Обсуждение: bug in float8in()

I run the following sql statements in linux and get the results:
postgres=# create table test_double(col1 float8);
CREATE TABLE
postgres=# insert into test_double values(1.7976931348623159E308);
INSERT 0 1
postgres=# select * from test_double;  col1
----------Infinity
(1 row)

but in windows:
postgres=# create table test_double(col1 float8);
CREATE TABLE
postgres=# insert into test_double values(1.7976931348623159E308);
ERROR: 
"17976931348623159000000000000000000000000000000000000000000000000000000
00000000000000000000000000000000000000000000000000000000000000000000000000000000
00000000000000000000000000000000000000000000000000000000000000000000000000000000
000000000000000000000000000000000000000000000000000000000000000000000000000000"
is out of range for type double precision
postgres=# select * from test_double;col1
------
(0 rows)

This is a bug, the behavior in windows and linux is not the smae

I check the code and find that float8in exists a bug:
the strtod() function in linux dosen't set errno when dealing with 
1.7976931348623159E308 but in windows it does
How should we improve it?

If you look into documentation, the behavior of float/double is platform depend.  If you want to same result on any
platform,use numeric instead.
 
    Zdenek

Richard Wang napsal(a):
> I run the following sql statements in linux and get the results:
> postgres=# create table test_double(col1 float8);
> CREATE TABLE
> postgres=# insert into test_double values(1.7976931348623159E308);
> INSERT 0 1
> postgres=# select * from test_double;
>    col1
> ----------
>  Infinity
> (1 row)
> 
> but in windows:
> postgres=# create table test_double(col1 float8);
> CREATE TABLE
> postgres=# insert into test_double values(1.7976931348623159E308);
> ERROR: 
> "17976931348623159000000000000000000000000000000000000000000000000000000
> 00000000000000000000000000000000000000000000000000000000000000000000000000000000
> 00000000000000000000000000000000000000000000000000000000000000000000000000000000
> 000000000000000000000000000000000000000000000000000000000000000000000000000000"
> is out of range for type double precision
> postgres=# select * from test_double;
>  col1
> ------
> (0 rows)
> 
> This is a bug, the behavior in windows and linux is not the smae
> 
> I check the code and find that float8in exists a bug:
> the strtod() function in linux dosen't set errno when dealing with 
> 1.7976931348623159E308 but in windows it does
> How should we improve it?
> 
> 
> 

Richard Wang wrote:
> I run the following sql statements in linux and get the results:
> postgres=# create table test_double(col1 float8);
> CREATE TABLE
> postgres=# insert into test_double values(1.7976931348623159E308);
> INSERT 0 1
> postgres=# select * from test_double;
>    col1
> ----------
>  Infinity
> (1 row)
> 
> but in windows:
> postgres=# create table test_double(col1 float8);
> CREATE TABLE
> postgres=# insert into test_double values(1.7976931348623159E308);
> ERROR: 
> "17976931348623159000000000000000000000000000000000000000000000000000000
> 00000000000000000000000000000000000000000000000000000000000000000000000000000000
> 00000000000000000000000000000000000000000000000000000000000000000000000000000000
> 000000000000000000000000000000000000000000000000000000000000000000000000000000"
> is out of range for type double precision
> postgres=# select * from test_double;
>  col1
> ------
> (0 rows)
> 
> This is a bug, the behavior in windows and linux is not the smae

float relies on platform behavior, AFAIK. You're better off using
numeric if you need to process these values without that.

Float is an approximate datatype. In this case, linux makes a fairly
wild approximation (that it's the same as infinity) whereas windows
says it just can't approximate it. You could argue for either one being
correct, but in the end you just need to use a different datatype if
you need consistent behavior.

> the strtod() function in linux dosen't set errno when dealing with 
> 1.7976931348623159E308 but in windows it does
> How should we improve it?

Not sure we should, really...

//Magnus

Magnus Hagander <magnus@hagander.net> writes:
> Float is an approximate datatype. In this case, linux makes a fairly
> wild approximation (that it's the same as infinity) whereas windows
> says it just can't approximate it.

Note that the behavior isn't consistent across Unixen either ---
my HPUX box acts like Windows here.  Basically Postgres just tries
to reflect the platform's floating-point support.
        regards, tom lane