The following bug has been logged on the website:
Bug reference: 16357
Logged by: Ivan Beloborodov
Email address: ivbel@ispras.ru
PostgreSQL version: 9.6.17
Operating system: Linux 5.0.0, Ubuntu 18.04.1
Description:
to_timestamp() returns incorrect results for some negative arguments:
set extra_float_digits to 1;
SET
select extract(epoch from timestamp '1920-10-24 12:11:13.975227');
date_part
--------------------
-1552218526.024773
(1 row)
select to_timestamp(-1552218526.024773) at time zone 'UTC';
timezone
----------------------------
1920-10-24 12:11:13.975228
(1 row)
Whereas for positive arguments of the same magnitude everything's ok:
select extract(epoch from timestamp '2019-03-10 11:48:46.024773');
date_part
-------------------
1552218526.024773
(1 row)
select to_timestamp(1552218526.024773) at time zone 'UTC';
timezone
----------------------------
2019-03-10 11:48:46.024773
(1 row)
The problem is not with extra_float_digits:
set extra_float_digits to 0;
SET
select to_timestamp(-1552218526.024773) at time zone 'UTC' = timestamp
'1920-10-24 12:11:13.975227';
?column?
----------
f
(1 row)
And, as far as I understand, not with insufficient precision of float8, here
is an example with the same behaviour with 15 significant digits:
select extract(epoch from timestamp '1853-08-15 10:04:15.35064');
date_part
-------------------
-3672568544.64936
(1 row)
select to_timestamp(-3672568544.64936) at time zone 'UTC';
timezone
----------------------------
1853-08-15 10:04:15.350639
(1 row)
select extract(epoch from timestamp '2086-05-18 13:55:44.64936');
date_part
------------------
3672568544.64936
(1 row)
select to_timestamp(3672568544.64936) at time zone 'UTC';
timezone
---------------------------
2086-05-18 13:55:44.64936
(1 row)
Tested this behaviour on PostgreSQL 9.6.17, Linux 5.0.0, Ubuntu 18.04.1 and
PostgreSQL 10.1, Linux 4.4.0, Ubuntu 14.04.1