Thread: Lack of detailed documentation

The following documentation comment has been logged on the website:

Page: https://www.postgresql.org/docs/12/functions-geometry.html
Description:

On:  https://www.postgresql.org/docs/current/functions-geometry.html

The functions aren't defined, nor have pointers.  For instance:
*    Scaling/rotation    box '((0,0),(1,1))' * point '(2.0,0)'

I see how it can be scaling (or shearing), but how does it do rotation?
Or maybe when the points x^2+y^2 == 1, it's interpreted at rotation?
Each operator needs a full explanation...
thx

"David G. Johnston" <david.g.johnston@gmail.com> writes:
> On Fri, Feb 7, 2020 at 4:53 PM PG Doc comments form <noreply@postgresql.org>
> wrote:
>> On:  https://www.postgresql.org/docs/current/functions-geometry.html
>> The functions aren't defined, nor have pointers.  For instance:
>> *       Scaling/rotation        box '((0,0),(1,1))' * point '(2.0,0)'
>> I see how it can be scaling (or shearing), but how does it do rotation?
>> Or maybe when the points x^2+y^2 == 1, it's interpreted at rotation?

> Google: "box point multiplication postgres" came up with the follow book
> section:
> Haven't independently confirm the math but it sounds reasonable...

I responded to the OP about this but unaccountably forgot to cc the list.
The two main points were:

* Per high-school-level vector geometry, rotation around the origin
by an angle theta is equivalent to multiplying by the vector
(cos(theta), sin(theta)).  (So no, the book section you quoted isn't
quite right :-(.)  But we can't go and offer a geometry course on
this page.

* While this works for rotating points, paths, and circles, it
doesn't really work for rotating boxes, because for any rotation
angle that's not a multiple of 90 degrees, you'd need to end up with
a tilted box ... and our box type has no notion of tilted.  The
specified corners rotate to the right places, but the box size isn't
preserved.

So it seems fairly unfortunate to me that the two examples of
scaling/rotation use a box as the LHS.  I could get behind changing
them to use, say, a path as LHS.  But I don't think there's room to
explain how to do rotation if you don't know that already, any more
than (say) our explanation of aggregate functions tries to explain
statistics.

            regards, tom lane

Yes, I figured this out.  People shouldn't have to figure out that the 2 numbers in the vector are the scale and
rotation.
They could be the rotation and scale.  The rotation could be in degrees or radians.
This is documentation- it should explain it.
And yes, does it rotate around its center, or the origin?
And this was just one example.  Every friggin' one of them needs to be documented.
I'm not demanding that they be done right away, but there should be a description of each.  They can all start with
TBD...
-r
Randy Strauss, desk: 650-604-0431 (Bldg N210, Rm 145, Cube 36)

On 2/9/20, 8:09 AM, "Tom Lane" <tgl@sss.pgh.pa.us> wrote:

    "David G. Johnston" <david.g.johnston@gmail.com> writes:
    > On Fri, Feb 7, 2020 at 4:53 PM PG Doc comments form <noreply@postgresql.org>
    > wrote:
    >> On:
https://urldefense.proofpoint.com/v2/url?u=https-3A__www.postgresql.org_docs_current_functions-2Dgeometry.html&d=DwIBAg&c=ApwzowJNAKKw3xye91w7BE1XMRKi2LN9kiMk5Csz9Zk&r=0XRgt7H2WP_TiC5qNss_VhvfCsa-0F_Mw2WUm8TRNiA&m=IDBm4StAmus_A2EYOJRshLFkAAflIS2G74fY7Dhjpag&s=ZuX_8udUWIRdOV5aFxQKHppjdkUgMrhz0mg68HmdTlo&e=

    >> The functions aren't defined, nor have pointers.  For instance:
    >> *       Scaling/rotation        box '((0,0),(1,1))' * point '(2.0,0)'
    >> I see how it can be scaling (or shearing), but how does it do rotation?
    >> Or maybe when the points x^2+y^2 == 1, it's interpreted at rotation?
    
    > Google: "box point multiplication postgres" came up with the follow book
    > section:
    > Haven't independently confirm the math but it sounds reasonable...
    
    I responded to the OP about this but unaccountably forgot to cc the list.
    The two main points were:
    
    * Per high-school-level vector geometry, rotation around the origin
    by an angle theta is equivalent to multiplying by the vector
    (cos(theta), sin(theta)).  (So no, the book section you quoted isn't
    quite right :-(.)  But we can't go and offer a geometry course on
    this page.
    
    * While this works for rotating points, paths, and circles, it
    doesn't really work for rotating boxes, because for any rotation
    angle that's not a multiple of 90 degrees, you'd need to end up with
    a tilted box ... and our box type has no notion of tilted.  The
    specified corners rotate to the right places, but the box size isn't
    preserved.
    
