"Andreas Raab" <andreas.raab at gmx.de> wrote:

> Hi Tim,
> 
> > The _down_ side is the loss of backwards compatability with all those
> > old images - but then they will still run on older VMs so maybe that
> > is not such a big problem. Really the only thing in the way is finally
> > making a decision to go forwards. Guides? Dan? Ted? Alan? Andreas?
> > Anyone?
> 
> A complex issue. From my point of view the "backward compatibility" thing
> isn't the real question here. This can be dealt with and given the current
> rate of change having an incompatible version every five years or so is
> quite acceptable ;-) So the real question is whether the changes buy us
> enough flexibility to get over the next five years or so. Some of the
> changes we're talking about here (like CM format, removing obsolete stuff)
> are absolutely worthwhile as they cost hardly anything (both in terms of
> speed and space)
Agreed; personally I'm not very worried about full backward
compatability of image file but I concede that some people might be.

> but some others (most notably the stack changes of the BC
> VM) aren't quite as obvious. Which leaves us in the somewhat inconvenient
> situation that while the "obvious stuff" doesn't really require an
> incompatible change the things that _do_ require an incompatible change
> aren't quite as obvious.
Ah, well the 'current' BC stuff doesn't need any stack related changes
(IIRC) and only needs the three or four extra prims we've already
discussed adding to the VM anyway. I postulate that a few extra
bytecodes might be of benefit later but the overall news is that no
image format change is required for some decent level of BC support.
Which ought not surprise us since BHH supported closures and a decent
context caching system nearly twenty years ago without any weird image
format.
> 
> Complex. Even more so in the light of Anthony's latest message. It makes me
> feel much more uneasy to know that he'll be going off another path and that
> we may be stuck with something that isn't quite complete (at least as far as
> I understand it right now) and hasn't been heavily tested.
I guess this is where we test the touted benefits of open source code;
we have the code, we're smart, we can ingest mushrooms, we can hack. 

I think it's well worth making the effort to put it all together and
give it a try. Old VMs can support the old images for some reasonable
time in the future and in fact I think I'll claim that with the VM code
in a package it ought to be possible (I'm not saying easy) to forward
support the older VM design for quite some time to come. Use the older
package to build old-style VMs, the new one for newer.

tim
-- 
Tim Rowledge, tim at sumeru.stanford.edu, http://sumeru.stanford.edu/tim
Any nitwit can understand computers.  Many do.  - Ted Nelson