Thread: Strange behavior with timestamptz
Hi Everyone,
I have a situation where two tables have a "Created" field defined as follows:
table seedlot "created timestamptz DEFAULT now(), "
table transaction "created timestamptz NOT NULL DEFAULT now(),"
The transaction table records when a seedlot record is created, with both events happening in the same plpgsql function. The seedlot record is created first, with the transaction table being updated later in the procedure.
Seedlot 153 was received on August 11 with the following result:
base=# select created from seedlot where syslotid=153;
created
------------------------
2003-11-08 12:13:39-06
(1 row)
created
------------------------
2003-11-08 12:13:39-06
(1 row)
base=# select created from transaction where syslotid=153 and transactiontypeid=22;
created
------------------------
2003-08-11 12:13:39-05
(1 row)
created
------------------------
2003-08-11 12:13:39-05
(1 row)
Does anyone have any idea why the default for seedlot recorded the time with the day and month switched, resulting in the seedlot record being stamped Nov 8, 2003 while the transaction was stamped correctly as Aug 11, 2003?
In the same function a number of other records are created and timestamped. In all cases where the table definition statement does not include NOT NULL, the time was stamped with the day and month reversed.
However! Two of the tables define created thus:
"created timestamptz DEFAULT now() NOT NULL,"
but still ended up with the day and month being switched.
The transaction record is the last one in the procedure to be created.
I am using version 7.3.2.
I hope I'm not missing something obvious...
Thanks for your help,
George
Hi Tom, I have written both the application and the PostgreSQL procedures (still in development). All the timestamps I referred to are created by default when the stored procedure is invoked. The problem may be due to some inconsistency in how PostgreSQL is interpreting what the operating system (Windows XP) is supplying. While I have not been able to recreate the problem directly, another aspect of the process does show an inconsistency. In the application the user enters the date the product was received (which may be different from the date the record is created). This is passed to the stored procedure as a parameter of type date, and is inserted into a date field (datereceived). By changing the system date and running the application the following occurred: Date received entered as August 11, 2003: ? me.datetimepicker1.value #8/11/2003 8:18:28 PM# base=# select datereceived, created from receiving where receivingid=56;datereceived | created --------------+------------------------------2003-11-08 | 2003-08-25 20:20:55.41425-05 (1 row) Date received entered as August 25, 2003: ? me.datetimepicker1.value #8/25/2003 8:22:37 PM# base=# select datereceived, created from receiving where receivingid=57;datereceived | created --------------+------------------------------2003-08-25 | 2003-08-25 20:22:39.68625-05 (1 row) base=# show datestyle; DateStyle ---------------------------------------ISO with US (NonEuropean) conventions (1 row) In both cases the month is being sent to the stored procedure first, but in the first instance (month < 13) it is being interpreted as the day. George ----- Original Message ----- From: "Tom Lane" <tgl@sss.pgh.pa.us> To: "George Weaver" <georgew1@mts.net> Cc: <pgsql-sql@postgresql.org> Sent: Monday, August 25, 2003 3:59 PM Subject: Re: [SQL] Strange behavior with timestamptz > "George Weaver" <georgew1@mts.net> writes: > > Does anyone have any idea why the default for seedlot recorded the time wit= > > h the day and month switched, resulting in the seedlot record being stamped= > > Nov 8, 2003 while the transaction was stamped correctly as Aug 11, 2003? > > It's really not possible to believe that both of those were loaded from > the defaults you show. now() doesn't ever break down the system clock > value into day/month/year --- it just takes the system clock time in > seconds-since-epoch and adds a constant to get the right zero offset. > So there's no credible mechanism for now() to make such a mistake. > > I think that your client software supplied a value for one field and > didn't supply a value for the other, and the supplied value was provided > in the wrong DateStyle. > > regards, tom lane >
"George Weaver" <georgew1@mts.net> writes:
> Does anyone have any idea why the default for seedlot recorded the time wit=
> h the day and month switched, resulting in the seedlot record being stamped=
> Nov 8, 2003 while the transaction was stamped correctly as Aug 11, 2003?
It's really not possible to believe that both of those were loaded from
the defaults you show. now() doesn't ever break down the system clock
value into day/month/year --- it just takes the system clock time in
seconds-since-epoch and adds a constant to get the right zero offset.
So there's no credible mechanism for now() to make such a mistake.
I think that your client software supplied a value for one field and
didn't supply a value for the other, and the supplied value was provided
in the wrong DateStyle.
regards, tom lane
> ... When the datereceived parameter was defined
> as type Date, it was actually arriving at the procedure as "11-08-2003" and
> "25-08-2003" in spite of showing up as indicated below. When I redefined
> the datereceived parameter as type OdbcDate, it arrived correctly at the
> procedure as "2003-08-11" and "2003-08-25".
Yeah. The first two formats are ambiguous, in Postgres' mind anyway.
With the US datestyle setting, it will first attempt to parse as mm-dd-yyyy
and if that fails try dd-mm-yyyy. So you'd need to change the datestyle
to Euro to get dd-mm-yyyy input to be parsed reliably.
As of 7.4 this is being tightened up, btw --- it'll be mm-dd-yyyy or error.
But AFAICS this has nothing to do with a default now() insertion,
because that value is never converted to a string before it gets into
the stored column.
regards, tom lane
Hi Tom, Not believing that PostgreSQL would be less consistent than Microsoft :-), I spent some time looking at how the application actually was transferring the date to the database procedure. When the datereceived parameter was defined as type Date, it was actually arriving at the procedure as "11-08-2003" and "25-08-2003" in spite of showing up as indicated below. When I redefined the datereceived parameter as type OdbcDate, it arrived correctly at the procedure as "2003-08-11" and "2003-08-25". Checking further, this appears to result from the computer's time settings (English - Canada) which I know realize use the European format. So it looks like this problem arose due to my inexperience. (Nonetheless I am still perplexed by the fact that the default values were assigned inconsistently as explained earlier and will try to puzzle this one through as well). Sorry for the confusion. George ----- Original Message ----- From: "George Weaver" <georgew1@mts.net> To: "Tom Lane" <tgl@sss.pgh.pa.us> Cc: <pgsql-sql@postgresql.org> Sent: Monday, August 25, 2003 9:51 PM Subject: Re: [SQL] Strange behavior with timestamptz > Hi Tom, > > I have written both the application and the PostgreSQL procedures (still in > development). All the timestamps I referred to are created by default when > the stored procedure is invoked. The problem may be due to some > inconsistency in how PostgreSQL is interpreting what the operating system > (Windows XP) is supplying. While I have not been able to recreate the > problem directly, another aspect of the process does show an inconsistency. > > In the application the user enters the date the product was received (which > may be different from the date the record is created). This is passed to > the stored procedure as a parameter of type date, and is inserted into a > date field (datereceived). By changing the system date and running the > application the following occurred: > > Date received entered as August 11, 2003: > > ? me.datetimepicker1.value > #8/11/2003 8:18:28 PM# > > base=# select datereceived, created from receiving where receivingid=56; > datereceived | created > --------------+------------------------------ > 2003-11-08 | 2003-08-25 20:20:55.41425-05 > (1 row) > > Date received entered as August 25, 2003: > > ? me.datetimepicker1.value > #8/25/2003 8:22:37 PM# > > base=# select datereceived, created from receiving where receivingid=57; > datereceived | created > --------------+------------------------------ > 2003-08-25 | 2003-08-25 20:22:39.68625-05 > (1 row) > > base=# show datestyle; > DateStyle > --------------------------------------- > ISO with US (NonEuropean) conventions > (1 row) > > In both cases the month is being sent to the stored procedure first, but in > the first instance (month < 13) it is being interpreted as the day. > > George > > ----- Original Message ----- > From: "Tom Lane" <tgl@sss.pgh.pa.us> > To: "George Weaver" <georgew1@mts.net> > Cc: <pgsql-sql@postgresql.org> > Sent: Monday, August 25, 2003 3:59 PM > Subject: Re: [SQL] Strange behavior with timestamptz > > > > "George Weaver" <georgew1@mts.net> writes: > > > Does anyone have any idea why the default for seedlot recorded the time > wit= > > > h the day and month switched, resulting in the seedlot record being > stamped= > > > Nov 8, 2003 while the transaction was stamped correctly as Aug 11, > 2003? > > > > It's really not possible to believe that both of those were loaded from > > the defaults you show. now() doesn't ever break down the system clock > > value into day/month/year --- it just takes the system clock time in > > seconds-since-epoch and adds a constant to get the right zero offset. > > So there's no credible mechanism for now() to make such a mistake. > > > > I think that your client software supplied a value for one field and > > didn't supply a value for the other, and the supplied value was provided > > in the wrong DateStyle. > > > > regards, tom lane > > > > > ---------------------------(end of broadcast)--------------------------- > TIP 1: subscribe and unsubscribe commands go to majordomo@postgresql.org