[Bug c++/61668] Division returns an error in some cases

[Bug c++/61668] Division returns an error in some cases

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Actually maybe 0.59 can be represented exactly, but that doesn't change the
fact that floating point numbers are not exact and calculations can accumulate
errors.

If you really think this is a compiler bug then please provide the information
requested at https://gcc.gnu.org/bugs not screenshots.

[Bug c++/61668] Division returns an error in some cases

[ketelbinkje at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

Peter Ketel <ketelbinkje at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical

--- Comment #4 from Peter Ketel <ketelbinkje at gmail dot com> ---
(In reply to Jonathan Wakely from comment #2)
> Floating point numbers are not exact. 0.59 cannot be exactly represented in
> a 64-bit double.

Which will render the the compiler useless because I cann't predict when a
number can't be written in 64 bit double format. All computations need to be
error free. Calculator will return the correct result no matter what!

If this isn't accepted as a bug then all computations done by programs based on
this compiler can no longer be trusted.

[Bug c++/61668] Division returns an error in some cases

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal

--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Please read http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html
before making silly claims.


I don't know any compiler that will print 0.59 for this program:

extern "C" int printf(const char*, ...);

int main()
{
  printf("%.32f\n", 0.59);
}

[Bug c++/61668] Division returns an error in some cases

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

--- Comment #6 from Jonathan Wakely <redi at gcc dot gnu.org> ---
See also https://gcc.gnu.org/bugs/#nonbugs

[Bug c++/61668] Division returns an error in some cases

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

--- Comment #7 from Andreas Schwab <schwab@linux-m68k.org> ---
Don't try to emulate decimal arithmetics with binary arithmetics.

[Bug c++/61668] Division returns an error in some cases

[ketelbinkje at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

--- Comment #8 from Peter Ketel <ketelbinkje at gmail dot com> ---
So what is the use of a math library that is only capable of binary
computations?
In the real world we use the decimal system for all computations.

Do we need binary computations? In most cases NO.

What is the use of a real or double if it's only capable to approximate
a decimal fraction?

This piece of code will calculate the value for s correctly on a IBM
mainframe. Please note that this code was written 1966!

C AREA OF A TRIANGLE - HERON'S FORMULA

C INPUT - CARD READER UNIT 5, INTEGER INPUT

C OUTPUT - LINE PRINTER UNIT 6, REAL OUTPUT

C INPUT ERROR DISPAY ERROR OUTPUT CODE 1 IN JOB
CONTROL LISTING

      INTEGER A,B,C

      READ(5,501) A,B,C

  501 FORMAT(3I5)

      IF(A.EQ.0 .OR. B.EQ.0 .OR. C.EQ.0) STOP 1

      S = (A + B + C) / 2.0

      AREA = SQRT( S * (S - A) * (S - B) * (S -
C))

      WRITE(6,601) A,B,C,AREA

  601 FORMAT(4H A= ,I5,5H  B= ,I5,5H  C= ,I5,8H 
AREA= ,F10.2,12HSQUARE UNITS)

      STOP

      END


This is the modern Fortan test program:

program main

  implicit none
  integer re_i
  real :: Test =0.59

  Test = (0.45 + 0.45 + 0.28) / 2

  re_i = system("pause")

end

It wil fail to compute the correct value because it too uses binary
computation.
So why do we have real and double data types for computational purposes
that will never return a correct result?

Thank God we didn't use the crappy compilers anno 2014 when we flew to
the moon.


On 02-07-14 12:03, schwab@linux-m68k.org wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668
>
> --- Comment #7 from Andreas Schwab <schwab@linux-m68k.org> ---
> Don't try to emulate decimal arithmetics with binary arithmetics.
>

[Bug c++/61668] Division returns an error in some cases

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61668

--- Comment #9 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Please go and complain somewhere else, it's not GCC's problem that you are
surprised by the behaviour of binary floating point logic.

There are several ways to work with exact decimal numbers. See
https://gcc.gnu.org/onlinedocs/gcc/Decimal-Float.html for one option. Another
is a maths library such a GMP for working with rational numbers. Another is
working with integers and scaling them in your code.