> : They should have _no_ correlation between ${MACHINE_ARCH}s whatsoever,
> : or all ports should share _exactly_ the same device switch.
> 
> Okay, I'll assume the former case, to minimize both table bloat and kernel
> load.  I'll shut up on this topic now.

Actually, I'd suggest that if no one wants to argue against it, make
all ports share exactly the same device switch.  That'll reduce
maintenance.

I can't think of any serious drawbacks to it.

If nobody cares to argue vehemently against it and presents a good
drawback, then i'd say it wins from the maintenance perspective.


> Now on to figure out the stat()/fstat() userland compatibility thing that no
> one but I seems to see as a problem....

I've heard you mention a problem, and try to give examples (some of
which i've replied to already).  I don't think you've demonstrated
that the need to do 32->16 and new->old conversions in the MI kernel
code.

If you'd like to try more examples, well, go for it.  8-)



chris