Scheme vs. CommonLisp vs. the World

David Gadbois
Tue, 13 May 1997 11:36:01 -0500 (CDT)

   Date: Tue, 13 May 1997 10:51 -0400
   From: Kalman Reti <reti@RIVERSIDE.SCRC.Symbolics.COM>

       Date: Tue, 13 May 1997 10:21 EDT
       From: (Henry G. Baker)

       Perhaps the 'digits'/'bigits' should be 64 bits instead of 32

   This would solve the problem I was talking about as well.

But, but, if we go with 32-bit object references on 32-machines, that
means these "bigits" will have to be heap consed.  And if you are
doing that, it is only a short step to having real bignums.

I can see using big 64- or 128-bit IDs regardless of the underlying
architecture (they would be great for really big, distributed OODBs)
so we could use an immediate representation for bigits, but 1) there
would be a big performance hit for not using the architectural sizes,
and 2) we'd have to do a *lot* more from scratch rather than adapting,
say, CMUCL and Flux.

--David Gadbois