Li Japin <japinli@hotmail.com> writes:
> Thanks for your confirm. Is there anything I can do?
No, I've got it.
In adding the test coverage I spoke of, I thought we should allow
the date_part tests to check all the entries in timestamp[tz]_tbl
not just those around current time, and I found an independent
problem:
timestamp | isoyear | week | isodow | dow | doy
-----------------------------+-----------+------+--------+-----+-----
...
Tue Feb 16 17:32:01 0097 BC | -96 | 7 | 2 | 2 | 47
Sat Feb 16 17:32:01 0097 | 97 | 7 | 6 | 6 | 47
that is, the ISOYEAR case is failing to correct for BC years.
We could imagine fixing this in date2isoyear() but I think it's
safer to leave that function alone and do the corrections
in timestamp[tz]_part. Note for example that formatting.c
already applies a BC correction to the result; and I think the
usage in date2isoyearday() requires sticking to the year-zero-exists
convention, too.
regards, tom lane
