Testing the waters.

Arthur A. Gleckler arthur@martigny.ai.mit.edu
Thu, 8 May 1997 09:10:48 -0700


Lars Thomas Hansen writes:
|  
|  From: Drew R Whitehouse <Drew.Whitehouse@anu.edu.au>
|  
|  >	Are there any decent native i386 scheme compilers at the
|  >moment ? 
|  
|  MIT Scheme comes to mind.  The latest version runs on both Win32 and
|  Linux, so one would imagine they've spent some time on making it
|  portable.  The one worry I have about MIT Scheme is its non-generational
|  GC, both because it doesn't have one (you definitely want generational
|  gc), but also because one wonders about _why_ it doesn't have one -- was
|  it because no-one bothered to sit down and do it, or because it would be
|  hard to fit into the existing code base?  I really don't know.

MIT Scheme was written not as an end in itself, but to support the
mostly non-language-related research efforts of Sussman's and
Abelson's group.  Therefore, new features were only added when they
directly supported what someone in the group needed.  Also, everyone
had plenty of RAM, so paging, at least, was mostly not an issue.

Switching to a generational garbage collector would probably have been
quite a bit of work (as it would have been for most implementations, I
suspect).  A good deal of time was spent tuning the garbage collector
to be very fast, and its performance was never a problem.  There is
even a version of MIT Scheme that uses disk as the second heap during
GC, a la Minsky.

In short, there was probably no deep technical reason that
generational GC was not added to MIT Scheme.