[GENERAL] Typecasting datetype as date. How do I cope with NULLs?

"SR" == Stuart Rison <stuart@ludwig.ucl.ac.uk> writes:

SR> Dear All,
SR> For those of you who don't want to wade through the details, here's the
SR> question: "How do I get the date portion of a datetime field for ALL
SR> ENTRIES in a table regardless of whether the entry is NULL or not? (N.B.
SR> Typecasting a datetime NULL as date generates an error)"

. . .

SR> I find myself having to do two selects to get all the people in the table.
SR> i.e.:

SR> patients=> SELECT surname,firstname,othernames,dob::date FROM patients
SR> WHERE dob
SR> IS NOT NULL;
SR> surname|firstname|othernames | date
SR> -------+---------+--------------------+----------
SR> Goose |Mother |Lay Golden Eggs |11-01-1923
SR> One |Un |Uno Ein |11-11-1111
SR> Light |Dee |Full |22-01-1933
SR> (3 rows)

SR> patients=> SELECT surname,firstname,othernames FROM patients WHERE dob IS NULL;
SR> surname|firstname|othernames
SR> -------+---------+----------
SR> Rison |Stuart |
SR> Rison |This |Pal
SR> Rison |Mark |
SR> (3 rows)

SR> My question is, how do I get surname,firstname,othername and the date
SR> portion of ALL people in table people regardless of whether the entry has
SR> an actual dob or a NULL dob.

There are, on my mind, at least two answers. For experience I use
small table 'create table a( dt datetime, i int)'. Hera are data in
this table (one row has NULL as dt value):

tolik=> select * from a;
dt | i
----------------------------+--
Thu Nov 26 16:35:23 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10

First use 'union':
-----------------------------------------------
select dt::date, i from a where dt is not null
union
select NULL, i from a where dt is null;
date| i
----------+--
11-25-1998| 2
11-26-1998| 1
11-27-1998| 3
|10
(4 rows)
-----------------------------------------------

Second, try use date_trunc('day', dt) instead date_part:
--------------------------------------------------------------
tolik=> select date_trunc('day', dt), i from a;
date_trunc | i
----------------------------+--
Thu Nov 26 00:00:00 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10
(4 rows)

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

Regards!

--
Anatoly K. Lasareff Email: tolik@icomm.ru
Senior programmer

There are, on my mind, at least two answers. For experience I use
small table 'create table a( dt datetime, i int)'. Hera are data in
this table (one row has NULL as dt value):

tolik=> select * from a;
dt | i
----------------------------+--
Thu Nov 26 16:35:23 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10

First use 'union':
-----------------------------------------------
select dt::date, i from a where dt is not null
union
select NULL, i from a where dt is null;
date| i
----------+--
11-25-1998| 2
11-26-1998| 1
11-27-1998| 3
|10
(4 rows)
-----------------------------------------------

Second, try use date_trunc('day', dt) instead date_part:
--------------------------------------------------------------
tolik=> select date_trunc('day', dt), i from a;
date_trunc | i
----------------------------+--
Thu Nov 26 00:00:00 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10
(4 rows)

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

I find the above interesting and I've tried something similar that won't
work.
I have two tables defined as follows

create table test1 (id1 int,link1 int);
create table test2 (id2 int, field2 varchar(5));

where link1 is a foreign key of test1 that should be linked to id2 of test2

Now when I execute the following query:

select id1,field2,link1 from test1,test2 where test1.link1=test2.id2
union
select id1,NULL,link1 from test1;

I always get the following error:

Each UNION query must have identical target types.

Why this error, and what does it mean?

TIA
Wim Ceulemans - wim.ceulemans@nice.be
Nice Software Solutions
Eglegemweg 3, 2811 Hombeek - Belgium
Tel +32(0)15 41 29 53 - Fax +32(0)15 41 29 54

"WC" == Wim Ceulemans <wim.ceulemans@nice.be> writes:

WC> I find the above interesting and I've tried something similar that won't
WC> work.
WC> I have two tables defined as follows

WC> create table test1 (id1 int,link1 int);
WC> create table test2 (id2 int, field2 varchar(5));

WC> where link1 is a foreign key of test1 that should be linked to id2 of test2

WC> Now when I execute the following query:

WC> select id1,field2,link1 from test1,test2 where test1.link1=test2.id2
WC> union
WC> select id1,NULL,link1 from test1;

WC> I always get the following error:

WC> Each UNION query must have identical target types.

WC> Why this error, and what does it mean?

Pehaps this is a bug fixed in 6.4 version? I saw this message when
before upgrating to 6.4. You can modify query so:

select id1,field2,link1 from test1,test2 where test1.link1=test2.id2
union
select id1, -1, link1 from test1;

where -1 is value non present in field2 domain.

