> From: Mike Long <mike.long@analog.com>
> X-Loop: current-users@NetBSD.ORG
> 
> 
> >From: Iain Hibbert <plunky@skate.demon.co.uk>
> >Date: Fri, 1 Sep 1995 10:28:18 +0100 (BST)
> 
> >this binary compatibility thing..  something that I found a while ago
> >was a paper by a french guy (I think) about something he called Semantic
> >Dictionary Encoding - basically, it is code generation at load time,
> >the compiler or whatever just creates a description of the algorithm
> >and the loader generates the code on the fly as it loads..
> 
> The OSF has something like this.  They call it ANDF
> (Architecture-Neutral Distribution format).  The intention is to be
> able to distribute one binary for multiple systems/architectures,
> without having to reveal proprietary source code.  It's similar to the
> RTL that gcc uses to communicate between its front- and back ends.
> 
> IMHO, the equivalent of a compiler back end would be a bit much to
> stick into the kernel. :-)

Which is why the SVR4 variants all have a user space loader.  Having
the kernel link/load programs has never seemed like a good idea to
me.  Newer systems start executing in a library requiring basically
no other services and this library then loads the intended program.
The actual execution time is similar (though you do have to add the
protection boundary crossing times) and its easy to add new loaders
(or even complete interpreters).

-everettm