Обсуждение: BUG #19367: typos in backend/utils/adt/timestamp.c

The following bug has been logged on the website:

Bug reference:      19367
Logged by:          Nikolay Fadeev
Email address:      fadeymail@rambler.ru
PostgreSQL version: 18.1
Operating system:   all OS
Description:

There are typos in return type of these functions:
1) timestamptz_pl_interval_at_zone(PG_FUNCTION_ARGS)
    NOW: PG_RETURN_TIMESTAMP(timestamptz_pl_interval_internal(timestamp,
span, attimezone));
    SHOULD:
PG_RETURN_TIMESTAMPTZ(timestamptz_pl_interval_internal(timestamp, span,
attimezone));
2) Datum timestamptz_mi_interval_at_zone(PG_FUNCTION_ARGS)
    NOW: PG_RETURN_TIMESTAMP(timestamptz_mi_interval_internal(timestamp,
span, attimezone));
    SHOULD:
PG_RETURN_TIMESTAMPTZ(timestamptz_mi_interval_internal(timestamp, span,
attimezone));

On Mon, 29 Dec 2025 at 15:55, Rahila Syed <rahilasyed90@gmail.com> wrote:
> Hi,
>
>  There are typos in return type of these functions:
>  1) timestamptz_pl_interval_at_zone(PG_FUNCTION_ARGS)
>      NOW: PG_RETURN_TIMESTAMP(timestamptz_pl_interval_internal(timestamp,
>  span, attimezone));
>      SHOULD:
>  PG_RETURN_TIMESTAMPTZ(timestamptz_pl_interval_internal(timestamp, span,
>  attimezone));
>  2) Datum timestamptz_mi_interval_at_zone(PG_FUNCTION_ARGS)
>      NOW: PG_RETURN_TIMESTAMP(timestamptz_mi_interval_internal(timestamp,
>  span, attimezone));
>      SHOULD:
>  PG_RETURN_TIMESTAMPTZ(timestamptz_mi_interval_internal(timestamp, span,
>  attimezone));
>
> You’re right — these are just typos, and they don’t affect correctness since both
> ultimately call Int64GetDatum().
> Still, +1 for fixing them for clarity.

The functions timestamptz_pl_interval() and timestamptz_mi_interval() have the
same typos, right?

--
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.

29.12.2025, 13:37, Japin Li <japinli@hotmail.com>

On Mon, 29 Dec 2025 at 15:55, Rahila Syed <rahilasyed90@gmail.com> wrote:
> Hi,
>
> There are typos in return type of these functions:
> 1) timestamptz_pl_interval_at_zone(PG_FUNCTION_ARGS)
>     NOW: PG_RETURN_TIMESTAMP(timestamptz_pl_interval_internal(timestamp,
> span, attimezone));
>     SHOULD:
> PG_RETURN_TIMESTAMPTZ(timestamptz_pl_interval_internal(timestamp, span,
> attimezone));
> 2) Datum timestamptz_mi_interval_at_zone(PG_FUNCTION_ARGS)
>     NOW: PG_RETURN_TIMESTAMP(timestamptz_mi_interval_internal(timestamp,
> span, attimezone));
>     SHOULD:
> PG_RETURN_TIMESTAMPTZ(timestamptz_mi_interval_internal(timestamp, span,
> attimezone));
>
> You’re right — these are just typos, and they don’t affect correctness since both
> ultimately call Int64GetDatum().
> Still, +1 for fixing them for clarity.

The functions timestamptz_pl_interval() and timestamptz_mi_interval() have the
same typos, right?

--
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.

Japin Li <japinli@hotmail.com> writes:
> On Mon, 29 Dec 2025 at 15:55, Rahila Syed <rahilasyed90@gmail.com> wrote:
>>> There are typos in return type of these functions:

>> You’re right — these are just typos, and they don’t affect correctness since both
>> ultimately call Int64GetDatum().
>> Still, +1 for fixing them for clarity.

