using non-standard optflags (-O3, in particular)

[Patrice Dumas]

On Fri, Feb 09, 2007 at 05:17:11PM +0100, Ralf Corsepius wrote:
> >  Sure, people have thought about the defaults, but it doesn't 
> > mean that the packager doesn't know what he's doing either.
> Rest assured: In 99% of all cases they don't know.
> 
> They test on their "Pentium IV" and claim something - They can't have
> any clues about what happens on a sparc, an i586 and 
> AMD X2 <what the heck>, or a ppc something.

That's a good point. It should indeed be verified/benchmarked on the
most common arches supported. It is theoretically possible that
contributors share their benchmarks on the different platforms to
validate the results, but it is clearly not commonly happening in
fedora...

> > Using -O3 in specific cases isn't that big of a deal.
> It is - It renders debugging impossible on many systems, strict-aliasing
> silently kills SW on some targets, it might trigger exotic
> target-specific bugs etc. etc. 

I admit that I am not familiar with this issue, but what you describe
don't seem to be very reproducible. Besides it seems that for some of
the issues you describe the issue is in the compiler, so it shouldn't
prevent from using it, but instead it should be fixed in the compiler.

--
Pat