On Wed, Jul 15, 2020 at 09:26:53AM -0700, David G. Johnston wrote:
> On Tue, May 12, 2020 at 8:56 PM Laurenz Albe <laurenz.albe@cybertec.at> wrote:
>
> On Tue, 2020-05-12 at 18:09 -0700, David G. Johnston wrote:
> > Redirecting to -hackers for visibility. I feel there needs to be
> something done here, even if just documentation (a bullet in the usage
> notes section - and a code comment update for the macro)
> > pointing this out and not changing any behavior.
>
> Since "to_date" is an Oracle compatibility function, here is what Oracle
> 18.4 has to say to that:
>
> SQL> SELECT to_date('0000', 'YYYY') FROM dual;
> SELECT to_date('0000', 'YYYY') FROM dual
> *
> ERROR at line 1:
> ORA-01841: (full) year must be between -4713 and +9999, and not be 0
>
>
>
> Attached is a concrete patch (back-patchable hopefully) documenting the current
> reality.
>
> As noted in the patch commit message (commentary really):
>
> make_timestamp not agreeing with make_date on how to handle negative years
> should probably just be fixed - but that is for someone else to handle.
>
> Whether to actually change the behavior of to_date is up for debate though I
> would presume it would not be back-patched.
OK, so, looking at this thread, we have to_date() treating -1 as -2 BC,
make_date() treating -1 as 1 BC, and we have Oracle, which to_date() is
supposed to match, making -1 as 1 BC.
Because we already have the to_date/make_date inconsistency, and the -1
to -2 BC mapping is confusing, and doesn't match Oracle, I think the
clean solution is to change PG 14 to treat -1 as 1 BC, and document the
incompatibility in the release notes.
--
Bruce Momjian <bruce@momjian.us> https://momjian.us
EnterpriseDB https://enterprisedb.com
The usefulness of a cup is in its emptiness, Bruce Lee