--
Anatoly K. Lasareff Email: tolik@icomm.ru
Senior programmer

SR> Dear All,
SR> For those of you who don't want to wade through the details, here's the
SR> question: "How do I get the date portion of a datetime field for ALL
SR> ENTRIES in a table regardless of whether the entry is NULL or not? (N.B.
SR> Typecasting a datetime NULL as date generates an error)"

<snip>

There are, on my mind, at least two answers. For experience I use
small table 'create table a( dt datetime, i int)'. Hera are data in
this table (one row has NULL as dt value):

tolik=> select * from a;
dt | i
----------------------------+--
Thu Nov 26 16:35:23 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10

First use 'union':
-----------------------------------------------
select dt::date, i from a where dt is not null
union
select NULL, i from a where dt is null;
date| i
----------+--
11-25-1998| 2
11-26-1998| 1
11-27-1998| 3
|10
(4 rows)

I had not thought of using a UNION, thanks for that (well that's not
strictly true, I was 'procedurally' implemeting it rather then using SQL!!)
so this is a definite possibility.
I still feel that this equates to two SQL queries since the backend will
have to process each individually and then UNION them (is this right?). My
suggestion of:

patients=> SELECT surname,firstname,othernames,date_part('day',dob) as dd,
patients-> date_part('month',dob) as mm, date_part('year',dob) as yyyy
patients-> FROM patients;
surname|firstname|othernames |dd|mm|yyyy
-------+---------+--------------------+--+--+----
Goose |Mother |Lay Golden Eggs |11| 1|1923
One |Un |Uno Ein |11|11|1111
Light |Dee |Full |22| 1|1933
Rison |Stuart | | | |
Rison |This |Pal | | |
Rison |Mark | | | |
(6 rows)

means only one query need to be executed by the backend and it can cope
with NULL, but it 'generates' three date fields and I would like only one.

Second, try use date_trunc('day', dt) instead date_part:
--------------------------------------------------------------
tolik=> select date_trunc('day', dt), i from a;
date_trunc | i
----------------------------+--
Thu Nov 26 00:00:00 1998 MSK| 1
Wed Nov 25 00:00:00 1998 MSK| 2
Fri Nov 27 00:00:00 1998 MSK| 3
|10
(4 rows)

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

Again, this solves one part of my problem (i.e. the query functions even if
I datetime is NULL) but what I'm looking for is something that ONLY shows
the date portion of a datetime (e.g. Thu Nov 26 16:35:23 1998 MSK becomes
26-11-1998) so the date_trunc masking is not quite what I'm looking for.

Would it be possible to do a select datetime and then output only the
dateday, datemonth and dateyear using an RE??

All the same, thanks for your suggestion Anatoly, UNION's the winner so far ;)

regards,

S.

+-------------------------+--------------------------------------+
| Stuart Rison            | Ludwig Institute for Cancer Research |
+-------------------------+ 91 Riding House Street               |
| Tel. (0171) 878 4041    | London, W1P 8BT, UNITED KINGDOM.     |
| Fax. (0171) 878 4040    | stuart@ludwig.ucl.ac.uk              |
+-------------------------+--------------------------------------+

At 13:12 +0200 on 27/11/98, Stuart Rison wrote:

Again, this solves one part of my problem (i.e. the query functions even if
I datetime is NULL) but what I'm looking for is something that ONLY shows
the date portion of a datetime (e.g. Thu Nov 26 16:35:23 1998 MSK becomes
26-11-1998) so the date_trunc masking is not quite what I'm looking for.

Would it be possible to do a select datetime and then output only the
dateday, datemonth and dateyear using an RE??

All the same, thanks for your suggestion Anatoly, UNION's the winner so
far ;)

Well, first I'd like to say that not converting null datetimes is
definitely a bug and should be placed in the ToDo list for the next version.

That said, I tried several conversions, using the following table:

testing=> select * from test1;
f1 |dt
------------+----------------------------
This year |Fri Jan 02 00:00:00 1998 IST
Next year |Mon Mar 15 00:00:00 1999 IST
Nothing |
Last year |Mon May 12 00:00:00 1997 IDT
More nothing|
(5 rows)

The one that finally worked is the following:

testing=> select f1, date( abstime( dt ) ) from test1;
f1 | date
------------+----------
This year |01-02-1998
Next year |03-15-1999
Nothing |
Last year |05-12-1997
More nothing|
(5 rows)

That is, convert to abstime, and then to date...

Herouth

--
Herouth Maoz, Internet developer.
Open University of Israel - Telem project
http://telem.openu.ac.il/~herutma