[MIT-Scheme-devel] new VM?

[Chris Hanson]

   Date: Fri, 11 Mar 2005 02:17:40 +0000
   From: Matt Birkholz <address@hidden>

   I was just re-familiarizing myself with s-code and the interpreter
   (e.g. mit-scheme/v7/src/microcode/interp.c, from the CVS repository on
   savannah.gnu.org).  I see a comment about punting the dangerous
   reference trap types in trap.h.  And I could not see where the
   interpreter sets the frame&offset in the TC_VARIABLE objects.  Is that
   because of the "new VM" mentioned in January?  Are some things being
   simplified in preparation for that?

The "dangerous" bits were dropped quite a while ago.  It had nothing
to do with the new VM.  I did it because I no longer saw a need for
the optimization, and it added considerable complexity to the lookup
code.

   I am about to dive into src/compiler/documentation/, and am wondering
   how much LIAR will change to accommodate a new VM.

I am hoping that the changes will be minimal.  The plan is to add the
VM as a new target "machine", and try to fit it into the current
framework.  However, I had been hoping that the VM would have the same
instruction set independent of the underlying machine's word size; if
that is done, the compiler will probably need some more extensive
work.  (Perhaps that is a reason not to do this right now.)