From: Tom Lane <tgl@sss.pgh.pa.us>
To: pgsql-hackers@lists.postgresql.org
Subject: Bogus documentation for bogus geometric operators
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <24347.1587444160.1@sss.pgh.pa.us>
Date: Tue, 21 Apr 2020 00:42:40 -0400
Message-ID: <24348.1587444160@sss.pgh.pa.us>
Precedence: bulk

While revising the docs for the geometric operators, I came across
these entries:

<^	Is below (allows touching)?	circle '((0,0),1)' <^ circle '((0,5),1)'
>^	Is above (allows touching)?	circle '((0,5),1)' >^ circle '((0,0),1)'

These have got more than a few problems:

1. There are no such operators for circles, so the examples are pure
fantasy.

2. What these operators do exist for is points (point_below, point_above
respectively) and boxes (box_below_eq, box_above_eq).  However, the
stated behavior is not what the point functions actually do:

point_below(PG_FUNCTION_ARGS)
...
	PG_RETURN_BOOL(FPlt(pt1->y, pt2->y));

point_above(PG_FUNCTION_ARGS)
...
	PG_RETURN_BOOL(FPgt(pt1->y, pt2->y));

So point_below would be more accurately described as "is strictly below",
so its operator name really ought to be <<|.  And point_above is "is
strictly above", so its operator name ought to be |>>.

3. The box functions do seem to be correctly documented:

box_below_eq(PG_FUNCTION_ARGS)
...
	PG_RETURN_BOOL(FPle(box1->high.y, box2->low.y));

box_above_eq(PG_FUNCTION_ARGS)
...
	PG_RETURN_BOOL(FPge(box1->low.y, box2->high.y));

But there are comments in the source code to the effect of

 * box_below_eq and box_above_eq are obsolete versions that (probably
 * erroneously) accept the equal-boundaries case.  Since these are not
 * in sync with the box_left and box_right code, they are deprecated and
 * not supported in the PG 8.1 rtree operator class extension.

I'm not sure how seriously to take this deprecation comment, but it
is true that box_below (<<|) and box_above (|>>) have analogs for
other data types while these functions don't.

4. Just for extra fun, these point operators are listed in some
GIST and SP-GIST opclasses; though the box ones are not, as per
that code comment.

Perhaps it's too late in the v13 cycle to actually do anything
about this code-wise, but what should I do documentation-wise?
I'm certainly not eager to document that these operators behave
inconsistently depending on which type you're talking about.

			regards, tom lane