Re: [PATCH] Improve geometric types

On 09/27/2018 07:21 PM, Alvaro Herrera wrote:
> If you look at the differing results carefully, there's this one:
> 
> *** 3249,3255 ****
> !  [(0,0),(3,0),(4,5),(1,6)] | (-5,-12)          | [(0,-0),(-15,-36),(40,-73),(67,-42)]
> --- 3249,3255 ----
> !  [(0,0),(3,0),(4,5),(1,6)] | (-5,-12)          | [(0,0),(-15,-36),(40,-73),(67,-42)]
> 
> (Third column is first multiplied by second).
> 
> I wonder why the expected file has a -0 only in the second position and
> not both first and second.  These are both positive zeroes being
> multiplied by a negative number.  Why is 0 * -12 = -0  yet  0 * -5 = 0?
> What is going on?  Is the sign suppressed for negative zeros only in the
> first coordinate?  I suppose this is just a side effect of how
> float8_mi, _pl, _mul work (in point_mul_point).
> 
> Anyway maybe your test case should use more of the float8 op
> combinations in order to show the difference.
> 

I may be missing what you're saying, but point_mul_point is not just a
simple multiplication of coordinates, i.e.

    (x1,y1) * (x2,y2) != (x1*x2, y1*y2)

It essentially does this:

    ((x1 * x2 - y1 * y2), (x1 * y2 + x2 * y1))

so I wouldn't be surprised if this was a difference between _pl and _mi.


regards

-- 
Tomas Vondra                  http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services