How to know if compiler is Safe?

[David Rees]

On Thu, 11 Jun 1998, Jeffrey A Law wrote:

> It depends on exactly what errors you got.  We can't help you if
> you don't give us the list of unexpected failures :-)
> 
> Some unexpected failures are normal on certain platforms.

I saw that there is some work being done on a webpage to help gather
results.  I've posted mine there (but the directions aren't very clear, so
I hope I did it right!).  I've posted my results using the supplied
test_summary program earlier today.  It doesn't look like it's showed up
yet (I'm not actually subscribed to the list, but it's not on the web page
yet) so I'll send it out again.  It would be helpful if there was a list
of expected non-fatal errors for each platform, or at least the expected
number of unexpected errors for various platforms.  For example, if I knew
what tests H.J.Lu's egcs.1.0.3 (available in binary form on sunsite and
works well) passed and failed, I'd be that many steps closer to being
confidant in using a self compiled snapshot.  For some reason, I haven't
had much luck compiling egcs (or pgcc) and then compiling a bootable Linux
kernel (currently 2.1.105).

>   > And what do you guys think about pgcc?  It appears to
>   > use the haifa flag when compiling.  Does it optimize much better than
>   > egcs?  Sorry for all these questions, but they don't seem to be answered
>   > on the FAQ.  ;-)
> Using haifa on the x86 processors may or may not be a win; I'm not
> sure if it's been throughly analyzed on those targets.
> 
> I'd be a little worried that haifa's more aggressive scheduling can
> lengthen register lifetimes which has more effect on the x86 targets
> than any other target which uses instruction scheduling.
> 
> pgcc generally optimizes better than egcs for specific (all?) x86
> platforms.  Many of the changes are not appropriate for a compiler
> that wants to work on many platforms like egcs.  The good news is
> we are slowly trying to provide equivalent pgcc functionality in a
> more maintainable and generally useful way.  It takes signifiant
> time though :-)

Completely understood.  Thanks for the tips!

-Dave