> I have tried something similar some time ago, building next68k -current
> with gcc-2.95 for i386 (BSD/OS). So far I haven't had any success.
> Problems ranged from being unable to compile certain assembler files to
> simply not working code (that failed in very strange ways). I didn't
> take the time to investigate further.

You have to update gas to get past the assembler errors.

In preparation for the eventual ELF migration, the assembly files are being
made ELF-compatible. This means adding % to all registers, and someday it
means removing underscores from C labels.

> My solution to that problem was to take the 1.4 binaries and start from
> there. That way I was getting a -current kernel compiled (at that time,
> that is), but unfortunatly -current isn't working right and even more
> unfortunatly Darrin does not have the time to fix things and neither 
> have I. :((

Argh. That's what I did, and man does it have problems. The clock chip
handling appears to be scrambled (it gets the time wrong, then leaves
the clock chip in a state such that NextStep says "CLOCK CHIP NOT SET
PROPERLY -- CHECK AND RESET THE DATE!"). Any attempt to use rdate to
set it crashes the kernel immediately. I filed a PR.

> I can't comment on that, because I haven't tried to build a recent
> -current as I know that it's broken in other ways.

The last one I tried was 1.4M.

> complaining that it breaks cross-compile. But then it's marked XXX
> anyway, and I don't really see any easy way out.

That's tftp.h !! How many hardcoded constants or struct definitions can
we be using from there? I think it'd be safe to put a literal copy of the
NetBSD header into the boot program...

Todd Whitesel
toddpw @ best.com