timestamp arithmetic (a possible bug?)

Interesting....

If I reverse the order it works...

agencysacks=# SELECT CAST('01.01.2004 10:00:00' AS
TIMESTAMP) + (CAST('01.01.2004 10:01:00' AS TIMESTAMP)
- CAST('01.01.2004 10:00:00' AS TIMESTAMP)) as answer;
answer
---------------------
2004-01-01 10:01:00
(1 row)

However your original...

agencysacks=# SELECT (CAST('01.01.2004 10:01:00' AS
TIMESTAMP) - CAST('01.01.2004 10:00:00' AS TIMESTAMP))
+ CAST('01.01.2004 10:00:00' AS TIMESTAMP);
ERROR: operator does not exist: interval + timestamp
without time zone
HINT: No operator matches the given name and argument
type(s). You may need to add explicit type casts.

agencysacks=# select version();

version

-------------------------------------------------------------------------------------------------------------------------
PostgreSQL 7.4.2 on powerpc-apple-darwin7.2.0,
compiled by GCC gcc (GCC) 3.3 20030304 (Apple
Computer, Inc. build 1495)
(1 row)

Looks like postgresql demands the order to be
timestamp +- interval.

Ted

--- Ilir Gashi <I.Gashi@city.ac.uk> wrote:

Hi,

I saw this behaviour in PostgreSQL 7.2. (Once again,
I know this is an old
release but I do not have a newer version installed,
and I am only using
the server for research purposes). If you execute
the following statement

SELECT (CAST('01.01.2004 10:01:00' AS TIMESTAMP) -
CAST('01.01.2004
10:00:00' AS TIMESTAMP)) + CAST('01.01.2004
10:00:00' AS TIMESTAMP);

The result returned is:

?column?
---------------------
2004-01-01 00:01:00
(1 row)

I was expecting: 2004-01-01 10:01:00.

Tried it on Oracle 8.0.5:

SELECT TO_DATE('01.01.2004 10:01:00', 'DD.MM.YYYY
HH:MI:SS') -
TO_DATE('01.01.2004 10:00:00', 'DD.MM.YYYY
HH:MI:SS') + TO_DATE('01.01.2004
10:00:00', 'DD.MM.YYYY HH:MI:SS') FROM DUAL;

---------------------------
2004-01-01 10:01:00
(1 row selected)

And MSSQL 7:

SELECT (CAST('01.01.2004 10:01:00' AS DATETIME) -
CAST('01.01.2004
10:00:00' AS DATETIME) + CAST('01.01.2004 10:00:00'
AS DATETIME));

---------------------------

2004-01-01 10:01:00.000

(1 row(s) affected)

Is this a bug? Same thing happens if I use
TimestampTZ rather than
Timestamp.

Best regards,

Ilir

____________________________________________

Ilir Gashi
PhD Student
Centre for Software Reliability
City University
Northampton Square, London EC1V 0HB
email: i.gashi@city.ac.uk
website:
http://www.csr.city.ac.uk/csr_city/staff/gashi/
____________________________________________

---------------------------(end of
broadcast)---------------------------
TIP 1: subscribe and unsubscribe commands go to
majordomo@postgresql.org

__________________________________
Do you Yahoo!?
Yahoo! Mail - 50x more storage than other providers!
http://promotions.yahoo.com/new_mail

On Jul 2 2004, Achilleus Mantzios wrote:

SELECT '01.01.2004 10:00:00'::timestamp + ('01.01.2004
10:01:00'::timestamp - '01.01.2004 10:00:00'::timestamp)::interval;
?column?
---------------------
2004-01-01 10:01:00
(1 row)

works fine in my 7.4.2

Yes. I've just tried it in PostgreSQL 7.2 and it works fine there as well.

On Jul 2 2004, Theodore Petrosky wrote:

Interesting....

If I reverse the order it works...

agencysacks=# SELECT CAST('01.01.2004 10:00:00' AS
TIMESTAMP) + (CAST('01.01.2004 10:01:00' AS TIMESTAMP)
- CAST('01.01.2004 10:00:00' AS TIMESTAMP)) as answer;
answer
---------------------
2004-01-01 10:01:00
(1 row)

Yes. The above works for me as well in 7.2.

However your original...

agencysacks=# SELECT (CAST('01.01.2004 10:01:00' AS
TIMESTAMP) - CAST('01.01.2004 10:00:00' AS TIMESTAMP))
+ CAST('01.01.2004 10:00:00' AS TIMESTAMP);
ERROR: operator does not exist: interval + timestamp
without time zone
HINT: No operator matches the given name and argument
type(s). You may need to add explicit type casts.

agencysacks=# select version();

version

-------------------------------------------------------------------------------------------------------------------------
PostgreSQL 7.4.2 on powerpc-apple-darwin7.2.0,
compiled by GCC gcc (GCC) 3.3 20030304 (Apple
Computer, Inc. build 1495)
(1 row)

Looks like postgresql demands the order to be
timestamp +- interval.

Ted

I am not getting the error above. I get the incorrect result as reported in
my original report. Does this mean that this was a buggy behaviour in 7.2
and then was changed in the latter releases to give the error message that
7.4.2 is giving?

Thanks for the quick responses.

Best regards,

Ilir
____________________________________________

Ilir Gashi
PhD Student
Centre for Software Reliability
City University
Northampton Square, London EC1V 0HB
email: i.gashi@city.ac.uk
website: http://www.csr.city.ac.uk/csr_city/staff/gashi/
____________________________________________

Ilir Gashi <I.Gashi@city.ac.uk> writes:

I saw this behaviour in PostgreSQL 7.2. (Once again, I know this is an old
release but I do not have a newer version installed, and I am only using
the server for research purposes). If you execute the following statement

SELECT (CAST('01.01.2004 10:01:00' AS TIMESTAMP) - CAST('01.01.2004
10:00:00' AS TIMESTAMP)) + CAST('01.01.2004 10:00:00' AS TIMESTAMP);

More recent releases fail with

regression=# SELECT (CAST('01.01.2004 10:01:00' AS TIMESTAMP) - CAST('01.01.2004 10:00:00' AS TIMESTAMP)) + CAST('01.01.2004 10:00:00' AS TIMESTAMP);
ERROR: operator does not exist: interval + timestamp without time zone
HINT: No operator matches the given name and argument type(s). You may need to add explicit type casts.

(There is a timestamp + interval operator, so you could make it work by
flipping around the outer addition.)

It looks like what is happening in 7.2 is the parser is choosing the
closest available operator, which happens to be time + date, and
then coercing the interval and timestamp to time and date respectively.
The former isn't so bad but the latter loses your 10AM data...

Since then (I think in 7.3) we tightened the rules so that
information-losing coercions, such as timestamp to date, won't be
invoked implicitly.

regards, tom lane

On Fri, 2 Jul 2004, Tom Lane wrote:

Ilir Gashi <I.Gashi@city.ac.uk> writes:

I saw this behaviour in PostgreSQL 7.2. (Once again, I know this is an old
release but I do not have a newer version installed, and I am only using
the server for research purposes). If you execute the following statement

SELECT (CAST('01.01.2004 10:01:00' AS TIMESTAMP) - CAST('01.01.2004
10:00:00' AS TIMESTAMP)) + CAST('01.01.2004 10:00:00' AS TIMESTAMP);

(There is a timestamp + interval operator, so you could make it work by
flipping around the outer addition.)

Should we be providing an interval + timestamp operator as well since it
looks like the spec implies both orderings should work?

Stephan Szabo <sszabo@megazone.bigpanda.com> writes:

On Fri, 2 Jul 2004, Tom Lane wrote:

(There is a timestamp + interval operator, so you could make it work by
flipping around the outer addition.)

Should we be providing an interval + timestamp operator as well since it
looks like the spec implies both orderings should work?

If you see spec support for it, then yes ... where do you read that
exactly?

regards, tom lane

On Fri, 2 Jul 2004, Tom Lane wrote:

Stephan Szabo <sszabo@megazone.bigpanda.com> writes:

On Fri, 2 Jul 2004, Tom Lane wrote:

(There is a timestamp + interval operator, so you could make it work by
flipping around the outer addition.)

Should we be providing an interval + timestamp operator as well since it
looks like the spec implies both orderings should work?

If you see spec support for it, then yes ... where do you read that
exactly?

SQL92 (draft) 4.5.3 Operators involving datetimes and intervals (the table
appears to be the same in SQL99 4.7.3)

The list given there is:
datetime - datetime -> interval
datetime +,- interval -> datetime
interval + datetime -> datetime
interval +,- interval -> interval
interval *,/ numeric -> interval
numeric * interval -> interval

Stephan Szabo <sszabo@megazone.bigpanda.com> writes:

Should we be providing an interval + timestamp operator as well since it
looks like the spec implies both orderings should work?

If you see spec support for it, then yes ... where do you read that
exactly?

SQL92 (draft) 4.5.3 Operators involving datetimes and intervals (the table
appears to be the same in SQL99 4.7.3)

Yeah. It looks like we have most of these, but would need to add
interval + date
interval + timetz
interval + timestamp
interval + timestamptz
and for consistency
integer + date
Curiously, we do have interval + time without time zone ... I guess Tom
Lockhart overlooked these when he was working in the area.

I notice also that date - date yields an integer (ie, number of days)
where I think that strict spec compliance would mandate yielding an
interval instead. I'm uneager to change this though.

regards, tom lane