Re: int/float binary conversion

[Lars Christian Jensen]

On Wed, 14 Sep 2005, Ulrich Eckhardt wrote:

> Lars Christian Jensen wrote:
> >     float f1 = 4.5;
> >     int i =    *((int*)&f1);
> >     float f2 = *((float*)&i);
> >
> >     printf("f1 = %f, i = 0x%08x, f2 = %f\n", f1, i, f2);
>
> You are using stupid, brute force casts to convert between totally
> unrelated objects and expect anything useful?

Assume that sizeof(int) == sizeof(float), which is the case on my 
platform. The binary representation of the float is put into an integer 
(the integer value would of course be platform dependent), and then back 
into another float. I would assume that those two floats should have the 
same binary representation, and hence have the same value, but they don't.

> Sorry, but what exactly is your problem?

I found an equivalent c-style cast that caused problems inside a library 
that used to work. A float/int union solves the problem (and is easier to 
read), but I can't see anything wrong in these casts.

> If you wanted to demonstrate that it is possible to write C++ programs 
> that don't behave consistently (i.e. show different implementations of 
> 'undefined behaviour'), fine, but we all knew that already.

The problem code uses pointer-to-pointer c-style casts, address-of 
operator(&) and indirection operator(*). Which of these gives un-defined 
behaviour in c++?

--
Lars Christian Jensen