Hackers,
I've just found a little bug: extracting "epoch" from the last 30
years before Postgres' "+Infinity" leads an integer overflow:
postgres=# SELECT x::timestamptz, extract(epoch FROM x::timestamptz)
postgres-# FROM
postgres-# (VALUES
postgres(# ('294247-01-10 04:00:54.775805'),
postgres(# ('294247-01-10 04:00:55.775806'),
postgres(# ('294277-01-09 04:00:54.775806'), -- the last value before 'Inf'
postgres(# ('294277-01-09 04:00:54.775807') -- we've discussed, it
should be fixed
postgres(# ) as t(x);
x | date_part
---------------------------------+-------------------
294247-01-10 04:00:54.775805+00 | 9223372036854.78
294247-01-10 04:00:55.775806+00 | -9223372036853.78
294277-01-09 04:00:54.775806+00 | -9222425352054.78
infinity | Infinity
(4 rows)
With the attached patch it becomes positive:
x | date_part
---------------------------------+------------------
294247-01-10 04:00:54.775805+00 | 9223372036854.78
294247-01-10 04:00:55.775806+00 | 9223372036855.78
294277-01-09 04:00:54.775806+00 | 9224318721654.78
infinity | Infinity
(4 rows)
--
Best regards,
Vitaly Burovoy