Strict aliasing and converting binary patterns between integer and float

[Olivier Galibert <galibert at pobox dot com>]

I need to convert back and forth between an unsigned int and the float
with the same binary pattern (admitting I knows they're the same size
- 32bits - in the first place).  Up until now I've been using the usual:

unsigned int f(void);

float g(void)
{
  unsigned int v1;
  float v2;
  v1 = f();
  v2 = *(float *)&v1;
  return v2;
}

WHile this very simple example may still work with -fstrict-aliasing,
as soon as you have a more complex function the value of v2 is not
always what is expected.

So my question is, what is a reliable way to do this conversion with
gcc and strict aliasing enabled?

  OG.