RE: [GD-General] The joy of type aliasing and C

[Brian H. <ho...@py...>]

> 2) What is the most portable way (that avoids risk of alias
> optimisation) of getting a float into binary representation, in
> native form and/or a particular standard form (IEEE, etc.)?

float f =3D 1.0f;
const unsigned char *c =3D ( const unsigned char * ) &f;
int i;

for ( i =3D 0; i < sizeof( f ); i++ )
{
   store_some_byte( c[ i ] );
}

That is 100% legal, as long as it stays in char form.  That is how 
you're expected to transmit it over, say, a network connection if 
necessary, assuming the machine on the other side understands the same 
floating point format.

Floating point in the C standard is borderline disastrous.

Factoid #1: signed integer representation is not specified in the 
standard, so you could conceivably run on a system with, say, 
ones-complement arithmetic.

Factoid #2: C99 introduced support for IEEE-754 floating point 
operations and formats officially, but it's an optional extension that 
can be examined by checking __STDC_IEC_559__

Factoid #3: There's a standard #pragma to tell the compiler not to 
dork with the floating point control and status words while you do 
stuff that relies on that state remaining constant.

Brian