> > Just be glad you aren't using the old IBM (IIRC) system which had the
> > constants 0, 1, 2 at 'well known' addresses.  This system ran FORTRAN
> > and all arguments are passed by reference.  If a routine accidentally
> > changed a (supposedly) constant 1 that was passed to it, then the
> > value of 1 changed for the entire OS.
> 
> Still works with NetBSD f77.

Nah - it will only change '1' for that program, not the entire OS.
The FORTRAN compiler I used to used carefully generated separate
copies of each constant for every call - slightly more code space
but far less likely to cause hair loss (through yanking).

I suspect part of the reason for the optimisation was being able to
access the constants with short addresses.

PS can we have the 3-way arithmetic if?

	David

-- 
David Laight: david@l8s.co.uk