Обсуждение: round(real,integer)

History question:
Why does select round(3,3) work, 
         select round(3.0,3) work, 
but      select round(3.0::real,1) not work?

There's a utility cast in the integer case (described here
http://www.postgresql.org/docs/9.3/static/typeconv-func.html),but not in the real case. 
Is this on purpose, or just an oversight? 
Obviously one can work around it, but I'm sick of doing so, it makes my SQL examples ugly. I’d like to fix it, but not
ifthere’s a reason it’s not supposed to be done. 
P.

--
Paul Ramsey
http://cleverelephant.ca
http://postgis.net

Paul Ramsey <pramsey@cleverelephant.ca> writes:
> History question:
> Why does select round(3,3) work, 
>          select round(3.0,3) work, 
> but      select round(3.0::real,1) not work?

The 2-argument round() function actually takes (numeric, integer).

There's an implicit cast from int to numeric, but not an implicit
cast from real to numeric (for that pair of types, the direction
of implicit casting is from numeric to real).

The choices we made for implicit casting behavior among the numeric
datatypes are based on what it says in the SQL standard about exact and
approximate numeric types.  There's probably room for argument about
the details, but it seems unlikely that we'd risk the breakage that'd
ensue from rejiggering the casting rules at this stage.

If this annoys you enough, a more realistic solution would be to add
an additional round(float8, int) function.  I think it'd have to return
numeric though if you don't want surprises.

            regards, tom lane

On January 10, 2014 at 12:37:04 PM, Tom Lane (tgl@sss.pgh.pa.us) wrote:

Paul Ramsey <pramsey@cleverelephant.ca> writes:
> History question:
> Why does select round(3,3) work, 
>          select round(3.0,3) work, 
> but      select round(3.0::real,1) not work?

The 2-argument round() function actually takes (numeric, integer).

There's an implicit cast from int to numeric, but not an implicit
cast from real to numeric (for that pair of types, the direction
of implicit casting is from numeric to real).

The choices we made for implicit casting behavior among the numeric
datatypes are based on what it says in the SQL standard about exact and
approximate numeric types. There's probably room for argument about
the details, but it seems unlikely that we'd risk the breakage that'd
ensue from rejiggering the casting rules at this stage.

If this annoys you enough, a more realistic solution would be to add
an additional round(float8, int) function. I think it'd have to return
numeric though if you don't want surprises.

regards, tom lane

Paul Ramsey <pramsey@cleverelephant.ca> writes:
> What surprises do you think would come from a
> round(real, integer) returns real
> function? 

People might expect that rounding to, say, 6 digits produces an exact
decimal answer.  Even if you're not exceeding 6 digits overall, it's
unlikely that the answer is *exact*, if it's represented as a float.

            regards, tom lane

On January 10, 2014 at 1:09:24 PM, Tom Lane (tgl@sss.pgh.pa.us) wrote:

to, say, 6 digits produces an exact 
decimal answer. Even if you're not exceeding 6 digits overall, it's 
unlikely that the answ