> The functions timestamptz_pl_interval() and timestamptz_mi_interval() have the
> same typos, right?

My recollection is that this code deliberately fuzzes the difference
between timestamp and timestamptz in many places.  An example is that
timestamp_eq and its sibling comparison functions are used as-is for
both timestamp and timestamptz --- note the comment in front of
timestamp_cmp_internal in timestamp.c.  (BTW, "thomas" there is
Lockhart not me.)

So I think that being cavalier about PG_RETURN_TIMESTAMP versus
PG_RETURN_TIMESTAMPTZ stems from that; in fact, if you go back
far enough in our git history you will find we didn't even have
the latter to begin with.

There may be places where we can clean this up and not simply be
reversing the direction in which we're type-punning, but I doubt
we'll be able to fix every one without duplicating functions.
So I can't get hugely excited about it.

            regards, tom lane

On Mon, 29 Dec 2025 at 10:37, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Japin Li <japinli@hotmail.com> writes:
>> On Mon, 29 Dec 2025 at 15:55, Rahila Syed <rahilasyed90@gmail.com> wrote:
>>>> There are typos in return type of these functions:
>
>>> You’re right — these are just typos, and they don’t affect correctness since both
>>> ultimately call Int64GetDatum().
>>> Still, +1 for fixing them for clarity.
>
>> The functions timestamptz_pl_interval() and timestamptz_mi_interval() have the
>> same typos, right?
>
> My recollection is that this code deliberately fuzzes the difference
> between timestamp and timestamptz in many places.  An example is that
> timestamp_eq and its sibling comparison functions are used as-is for
> both timestamp and timestamptz --- note the comment in front of
> timestamp_cmp_internal in timestamp.c.  (BTW, "thomas" there is
> Lockhart not me.)
>
> So I think that being cavalier about PG_RETURN_TIMESTAMP versus
> PG_RETURN_TIMESTAMPTZ stems from that; in fact, if you go back
> far enough in our git history you will find we didn't even have
> the latter to begin with.
>
> There may be places where we can clean this up and not simply be
> reversing the direction in which we're type-punning, but I doubt
> we'll be able to fix every one without duplicating functions.
> So I can't get hugely excited about it.
>

After greping, I found the following:

  $ grep -e '^timestamp_pl_interval' -e '^timestamptz_pl_interval' -rn src/
  src/backend/utils/adt/timestamp.c:3090:timestamp_pl_interval(PG_FUNCTION_ARGS)
  src/backend/utils/adt/timestamp.c:3233:timestamptz_pl_interval_internal(TimestampTz timestamp,
  src/backend/utils/adt/timestamp.c:3380:timestamptz_pl_interval(PG_FUNCTION_ARGS)
  src/backend/utils/adt/timestamp.c:3401:timestamptz_pl_interval_at_zone(PG_FUNCTION_ARGS)
  $ grep -e '^timestamp_mi_interval' -e '^timestamptz_mi_interval' -rn src/
  src/backend/utils/adt/timestamp.c:3207:timestamp_mi_interval(PG_FUNCTION_ARGS)
  src/backend/utils/adt/timestamp.c:3365:timestamptz_mi_interval_internal(TimestampTz timestamp,
  src/backend/utils/adt/timestamp.c:3389:timestamptz_mi_interval(PG_FUNCTION_ARGS)
  src/backend/utils/adt/timestamp.c:3412:timestamptz_mi_interval_at_zone(PG_FUNCTION_ARGS)

Since timestamptz_pl_interval() and timestamptz_mi_interval() are independent
functions (not just aliases or direct calls to the timestamp versions), their
return macros should be updated for accuracy: change any PG_RETURN_TIMESTAMP()
to the correct PG_RETURN_TIMESTAMPTZ().

The timestamptz_pl_interval_at_zone() and timestamptz_mi_interval_at_zone() may
still intentionally blur the distinction between timestamp and timestamptz
-- consistent with the historical design choices discussed earlier -- so they
can remain unchanged for now.

--
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.