Re: gnu software bugs - long double

[Mischa Baars <mjbaars1977 at gmail dot com>]

On 11/02/2013 11:19 PM, Jonathan Wakely wrote:
> On 2 November 2013 22:12, Mischa Baars wrote:
> > On 11/02/2013 11:06 PM, Jonathan Wakely wrote:
> > > On 2 November 2013 21:52, Mischa Baars wrote:
> > > > You are mistaken :)
> > > >
> > > > Indeed some rational numbers can only be represented up to a certain
> > > > number
> > > > of bits, like 1 / 3. Others can be exactly represented, like 1 / 8.
> > > >
> > > > All real numbers, and therefore all rational numbers, can be represented
> > > > up
> > > > to a certain number of bits.
> > > i.e. not exactly.
> > >
> > > > Converting 1.1 from string to double should not be a problem.
> > > That's not what your program does. It converts (long double)1.1, which
> > > is not equal to 1.1, to a string.
> >
> > I'm trying to pass a completely normal value to the debugger and to the
> > standard output, and do not convert any value to a string.
> Seriously?  How do you think printf writes to standard output if not
> converting to a string?

So that what glibc does :) That's not my doing.

> And 1.1 is not representable as long double.
>
> If you are willing to stop being so arrogant for a few minutes and
> learn something try running this program and think about what the
> result means:
>
> #include <assert.h>
>
> int main()
> {
>    long double l = 1.1;
>    long double ll = 10 * l;
>    assert( ll == 11 );
> }
>
> If you think GCC gets this result wrong then please use a different
> compiler and stop wasting everyone's time, this is off-topic for this
> mailing list.

There's no converting to any string in your example. You only convert 
source code strings into their corresponding doubles.

What I'm trying to point out is that the output differs from the value 
entered in the source. The string 1.1 from the source is not correctly 
converted to it's corresponding double, or the double is not correctly 
converted back into it's corresponding string.