Re: Re: BUG #12885: The result of casting a double to an integer depends on the database version

On Thu, Mar 26, 2015 at 11:15 AM, Michael Paquier
<michael(dot)paquier(at)gmail(dot)com> wrote:
> On Thu, Mar 26, 2015 at 11:12 AM, Andrew Gierth wrote:
>>>>>>> "Michael" == Michael Paquier writes:
>>
>> Michael> Indeed...
>> Michael> =# select column1 AS double_value, cast(column1 AS INT) AS
>> Michael> int_value
>>
>> The complication for numeric is that there's also the case of round(x,n)
>> and casting to numeric(m,n) to consider.
>
> OK, thanks for the reminder... I forgot this case.

Attached is a patch that adds nearest-to-even rounding for numeric
when dscale = 0. This gives the following results with the previous
query:
=# select column1 AS double_value, cast(column1 AS INT) AS int_value
FROM (VALUES
(-2.5::numeric),

(-1.5::numeric),

(-0.5::numeric),

(0.5::numeric),

(1.5::numeric),

2.5::numeric)) t;
double_value | int_value
--------------+-----------
-2.5 | -2
-1.5 | -2
-0.5 | 0
0.5 | 0
1.5 | 2
2.5 | 2
(6 rows)

For the case where dscale != 0 case, what would we actually want? I
have been staring at the code around round_var for some time but I
could not really get if we'd really want to change something there...
For example round(2.5::numeric, 1) should be 2.4 and not 2.5?
Regards.
--
Michael