Re: Mike and Ward, what next?

Erik <groups@[EMAIL PROTECTED]
> wrote:

> > ... maybe we should go back to a tokenized model with the token
> > displatching engine written as a Universal Binary and just abandon
> > including an assembler or attempting to do non-trivial optimizing...
> >
> 
> What would happen if you used AS, the assembler that's included with
> the OS?  If my understanding is correct, AS is the optimized version of
> GAS, the GNU Assembler.  I believe GAS tries to keep instructions
> consistent across architectures, so you would have some measure of
> ****tability and optimization at the same time.  It's just a guess,
> though.  I'm not that well-versed with assembly.

I prefer Forth-based Assemblers, myself ;-)

One issue is building the core of a x86 Forth, which could be done in
any ASM, I suppose.

I was referring to the difficulty of providing a single Forth that could
build a Universal Binary snapshot/turnkey.  Having to do two compiles to
aaccomplish that sort of interferes with the incremental compile/test
style we all know and love. 

Any assembly used in user source would have to be redone in the other
CPU's instruction set, meaning one source code could not be used for
both builds. 

Non-minimal optimization for PPC and x86 would be entirely different as
well.

A tokenized Forth with some load or run time optimization is an
interesting alternative, although it starts to sound like RPN Java.

  -- w