Earliest Timestamp supported in PostgreSQL - sql

I work with different databases in a number of different time zones (and periods of time) and one thing that normally originates problems, is the date/time definition.
For this reason, and since a date is a reference to a starting value, to keep track of how it was calculated, I try to store the base date; i.e.: the minimum date supported in that particular computer/database;
If I am seeing it well, this depends on the RDBMS and on the particular storage of the type.
In SQL Server, I found a couple of ways of calculating this "base date";
SELECT CONVERT(DATETIME, 0)
or
SELECT DATEADD(MONTH, 0, 0 )
or even a cast like this:
DECLARE #300 BINARY(8)
SET #300 = 0x00000000 + CAST(300 AS BINARY(4))
set #dt=(SELECT CAST(#300 AS DATETIME) AS BASEDATE)
print CAST(#dt AS NVARCHAR(100))
(where #dt is a datetime variable)
My question is, is there a similar way of calculating the base date in PostgreSQL, i.e.: the value that is the minimum date supported and is on the base of all calculations?
From the description of the date type, I can see that the minimum date supported is 4713 BC, but is there a way of getting this value programmatically (for instance as a formatted date string), as I do in SQL Server?

The manual states the values as:
Low value: 4713 BC
High value: 294276 AD
with the caveat, as Chris noted, that -infinity is also supported.
See the note later in the same page in the manual; the above is only true if you are using integer timestamps, which are the default in all vaguely recent versions of PostgreSQL. If in doubt:
SHOW integer_datetimes;
will tell you. If you're using floating point datetimes instead, you get greater range and less (non-linear) precision. Any attempt to work out the minimum programatically must cope with that restriction.
PostgreSQL does not just let you cast zero to a timestamp to get the minimum possible timestamp, nor would this make much sense if you were using floating point datetimes. You can use the julian date conversion function, but this gives you the epoch not the minimum time:
postgres=> select to_timestamp(0);
to_timestamp
------------------------
1970-01-01 08:00:00+08
(1 row)
because it accepts negative values. You'd think that giving it negative maxint would work, but the results are surprising to the point where I wonder if we've got a wrap-around bug lurking here:
postgres=> select to_timestamp(-922337203685477);
to_timestamp
---------------------------------
294247-01-10 12:00:54.775808+08
(1 row)
postgres=> select to_timestamp(-92233720368547);
to_timestamp
---------------------------------
294247-01-10 12:00:54.775808+08
(1 row)
postgres=> select to_timestamp(-9223372036854);
to_timestamp
------------------------------
294247-01-10 12:00:55.552+08
(1 row)
postgres=> select to_timestamp(-922337203685);
ERROR: timestamp out of range
postgres=> select to_timestamp(-92233720368);
to_timestamp
---------------------------------
0954-03-26 09:50:36+07:43:24 BC
(1 row)
postgres=> select to_timestamp(-9223372036);
to_timestamp
------------------------------
1677-09-21 07:56:08+07:43:24
(1 row)
(Perhaps related to the fact that to_timestamp takes a double, even though timestamps are stored as integers these days?).
I think it's possibly wisest to just let the timestamp range be any timestamp you don't get an error on. After all, the range of valid timestamps is not continuous:
postgres=> SELECT TIMESTAMP '2000-02-29';
timestamp
---------------------
2000-02-29 00:00:00
(1 row)
postgres=> SELECT TIMESTAMP '2001-02-29';
ERROR: date/time field value out of range: "2001-02-29"
LINE 1: SELECT TIMESTAMP '2001-02-29';
so you can't assume that just because a value is between two valid timestamps, it is its self valid.

The earliest timestamp is '-infinity'. This is a special value. The other side is 'infinity' which is later than any specific timestamp.
I don't know of a way of getting this programaticly. I would just use the value hard-coded the way you might use NULL. That means you have to handle infinities on the client side though.

Related

How does ORACLE database convert DATE to NUMBER implicitly?

