Re: [patch] m68k extendqidi2 fix
> >
> > I think you should apply the patch, because the build daemons use gcc 3.3
> > for building the packages. I can provide an indirect test to the patch in
> > the next upload of the octave2.1 package. In the last upload, I changed a
> > CFLAG from -O1 to -O0 for m68k and the compilation succeeded (I have no idea
> > whether this really contributed to fix the problem). I can restore the old
> > flag and we will see what happens.
>
> Now that explains why it suddenly built!
>
> So let's ask the other m68k people if it should be included. Could it break
> more than it fixes?
Unlikely. Code that breaks when less optimization is used would qualify as
horribly broken by design. (There's an exception to this rule, as always:
gross misuse of varargs on PowerPC is flagged as an error by the compiler
when optimizing, while tolerated without optimization. So make the -O0
m68k only).
We've been using hacks like this for a long time for m68k, at times even
without corresponding patches in the source (now heavily frowned upon).
They make packages build but don't fix the underlying gcc bugs. I'd
suggest sending the assembly code generated by -O0 to the gcc team,
together with the ICE message.
Michael
Reply to: