[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[MIT-Scheme-devel] Building with 32 bit pointers?

From: Chris Hanson
Subject: [MIT-Scheme-devel] Building with 32 bit pointers?
Date: Wed, 21 Jan 2004 01:16:08 -0500
User-agent: IMAIL/1.21; Edwin/3.115; MIT-Scheme/7.7.91.pre

   Date: Sat, 11 Oct 2003 19:52:57 +1000
   From: Tim Josling <address@hidden>

   1. I have Gnu/linux (Red hat 8) and have been using mit-gnu-scheme for a
   while. I am running into this 16mb heap issue, running out of memory on
   big genetic algorithm runs...

   I have the impression this is due to the use of 8 bits from the pointer
   as a type. Can I rebuild with 32 bit pointers so I get a large heap?
   Maybe there is a configuration option or do I need to patch something?

   I am on 7.5.17 of scheme but I have downloaded the patest stable release
   7.7.1 and can install that if it helps.

   I read all the doc and browsed the source for 1/2 hour or so without
   running into anything obvious. Any help appreciated,

Actually, there are only 6 bits of type code, not 8, but yes, that is
why there is a restricted amount of address space.

This is a long-standing problem, and there is no simple solution.  I
am planning to work on it this year, but it will take a while.

   2. In windows one can call arbitrary functions, but not in Unix/Linux.
   Is there any particular problem with adding this capability to Unix? Has
   any design or discussion occured about this? Would a patch to add it be
   in principle acceptable? Do you need a copyright assignment to the FSF?

There's no particular problem doing this, other than that calls to
certain kinds of C procedures require some care to get right.  There
are also some techniques that can be used to make the end result more
robust, at the expense of some efficiency.

There is an undergraduate planning to work on this during the term
starting in February.  If you wanted to work with him (or separately)
that is OK with me.  No copyright assignment is needed, but the end
result must have a GPL-compatible license.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]