I'm trying to understand this function:
NVL2( NULL, ( SYSDATE - SYSDATE ), DATE '2020-05-24' ))
And its returned value:
NVL2(NULL,(SYSDATE-SYSDATE),DATE '2020-05-24'))
2458994
I'm having trouble understanding where that number, 2458994, comes from, as SYSDATE-SYSDATE is a NUMBER, and you cannot implicity convert a DATE to a NUMBER:
TO_NUMBER(DATE '2020-05-24')
ORA-01722: invalid number
The ORACLE SQL Language Reference NVL2 states:
If expr2 is numeric data, then Oracle Database determines which argument has the highest numeric precedence, implicitly converts the other argument to that data type, and returns that data type.
So my question is, what form of conversion is ORACLE SQL using on the DATE datatype to make it a NUMBER datatype?
It's effectively doing:
to_number(to_char(DATE '2020-05-24','J'))
The 'J' is (from the docs):
Julian day; the number of days since January 1, 4712 BC. Number specified with J must be integers.
If you run that manually it gets the same value you see:
select to_number(to_char(DATE '2020-05-24','J')) from dual;
2458994
It isn't obvious that it should be doing that, but it is. If the second argument is a plain (type-2) number then you get an error:
select NVL2(NULL,42,to_date('2020-05-24','YYYY-MM-DD')) from dual;
ORA-00932: inconsistent datatypes: expected NUMBER got DATE
If you dump the date subtraction result it comes back as a different (internal, undocumented as far as I can see) data type:
select dump(SYSDATE-SYSDATE) from dual;
Typ=14 Len=8: 114,133,37,0,0,0,0,0
That seems to cause the third argument to be converted to that same type; it's almost equivalent to:
select DATE '2020-05-24' - DATE '-4712-01-01' from dual;
2458993
So it looks like it's either doing something similar to that but adjusting it, or doing an internal version of the 'J' conversion, or... something else vaguely similar. It doesn't seem to be documented behaviour.

SELECT matching dates from a timestamp with time zone column

I have a timestamp with time zone column within which I'd like to run a query returning all matching dates. eg. I want all rows which have a timestamp with date 2019-09-30. I'm trying something like this but haven't been able to figure it out:
SELECT * FROM table WHERE
x='1277' AND
date='2019-09-30 21:40:01.316240 +00:00'::DATE;
There are two options:
range search:
WHERE timestampcol >= TIMESTAMPTZ '2019-09-30'
AND timestampcol < (TIMESTAMPTZ '2019-09-30' + INTERVAL '1 day')
The proper index to make this fast is
CREATE INDEX ON atable (timestampcol);
conversion to date:
WHERE CAST(timestampcol AS date) = '2019-09-30'
The proper index to make this fast is
CREATE INDEX ON atable ((CAST(timestampcol AS date)));
Both methods work equally well. The second method has a shorter WHERE clause, but a specialized index that maybe no other query can benefit from.
You can use such a collation among your date converted column value and fixed date value :
with tab( x, date ) as
(
select 1277, '2019-09-30 21:40:01.316240 +00:00'::timestamp with time zone union all
select 1278, '2019-09-29 21:40:01.316240 +00:00'::timestamp with time zone
)
select *
from tab
where date::date = date'2019-09-30';
Demo
I am not familiar with postgresql, but the following page https://www.postgresql.org/docs/9.1/datatype-datetime.html indicates that there is an optional precision value p that is used to specify the number of fractional digits retained in the seconds field. How was the data type configured for your date field? Is the timestamp column being stored as an eight byte integer or a floating-point number? If the later, the effective limit of precision might be less than 6. All of this info is on the above linked page.
You have several options, as they have been mentioned, and may vary depending on the data type of your field named "date".
For example,
Turn your field to date, returning format 'yyyy-mm-dd':
SELECT * FROM table WHERE x='1277' AND date::DATE='2019-09-30';
"date" just after field name.
Convert it to char and retrieve 10 characters:
SELECT * FROM table WHERE x='1277' AND LEFT(date::varchar,10)='2019-09-30';
Like the previous:
SELECT * FROM table WHERE x='1277' AND to_char(date,'yyyymmdd')='20190930';
And there are many others. For more specific info, you have to check PostgreSQL documentation to check which one is best for you or post more information in order we can guess more about your problem.
https://www.postgresql.org/docs/9.4/functions-datetime.html

Issue querying date from oracle.

I understand that querying a date will fail as its comparing a string to date and that can cause an issue.
Oracle 11.2 G
Unicode DB
NLS_DATE_FORMAT DD-MON-RR
select * from table where Q_date='16-Mar-09';
It can be solved by
select * from table where trunc(Q_date) = TO_DATE('16-MAR-09', 'DD-MON-YY');
What I don't get is why this works.
select* from table where Q_date='07-JAN-08';
If anyone can please elaborate or correct my mindset.
Thanks
Oracle does allow date literals, but they depend on the installation (particularly the value of NLS_DATE_FORMAT as explained here). Hence, there is not a universal format for interpreting a single string as a date (unless you use the DATE keyword).
The default format is DD-MM-YY, which seems to be the format for your server. So, your statement:
where Q_date = '07-JAN-08'
is interpreted using this format.
I prefer to use the DATE keyword with the ISO standard YYYY-MM-DD format:
where Q_Date = DATE '2008-01-07'
If this gets no rows returned:
select * from table where Q_date='16-Mar-09';
but this does see data:
select * from table where trunc(Q_date) = TO_DATE('16-MAR-09', 'DD-MON-YY');
then you have rows which have a time other than midnight. At this point in the century DD-MON-RR and DD-MON-YY are equivalent, and both will see 09 as 2009, so the date part is right. But the first will only find rows where the time is midnight, while the second is stripping the time off via the trunc, meaning the dates on both sides are at midnight, and therefore equal.
And since this also finds data:
select* from table where Q_date='07-JAN-08';
... then you have rows at midnight on that date. You might also have rows with other times, so checking the count with the trunc version might be useful.
You can check the times you actually have with:
select to_char(q_date, 'YYYY-MM-DD HH24:MI:SS') from table;
If you do want to make sure you catch all times within the day you can use a range:
select * from table where
q_date >= date '2009-03-16'
and q_date < date '2009-03-17';
Quick SQL Fiddle demo.
Although it sounds like you're expecting all the times to be midnight, which might indicate a data problem.

