Using make_timestamp() to create a BC value

[Bryn Llewellyn <bryn@xxxxxxxxxxxx>]

Everything that I say here applies to "make_timestamptz()", too. But my code examples need less typing when I use plain "make_timestamp()".

It would seem to me that ordinary principles of good software design let one expect that the rule that this procedure tests ought to hold:

create procedure assert_make_ts_from_extracted_fields_from_ts_ok(ts in timestamp)
  language plpgsql
as $body$
declare
  year   constant int              not null := extract(year  from ts);
  month  constant int              not null := extract(month from ts);
  mday   constant int              not null := extract(day   from ts);
  hour   constant int              not null := extract(hour  from ts);
  min    constant int              not null := extract(min   from ts);
  sec    constant double precision not null := extract(sec   from ts);
begin
  raise info 'year: %', year;

  declare
    ts_from_extracted_fields constant timestamp not null :=
      make_timestamp(year, month, mday, hour, min, sec);
  begin
    assert (ts_from_extracted_fields = ts), 'assert failed';
  end;
end;

$body$;

Indeed, it does hold—for AD timestamps:

call assert_make_ts_from_extracted_fields_from_ts_ok('0001-01-01 00:00:00 AD');
call assert_make_ts_from_extracted_fields_from_ts_ok(localtimestamp);

The assertion holds for each invocation. And the expected "year" values, "1" and "2021", are reported.

But this invocation makes the assertion fail:

call assert_make_ts_from_extracted_fields_from_ts_ok('0001-01-01 00:00:00 BC');

The 22008 error is reported thus:

date field value out of range: -1-01-01

(And sure enough, "raise info 'year: %', year" reports "-1".)

The doc for "make_timestamp()" here:

www.postgresql.org/docs/14/functions-datetime.html

is totally silent about the limitation that a positive year argument must be used.

This means that I need to implement my own baroque version thus:

create function my_make_timestamp(
  year int, month int, mday int, hour int, min int, sec double precision)
  returns timestamp
  language plpgsql
as $body$
declare
  bc  constant boolean   not null := year < 0 ;
  t   constant timestamp not null := make_timestamp(abs(year), month, mday, hour, min, sec);
begin
  return case bc
           when true then (t::text||' BC')::timestamp
           else           t
         end;
end;
$body$;

If I replace "make_timestamp()" in "assert_make_ts_from_extracted_fields_from_ts_ok()" with "my_make_timestamp()"  then (of course) I get (what I regard as) the proper behavior.

Why must I do this? Am I missing something?