Paranoia on x86 question

[Ralf Corsepius]

On Tue, 2004-08-17 at 21:20, Angelo Fraietta wrote:
> Peter Dufault wrote:
> 
> > I built the development tools per the instructions and note that the 
> > paranoia test still fails, exactly as in the archives.  This is on a 
> > 400MHz Celeron using pc586.
> >
> > Is this expected, and should I re-build with less optimization?  I 
> > want to be sure I've built the right tool chain, I followed the 
> > directions in the "Getting Started" document.
> 
> The paranoia test has failed for as long as I can remember on pc386. I 
> have tested it successfully using the pc386dx bsp on a 386sx processor. 
> I don't know if paranoi was ever fixed for the remainer of the 386 
> derivatives

Earlier this year, there had been an extensive discussion of paranoia on
the gcc-list (gcc-list at gnu.org).

In a nutshell, the result was: Different GCC versions on ix86's w/ FPU
have different precision/float accuracy problems. Depending on the
compiler options and compiler version you are using, they often break
paranoia, including the version in RTEMS.

=> Though paranoia failures are not "expected", they also are not
"unexpected".

Ralf