Обсуждение: float conversion / presentation problem ?
Hi folks, i've noticed an problem with conversion from real to float - this produces ugly values. (in fact it seems to be real->float8) > fxignal=# SELECT 12.345::real::float; > float8 > ------------------ > 12.3450002670288 Maybe its not really an conversion problem, but an bug in sprintf(), (there were some other float related bugs known in glibc), as these examples let me suspect: > fxignal=# SELECT 12.345::real::float::real; > float4 > -------- > 12.345 > fxignal=# SELECT ((12.345::real::float8)-(0.345::float8))::real; > float4 > -------- > 12 I noticed this behaviour on postgresql-7.4.1 and 7.4.3 on glibc-2.3.2 BTW: real is an alias to float4 ? regards, -- --------------------------------------------------------------------- Enrico Weigelt == metux IT service phone: +49 36207 519931 www: http://www.metux.de/ fax: +49 36207 519932 email: contact@metux.de cellphone: +49 174 7066481 --------------------------------------------------------------------- -- DSL ab 0 Euro. -- statische IP -- UUCP -- Hosting -- Webshops -- ---------------------------------------------------------------------
Enrico Weigelt <weigelt@metux.de> writes:
> i've noticed an problem with conversion from real to float -
> this produces ugly values. (in fact it seems to be real->float8)
You do realize that float4/real is only good to six decimal places
on most platforms?
You can take the issue up with the glibc boys if you like, but
I think you'll get laughed off.
regards, tom lane
Tom Lane wrote: > Enrico Weigelt <weigelt@metux.de> writes: > >>i've noticed an problem with conversion from real to float - >>this produces ugly values. (in fact it seems to be real->float8) > > > You do realize that float4/real is only good to six decimal places > on most platforms? > > You can take the issue up with the glibc boys if you like, but > I think you'll get laughed off. and importantly, are stored as binary fractions. 0.10000 decimal is a repeating fraction in binary, its like .1100110011.... x 2^-2 or something as a FLOAT. Float4 has like 22 bits of significants for the mantissa, while Float8 has like 51 bits or something (this is from rusty memory, so I could have slipped a few bits here).
John R Pierce <pierce@hogranch.com> writes:
> Tom Lane wrote:
>> You do realize that float4/real is only good to six decimal places
>> on most platforms?
> and importantly, are stored as binary fractions. 0.10000 decimal is a
> repeating fraction in binary, its like .1100110011.... x 2^-2 or
> something as a FLOAT.
Right, and conversely a fraction that's cut off at a certain number of
base-2 digits often corresponds to a repeating pattern of base-10
digits. I once had a clue about the exact laws for this, but it was
many years ago :-(
regards, tom lane
Enrico Weigelt wrote: > BTW: real is an alias to float4 ? Yes. -Neil
On Thu, Apr 07, 2005 at 01:02:50AM -0400, Tom Lane wrote: > John R Pierce <pierce@hogranch.com> writes: > > Tom Lane wrote: > >> You do realize that float4/real is only good to six decimal > >> places on most platforms? > > > and importantly, are stored as binary fractions. 0.10000 decimal > > is a repeating fraction in binary, its like .1100110011.... x 2^-2 > > or something as a FLOAT. > > Right, and conversely a fraction that's cut off at a certain number > of base-2 digits often corresponds to a repeating pattern of base-10 > digits. I once had a clue about the exact laws for this, but it was > many years ago :-( As with base-10, any denominator that can't be expressed as a power of the prime factors in the base results in a repeating number. So basically any denominator that can't be expressed as 2^n turns into a repeating binary doohicky. Cheers, D -- David Fetter david@fetter.org http://fetter.org/ phone: +1 510 893 6100 mobile: +1 415 235 3778 Remember to vote!
* Tom Lane <tgl@sss.pgh.pa.us> wrote: > Enrico Weigelt <weigelt@metux.de> writes: > > i've noticed an problem with conversion from real to float - > > this produces ugly values. (in fact it seems to be real->float8) > > You do realize that float4/real is only good to six decimal places > on most platforms? Thats okay for me. I need only four. > You can take the issue up with the glibc boys if you like, but > I think you'll get laughed off. ehm, why ? cu -- --------------------------------------------------------------------- Enrico Weigelt == metux IT service phone: +49 36207 519931 www: http://www.metux.de/ fax: +49 36207 519932 email: contact@metux.de cellphone: +49 174 7066481 --------------------------------------------------------------------- -- DSL ab 0 Euro. -- statische IP -- UUCP -- Hosting -- Webshops -- ---------------------------------------------------------------------