Обсуждение: select t.name from tbl t (where "name" is not a column name)

hi,

i've just noticed the following behaviour and was wondering
if there's any documentation to explain what it's for.

  create table tbl(id serial primary key, a text, b text, c text);
  insert into tbl(a, b, c) values ('abc', 'def', 'ghi');
  insert into tbl(a, b, c) values ('jkl', 'mno', 'pqr');
  insert into tbl(a, b, c) values ('stu', 'vwx', 'yza');
  select t.name from tbl t;
  drop table tbl;

results in:

        name
  -----------------
   (1,abc,def,ghi)
   (2,jkl,mno,pqr)
   (3,stu,vwx,yza)

you can select t.name from tbl t even though "name" isn't the name
of a column in the table and you get this tuple-like version of the
each row. it doesn't work without the "t" alias and of course it
doesn't work if name really is the name of a column in the table.

so, what's this behaviour for, and, is there an equivalent
way to do it when "name" is the name of a column? i doubt
that i need this behaviour for anything. i'm just curious
and didn't see any mention of this in the documentation for
the select statement.

cheers,
raf

On 02/23/2010 05:07 PM, raf wrote:
> i've just noticed the following behaviour and was wondering
> if there's any documentation to explain what it's for.
>
>   create table tbl(id serial primary key, a text, b text, c text);
>   insert into tbl(a, b, c) values ('abc', 'def', 'ghi');
>   insert into tbl(a, b, c) values ('jkl', 'mno', 'pqr');
>   insert into tbl(a, b, c) values ('stu', 'vwx', 'yza');
>   select t.name from tbl t;

I forget exactly where this is documented (and could not find it with a
quick look), but calling t.name is the same as name(t) if a column
reference is not found, and name is a function, which it is.

So t.name is essentially casting the whole row as a name datatype and
outputting the result. Try it with text:

test=# \d foo
      Table "public.foo"
 Column |  Type   | Modifiers
--------+---------+-----------
 f      | integer |

test=# select foo.text from foo;
 text
------
 (-1)
(1 row)

test=# drop TABLE foo;
DROP TABLE

test=# create table foo(f int, text text);
CREATE TABLE

test=# insert into foo values(-1,'abc');
INSERT 0 1

test=# select foo.text from foo;
 text
------
 abc
(1 row)

test=# select foo.name from foo;
   name
----------
 (-1,abc)
(1 row)

HTH,

Joe

Joe,

What PG version are running?

8.2 here complains when running your example:

ERROR:  column foo.name does not exist
LINE 6: select foo.name from foo;
               ^

********** Error **********

ERROR: column foo.name does not exist
SQL state: 42703


Igor Neyman


> -----Original Message-----
> From: Joe Conway [mailto:mail@joeconway.com]
> Sent: Tuesday, February 23, 2010 9:19 PM
> To: pgsql-general@postgresql.org
> Subject: Re: select t.name from tbl t (where "name" is not a
> column name)
>
> On 02/23/2010 05:07 PM, raf wrote:
> > i've just noticed the following behaviour and was wondering
> if there's
> > any documentation to explain what it's for.
> >
> >   create table tbl(id serial primary key, a text, b text, c text);
> >   insert into tbl(a, b, c) values ('abc', 'def', 'ghi');
> >   insert into tbl(a, b, c) values ('jkl', 'mno', 'pqr');
> >   insert into tbl(a, b, c) values ('stu', 'vwx', 'yza');
> >   select t.name from tbl t;
>
> I forget exactly where this is documented (and could not find
> it with a quick look), but calling t.name is the same as
> name(t) if a column reference is not found, and name is a
> function, which it is.
>
> So t.name is essentially casting the whole row as a name
> datatype and outputting the result. Try it with text:
>
> test=# \d foo
>       Table "public.foo"
>  Column |  Type   | Modifiers
> --------+---------+-----------
>  f      | integer |
>
> test=# select foo.text from foo;
>  text
> ------
>  (-1)
> (1 row)
>
> test=# drop TABLE foo;
> DROP TABLE
>
> test=# create table foo(f int, text text); CREATE TABLE
>
> test=# insert into foo values(-1,'abc'); INSERT 0 1
>
> test=# select foo.text from foo;
>  text
> ------
>  abc
> (1 row)
>
> test=# select foo.name from foo;
>    name
> ----------
>  (-1,abc)
> (1 row)
>
> HTH,
>
> Joe
>
>

On 02/24/2010 07:16 AM, Igor Neyman wrote:
> Joe,
>
> What PG version are running?
>
> 8.2 here complains when running your example:
>
> ERROR:  column foo.name does not exist
> LINE 6: select foo.name from foo;
>                ^
>
> ********** Error **********
>
> ERROR: column foo.name does not exist
> SQL state: 42703

Prior to 8.3 you aren't able to cast a rowtype as text or name datatype,
so no matching function is found.

-------------
in 8.3.x
-------------
contrib_regression=# select text(foo) from foo;
 text
------
 (-1)
(1 row)

-------------
in 8.2.x
-------------
contrib_regression=# select text(foo) from foo;
ERROR:  function text(foo) does not exist
LINE 1: select text(foo) from foo;
               ^
HINT:  No function matches the given name and argument types. You may
need to add explicit type casts.


Joe