> >
> > Independent of this, I think that my patch for "upper" and "lower" would make the
> > documentation clearer.
>
> Yes, I agree this documentation needs help.
>
> For upper/lower(), it is clear that the documentation is better saying
> "unspecified" rather than infinite. The fact that upper/lower_inf()
> returns false for +/-Infinity is quite odd, but should at least be
> documented.
>
> Patch attached. It is odd that +Infinity (vs. Infinity) wasn't
> supported for datetime input until PG 16, but I think we have to say
> +/-infinity vs (blank)/-Infinity.
>
> Patch attached.

I am unhappy with "unspecified". A NULL value as upper or lower bound has a very
specific meaning, namely that the range is unbounded in that direction. This is
a bit confusing, since NULL is typically used for unknown or undefined values.

I think it would be better to say "returns NULL if the range is empty or unbounded"
and "is the range unbounded on the upper end?".

Yours,
Laurenz Albe