Equals(=) vs. LIKE for date data type

First, I am aware that this question has been posted generally Equals(=) vs. LIKE.
Here, I query about date type data on ORACLE database, I found the following, when I write select statment in this way:
SELECT ACCOUNT.ACCOUNT_ID, ACCOUNT.LAST_TRANSACTION_DATE
FROM ACCOUNT
WHERE ACCOUNT.LAST_TRANSACTION_DATE LIKE '30-JUL-07';
I get all rows I'm looking for. but when I use the sign equal = instead :
SELECT ACCOUNT.ACCOUNT_ID, ACCOUNT.LAST_TRANSACTION_DATE
FROM ACCOUNT
WHERE ACCOUNT.LAST_TRANSACTION_DATE = '30-JUL-07';
I get nothing even though nothing is different except the equal sign. Can I find any explanation for this please ?
Assuming LAST_TRANSACTION_DATE is a DATE column (or TIMESTAMP) then both version are very bad practice.
In both cases the DATE column will implicitly be converted to a character literal based on the current NLS settings. That means with different clients you will get different results.
When using date literals always use to_date() with(!) a format mask or use an ANSI date literal. That way you compare dates with dates not strings with strings. So for the equal comparison you should use:
LAST_TRANSACTION_DATE = to_date('30-JUL-07', 'dd-mon-yy')
Note that using 'MON' can still lead to errors with different NLS settings ('DEC' vs. 'DEZ' or 'MAR' vs. 'MRZ'). It is much less error prone using month numbers (and four digit years):
LAST_TRANSACTION_DATE = to_date('30-07-2007', 'dd-mm-yyyy')
or using an ANSI date literal
LAST_TRANSACTION_DATE = DATE '2007-07-30'
Now the reason why the above query is very likely to return nothing is that in Oracle DATE columns include the time as well. The above date literals implicitly contain the time 00:00. If the time in the table is different (e.g. 19:54) then of course the dates are not equal.
To workaround this problem you have different options:
use trunc() on the table column to "normalize" the time to 00:00
trunc(LAST_TRANSACTION_DATE) = DATE '2007-07-30
this will however prevent the usage of an index defined on LAST_TRANSACTION_DATE
use between
LAST_TRANSACTION_DATE between to_date('2007-07-30 00:00:00', 'yyyy-mm-dd hh24:mi:ss') and to_date('2007-07-30 23:59:59', 'yyyy-mm-dd hh24:mi:ss')
The performance problem of the first solution could be worked around by creating an index on trunc(LAST_TRANSACTION_DATE) which could be used by that expression. But the expression LAST_TRANSACTION_DATE = '30-JUL-07' prevents an index usage as well because internally it's processed as to_char(LAST_TRANSACTION_DATE) = '30-JUL-07'
The important things to remember:
Never, ever rely on implicit data type conversion. It will give you problems at some point. Always compare the correct data types
Oracle DATE columns always contain a time which is part of the comparison rules.
You should not compare a date to a string directly. You rely on implicit conversions, the rules of which are difficult to remember.
Furthermore, your choice of date format is not optimal: years have four digits (Y2K bug?), and not all languages have the seventh month of the year named JUL. You should use something like YYYY/MM/DD.
Finally, dates in Oracle are points in time precise to the second. All dates have a time component, even if it is 00:00:00. When you use the = operator, Oracle will compare the date and time for dates.
Here's a test case reproducing the behaviour you described:
SQL> create table test_date (d date);
Table created
SQL> alter session set nls_date_format = 'DD-MON-RR';
Session altered
SQL> insert into test_date values
2 (to_date ('2007/07/30 11:50:00', 'yyyy/mm/dd hh24:mi:ss'));
1 row inserted
SQL> select * from test_date where d = '30-JUL-07';
D
-----------
SQL> select * from test_date where d like '30-JUL-07';
D
-----------
30/07/2007
When you use the = operator, Oracle will convert the constant string 30-JUL-07 to a date and compare the value with the column, like this:
SQL> select * from test_date where d = to_date('30-JUL-07', 'DD-MON-RR');
D
-----------
When you use the LIKE operator, Oracle will convert the column to a string and compare it to the right-hand side, which is equivalent to:
SQL> select * from test_date where to_char(d, 'DD-MON-RR') like '30-JUL-07';
D
-----------
30/07/2007
Always compare dates to dates and strings to strings. Related question:
How to correctly handle dates in queries constraints
The date field is not a string. Internally an implicit conversion is made to a string when you use =, which does not match anything because your string does not have the required amount of precision.
I'd have a guess that the LIKE statement behaves somewhat differently with a date field, causing implicit wildcards to be used in the comparison that eliminates the requirement for any precision. Essentially, your LIKE works like this:
SELECT ACCOUNT.ACCOUNT_ID, ACCOUNT.LAST_TRANSACTION_DATE
FROM ACCOUNT
WHERE ACCOUNT.LAST_TRANSACTION_DATE BETWEEN DATE('30-JUL-07 00:00:00.00000+00:00') AND DATE('30-JUL-07 23:59:59.99999+00:00');

