Thread: casts: max double precision > text > double precision fails with out or range error

Tried asking this in pgsql-general but I got no response, so I thought
I'd give hackers a shot:

postgres=# select (((1.7976931348623157081e+308)::double
precision)::text)::double precision;
ERROR:  "1.79769313486232e+308" is out of range for type double precision

I'm working on a pg driver and in my float data decoder functional
tests, I ran into some errors that I eventually traced back to this
behavior. Essentially, postgres seems to cast the max normal double
(i.e., the bits of ~(1ULL<<52 | 1ULL<<63)) to text in such a manner
that it's rounded up, and the reverse cast, text-to-double-precision,
does not recognize it as being in range. Curiously, pg_dump seems to
print doubles with more precision (in both COPY and INSERT modes),
avoiding this issue. Of course I'm not expecting perfect precision in
round-tripping doubles like this (this is always dicey with IEEE
floating point anyway), but failing outright is a little ugly. Any
thoughts? Version is PostgreSQL 8.4.6 on i486-pc-linux-gnu, compiled
by GCC gcc-4.4.real (Ubuntu 4.4.3-4ubuntu5) 4.4.3, 32-bit.

Also, although the simplest way to illustrate this problem is with
this round-trip set of casts, that's obviously a contrived use case.
However, given that the same behavior is seen in the TEXT mode output
for doubles of the FEBE protocol, I think it's a little more
noteworthy.

Thanks,
Maciek Sakrejda

Excerpts from Maciej Sakrejda's message of mar ene 11 03:28:13 -0300 2011:
> Tried asking this in pgsql-general but I got no response, so I thought
> I'd give hackers a shot:
> 
> postgres=# select (((1.7976931348623157081e+308)::double
> precision)::text)::double precision;
> ERROR:  "1.79769313486232e+308" is out of range for type double precision
> 
> I'm working on a pg driver and in my float data decoder functional
> tests, I ran into some errors that I eventually traced back to this
> behavior. Essentially, postgres seems to cast the max normal double
> (i.e., the bits of ~(1ULL<<52 | 1ULL<<63)) to text in such a manner
> that it's rounded up, and the reverse cast, text-to-double-precision,
> does not recognize it as being in range. Curiously, pg_dump seems to
> print doubles with more precision (in both COPY and INSERT modes),
> avoiding this issue.

Yeah, it sets the extra_float_digits parameter.

alvherre=# set extra_float_digits to 3;
SET
alvherre=# select (((1.7976931348623157081e+308)::double precision)::text)::double precision;         float8          
──────────────────────────1.79769313486231571e+308
(1 fila)

-- 
Álvaro Herrera <alvherre@commandprompt.com>
The PostgreSQL Company - Command Prompt, Inc.
PostgreSQL Replication, Consulting, Custom Development, 24x7 support