[Qemu-devel] Re: [PATCH 2/5] Support human unit formats in strtobytes, eg. 1.0G

[Paolo Bonzini]

On 09/16/2010 12:42 PM, Jes Sorensen wrote:
> On 09/16/10 12:40, Avi Kivity wrote:
> >   On 09/15/2010 09:31 PM, Jes Sorensen wrote:
> > > Floating point is just plain wrong.
> >
> > Why?  If command-line processing becomes too slow, you can always buy a
> > math co-processor.
>
> Because it's imprecise anyway

As Avi mentioned, this is only true if you need byte precision beyond 4 
PB.  But most of the time byte precision is not necessary so in practice 
floating-point will be indistinguishable: all exact powers of 10 up to 
10^22 (beyond 64-bits) can be represented correctly by an IEEE double.

There's also strtold, if you're worried about precision...

> Besides, most users will probably hit their shell command line limit
> before hitting the problem with the decimals.

Value is first shifted and then multiplied, so that 6-7 digits may 
already overflow if the unit is terabytes.

Paolo