Hi all,
A couple of days ago a bug has showed up regarding rounding of float here:
http://www.postgresql.org/message-id/flat/20150320194337.2573.72944@wrigleys.postgresql.org#20150320194337.2573.72944@wrigleys.postgresql.org
The result being that the version of rint() shipped in src/port was
not IEEE compliant when rounding to even (MSVC < 2013 at least using
it), leading to inconsistent results depending on if the platform uses
src/port's rint() or the platform's one.
During this thread, Tom has raised as well that rounding for numeric
is not that IEEE-compliant:
http://www.postgresql.org/message-id/22366.1427313454@sss.pgh.pa.us
For example:
=# SELECT round(2.5::numeric), round(1.5::numeric),
round(0.5::numeric), round(-2.5::numeric);round | round | round | round
-------+-------+-------+------- 3 | 2 | 1 | -3
(1 row)
=# SELECT round(2.5::float), round(1.5::float), round(0.5::float),
round(-2.5::float);round | round | round | round
-------+-------+-------+------- 2 | 2 | 0 | -2
(1 row)
It sounds appealing to switch the default behavior to something that
is more IEEE-compliant, and not only for scale == 0. Now one can argue
as well that changing the default is risky for existing applications,
or the other way around that other RDBMs (?) are more compliant than
us for their equivalent numeric data type, and people get confused
when switching to Postgres.
An idea, from Dean, would be to have a new specific version for
round() able to do compliant IEEE rounding to even as well...
Opinions?
Regards,
--
Michael