Re: Window function docs

On Thu, Apr 6, 2017 at 06:13:44PM -0400, Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > As part of writing a talk about window functions, I have done some
> > cleanups of the window function docs, attached, that I would like to
> > apply to head.
>
> I think this change is just wrong:
>
> aggregate function, but the <literal>OVER</> clause causes it to be
> - treated as a window function and computed across an appropriate set of
> - rows.)
> + treated as a window function and computed across the entire partition.)
> </para>
>
> avg() is going to be evaluated across the current frame, not the whole
> partition. I don't mind trying to be more precise than "appropriate set",
> but it doesn't help to be apparently precise but wrong.

Good point. I have adjusted the text to say "window frame".

> Not really convinced by the s/regular/normal/g changes. That doesn't seem
> to clarify much either. Maybe it would help to define "normal" near here.
> There's a definition in func.sgml (that they are not ordered-set or
> hypothetical-set aggregates) but that's too far away.

Yeah, I went back and forth on that, and it seems like a mine-field.
The problem is that SUM() is both a non-window aggregate, and a window
function, and there isn't a clear terminology for non-window aggregates,
so changed the patch to just call them that. Hopefully the dual nature
of this is now clearer.

> In this bit:
>
> - definition. Rows that are not distinct in the <literal>ORDER BY</>
> - ordering are said to be <firstterm>peers</>; the four ranking functions
> - are defined so that they give the same answer for any two peer rows.
> + definition. Rows that are not distinct based on <literal>ORDER BY</>
> + ordering are said to be <firstterm>peers</>. The four ranking
> + functions (includes <function>cume_dist</>) are defined so that they
> + give the same answer for any two peer rows.
>
> "based on" seems little better than the previous wording. Maybe it'd be
> better to say "Rows that are not distinct when considering only the
> <literal>ORDER BY</> columns are said to be ..."

OK, I used your wording.

> sometimes also <function>nth_value</>. You can redefine the frame by
> - adding a suitable frame specification (<literal>RANGE</> or
> - <literal>ROWS</>) to the <literal>OVER</> clause.
> - See <xref linkend="syntax-window-functions"> for more information
> - about frame specifications.
> + modifying the frame specification; see <xref
> + linkend="syntax-window-functions"> for more information about frame
> + specifications.
> </para>
>
> Don't really like this change; it makes it vaguer for no apparent benefit.

Reverted. I had changed it because it is more the frame definition that
we want to highlight here, not ROW vs. RANGE, but there is no clearer
way to specify that clause.

> + <para>
> + <function>cume_dist</> computes the ratio of partition rows that
> + are less than or equal to the current row and its peers, while
> + <function>percent_rank</> computes the ratio of lesser partition rows,
> + assuming the current row is does not exist in the partition.
> + </para>
>
> The first one probably ought to be "computes the fraction of partition
> rows". The second definition seems nonsensical --- ratio to what? Or
> probably better, fraction of what? (And "is does not" needs help too.)

Updated.

> the user. Also, any built-in or user-defined normal aggregate function
> - can be used as a window function. Ordered-set aggregates presently
> - cannot be used as window functions, however.
> + can be used as a window function. However, ordered-set aggregate
> + cannot presently be used as window functions.
> </para>
>
> Grammar (don't lose the "s" in ordered-set aggregates).

Thanks for the review. Updated patch attached.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +