> > That sounds like making life hard for the many in order to make life 
> > easier for the few.
> > 
> > That's backwards in my opinion.  Anyone doing a retrofit as you suggest 
> > isn't going to be worried anywhere near as much about such an anomaly as a 
> > normal day-to-day user, with a correctly installed system, who can't get 
> > proper system dumps or whatever.
> > 
> > Anyway, why would anyone installing a new kernel on an a.out system want 
> > to install their kernel under any name other than "netbsd"? -- the patch 
> > only removes a ".aout" extension if it is present.
> 
> The case that mrg's comment made me worry about was this:
> 
> 	start with a stock 1.5.x system (/netbsd is a.out)
> 	install /netbsd.aout from 1.6 to test retrofit
> 	boot disk:\netbsd.aout
> 	machdep.booted_kernel claims the OLD kernel was booted

But that's easily avoided by installing your test kernel as, say, 
/testnetbsd (or indeed, anything that doesn't end in .aout).

> 
> Is it safe to call printf() from ofw_device_register()?  If we can add a
> status line to note that the kernel name is being changed, then I'm happy.

I'm not sure, shouldn't be hard to test though -- kernel will either print 
things correctly or lock up.

> BTW this is pullup-1-6 #1081, so if the consensus is that shark owners are
> expected to be clueful, that is fine too and I'll tell them to take it as is.

Something similar on Cats would be useful as well.

R.