In Oracle, convert number(5,10) to date

When ececute the following SQL syntax in Oracle, always not success, please help.
40284.3878935185 represents '2010-04-16 09:18:34', with microsecond.
an epoch date of 01 January 1900 (like Excel).
create table temp1 (date1 number2(5,10));
insert into temp1(date1) values('40284.3878935185');
select to_date(date1, 'yyyy-mm-dd hh24:mi:ssxff') from temp1
Error report: SQL Error: ORA-01861: literal does not match format
string
01861. 00000 - "literal does not match format string"
*Cause: Literals in the input must be the same length as literals in
the format string (with the exception of leading whitespace). If the
"FX" modifier has been toggled on, the literal must match exactly,
with no extra whitespace.
*Action: Correct the format string to match the literal.
Thanks to Mark Bannister
Now the SQL syntax is:
select to_char(to_date('1899-12-30','yyyy-mm-dd') +
date1,'yyyy-mm-dd hh24:mi:ss') from temp1
but can't fetch the date format like 'yyyy-mm-dd hh24:mi:ss.ff'. Continue look for help.
Using an epoch date of 30 December 1899, try:
select to_date('1899-12-30','yyyy-mm-dd') + date1
Simple date addition doesn't work with timestamps, at least if you need to preserve the fractional seconds. When you do to_timestamp('1899-12-30','yyyy-mm-dd')+ date1 (in a comment on Mark's answer) the TIMESTAMP is implicitly converted to a DATE before the addition, to the overall answer is a DATE, and so doesn't have any fractional seconds; then you use to_char(..., '... .FF') it complains with ORA-01821.
You need to convert the number of days held by your date1 column into an interval. Fortunately Oracle provides a function to do exactly that, NUMTODSINTERVAL:
select to_timestamp('1899-12-30','YYYY-MM-DD')
+ numtodsinterval(date1, 'DAY') from temp3;
16-APR-10 09.18.33.999998400
You can then display that in your desired format, e.g. (using a CTE to provide your date1 value):
with temp3 as ( select 40284.3878935185 as date1 from dual)
select to_char(to_timestamp('1899-12-30','YYYY-MM-DD')
+ numtodsinterval(date1, 'DAY'), 'YYYY-MM-DD HH24:MI:SSXFF') from temp3;
2010-04-16 09:18:33.999998400
Or to restrict to thousandths of a second:
with temp3 as ( select 40284.3878935185 as date1 from dual)
select to_char(to_timestamp('1899-12-30','YYYY-MM-DD')+
+ numtodsinterval(date1, 'DAY'), 'YYYY-MM-DD HH24:MI:SS.FF3') from temp3;
2010-04-16 09:18:33.999
An epoch of 1899-12-30 sounds odd though, and doesn't correspond to Excel as you stated. It seems more likely that your expected result is wrong and it should be 2010-04-18, so I'd check your assumptions. Andrew also makes some good points, and you should be storing your value in the table in a TIMESTAMP column. If you receive data like this though, you still need something along these lines to convert it for storage at some point.
Don't know the epoch date exactly, but try something like:
select to_date('19700101','YYYYMMDD')+ :secs_since_epoch/86400 from dual;
Or, cast to timestamp like:
select cast(to_date('19700101', 'YYYYMMDD') + :secs_since_epoch/86400 as timestamp with local time zone) from dual;
I hope this doesn't come across too harshly, but you've got to totally rethink your approach here.
You're not keeping data types straight at all. Each line of your example misuses a data type.
TEMP1.DATE1 is not a date or a varchar2, but a NUMBER
you insert not the number 40284.3878935185, but the STRING >> '40284.3878935185' <<
your SELECT TO_DATE(...) uses the NUMBER Temp1.Date1 value, but treats it as a VARCHAR2 using the format block
I'm about 95% certain that you think Oracle transfers this data using simple block data copies. "Since each Oracle date is stored as a number anyway, why not just insert that number into the table?" Well, because when you're defining a column as a NUMBER you're telling Oracle "this is not a date." Oracle therefore does not manage it as a date.
Each of these type conversions is calculated by Oracle based on your current session variables. If you were in France, where the '.' is a thousands separator rather than a radix, the INSERT would completely fail.
All of these conversions with strings are modified by the locale in which Oracle thinks your running. Check dictionary view V$NLS_PARAMETERS.
This gets worse with date/time values. Date/time values can go all over the map - mostly because of time zone. What time zone is your database server in? What time zone does it think you're running from? And if that doesn't spin your head quite enough, check out what happens if you change Oracle's default calendar from Gregorian to Thai Buddha.
I strongly suggest you get rid of the numbers ENTIRELY.
To create date or date time values, use strings with completely invariant and unambiguous formats. Then assign, compare and calculate date values exclusively, e.g.:
GOODFMT constant VARCHAR2 = 'YYYY-MM-DD HH24:MI:SS.FFF ZZZ'
Good_Time DATE = TO_DATE ('2012-02-17 08:07:55.000 EST', GOODFMT);