Re: Seeking other archs to build packages on
Roman Hodek <Roman.Hodek@informatik.uni-erlangen.de> writes:
> > Another point in favour of crosscompiling is, that some errors will
> > occur that don´t occur on i386. *Thinking about those smart i486
> > optimisations flags that fail on non-i386*
>
> You're right, but in another way than you think :-) Yes, some errors
> will only show up on cross compiling. But those -m486 & Co. flags will
> also fail if you compile natively on e.g. m68k... :-) The main point
Yes they will, and they do. And every time I'm pissed of. Maybe it
should be policy not to use those optimisations, at least not
hardcoded. Optimizing for i486 is stupid for i386, doesnt gain much on
i486 and (I think) even slows down on 586.
> is that cross-compiling code with floating point calculations is
> dangerous to impossible. You should be sure that the source doesn't
> include float or double vars....
That shouldn't be a problem. The ansi floats and doubles are perfectly
fine. I reinstalled my system after a crash with a crosscompiled
unstable kernel (the only possibility at that time), no problems
there. Also the m68k (or other) asm will scream when you use long
doubles or other i386 only stuff. gcc shouldn't generate such things
anyway and if it does, its good to find that bug.
> So this argument again speaks *against* cross-compiling.
I don't see any reason against it.
> But, BTW, I'm not against cross-compiling in general, only in this
> case (mass recompilation). I cross-compile most of my packages for
> m68k, too, but I can test the results on a m68k machine, and I know my
> packages so far that I can tell cross-compiling doesn't cause any
> special harm.
Packages should be marked as untested, if they are untested. But thats
the only thing I see needed for crosscompiled packages.
> Roman
May the Source be with you.
Mrvn
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: