Do we need multi-column frequency/histogram stats? WAS Re:[HACKERS] Statistics "dependency"

On Sun, Apr 23, 2017 at 10:01:16AM -0400, Bruce Momjian wrote:
> On Sun, Apr 23, 2017 at 11:44:12AM +0100, Simon Riggs wrote:
> > For us "functional dependency" would sound like something to do with
> > functions (e.g. CREATE FUNCTION), so just "dependency" appears to me
> > to be the best term for this.
> > 
> > There are multiple statistics for dependency stored, hence
> > "dependencies". I don't like it, but its the best term I can see at
> > present.
> 
> OK, thank you for the reply, and I am sorry I forgot the previous
> discussion.  I just wanted to re-check we had research this.  Thanks.

(Email subject updated.)

Actually, I have a larger question that I was thinking about.  Because
we already have lots of per-column stats, and now the dependency score,
is it possible to mix the per-column stats and dependency score in a way
that multi-column frequency/histogram stats are not necessary?  That
might be a less costly approach I had not considered.

Or did I miss that discussion too?  ;-)

--  Bruce Momjian  <bruce@momjian.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 +