    So it seems fairly unfortunate to me that the two examples of
    scaling/rotation use a box as the LHS.  I could get behind changing
    them to use, say, a path as LHS.  But I don't think there's room to
    explain how to do rotation if you don't know that already, any more
    than (say) our explanation of aggregate functions tries to explain
    statistics.
    
                regards, tom lane
    

Yes, I figured this out.  People shouldn't have to figure out that the 2 numbers in the vector are the scale and rotation.
They could be the rotation and scale.  The rotation could be in degrees or radians.
This is documentation- it should explain it.
And yes, does it rotate around its center, or the origin?
And this was just one example.  Every friggin' one of them needs to be documented.
I'm not demanding that they be done right away, but there should be a description of each.  They can all start with TBD...

"David G. Johnston" <david.g.johnston@gmail.com> writes:
> I doubt anyone disagrees with the sentiment but the last time the lack of
> documentation was brought up here was years ago and no one stepped up then
> to volunteer their time to improve this lesser used area of the database
> and it doesn't seem likely to happen organically here either as this
> message seems to imply a lack of interest in being part of the solution.

Yeah.  Attached is my proposal for clarifying what to do to rotate.
There's a lot more that could be done to improve this page --- for
one thing, most of the other similar tables show the results of the
examples, and for another, it's pretty unclear which data types
each operator takes.  But I don't have the time or interest to
tackle a major rewrite, and so far nobody else has stepped up either.

One thing that's sort of blocking any real progress on this is the
draconian space constraints imposed by the tabular format, which is
hurting us on a lot of these pages, not just this one.  Alvaro did
some preliminary investigation towards finding a better way [1],
but nobody's tried to push that forward.

            regards, tom lane

[1] https://www.postgresql.org/message-id/20200116184444.GA25792%40alvherre.pgsql

diff --git a/doc/src/sgml/func.sgml b/doc/src/sgml/func.sgml
index ceda48e..400b21d 100644
--- a/doc/src/sgml/func.sgml
+++ b/doc/src/sgml/func.sgml
@@ -8363,12 +8363,18 @@ CREATE TYPE rainbow AS ENUM ('red', 'orange', 'yellow', 'green', 'blue', 'purple
        <row>
         <entry> <literal>*</literal> </entry>
         <entry>Scaling/rotation</entry>
-        <entry><literal>box '((0,0),(1,1))' * point '(2.0,0)'</literal></entry>
+        <entry>
+         <para><literal>path '(0,0),(1,0),(1,1)' * point '(3.0,0)'</literal></para>
+         <para><literal>path '(0,0),(1,0),(1,1)' * point(cosd(45), sind(45))</literal></para>
+        </entry>
        </row>
        <row>
         <entry> <literal>/</literal> </entry>
         <entry>Scaling/rotation</entry>
-        <entry><literal>box '((0,0),(2,2))' / point '(2.0,0)'</literal></entry>
+        <entry>
+         <para><literal>path '(0,0),(1,0),(1,1)' / point '(2.0,0)'</literal></para>
+         <para><literal>path '(0,0),(1,0),(1,1)' / point(cosd(45), sind(45))</literal></para>
+        </entry>
        </row>
        <row>
         <entry> <literal>#</literal> </entry>

I wrote:
> One thing that's sort of blocking any real progress on this is the
> draconian space constraints imposed by the tabular format, which is
> hurting us on a lot of these pages, not just this one.  Alvaro did
> some preliminary investigation towards finding a better way,
> but nobody's tried to push that forward.

I've been making an attempt to improve that situation, and along
the way just pushed an expansion of the geometric-operator docs:

https://www.postgresql.org/docs/devel/functions-geometry.html

There's probably still some things to be desired, but it's a lot
less fuzzy than before.

One thing that surprised me is that I couldn't find any well-known
name for what the * and / operators are doing; digging around on
the net and in some dusty old math textbooks didn't yield any exact
matches.  I ended up adding footnotes with the actual computations,
but I'm not very happy with that approach.  Surely Lockhart[1] got this
definition from someplace, though, and didn't invent it out of thin air.

            regards, tom lane

[1] I'd supposed we could blame this stuff on Berkeley, but excavation
in our git history shows it came in at
https://git.postgresql.org/gitweb/?p=postgresql.git&a=commitdiff&h=9e2a87b62db87fc4175b00dabfd26293a2d072fa
... sans documentation of course.

"David G. Johnston" <david.g.johnston@gmail.com> writes:
> It is basically treating the points as if they were a representation of
> complex numbers with x being real and y being imaginary.

Hah!  I was too focused on looking for definitions involving vectors,
and didn't think of plain ol' complex numbers.  But yes, that's exactly
what it is.  I'll go improve the docs.

            regards, tom lane