cg at cdegroot.com (Cees de Groot) wrote:
	To me, sounds like a perfect model for the Squeak VM. Dynamically loaded
	objects aren't any slower than statically linked librarys on the OSes I 
	know, and on the Smalltalk level the issue is non-existant anyway. 
	
SPARC Solaris, Alpha OSF/1: statically linked libraries are faster on those
machines.  PPC MacOs 8: the procedure calling convention is set up to
favour dynamically linked code, a calling convention for statically linked
code _could_ be quite a bit faster but I don't know whether any compilers
exploit that.

The SmallEiffel compiler does whole-program analysis and gets very worthwhile
speedups by doing so.  The Stalin Scheme compiler does whole-program
analysis and really agressive inlining which wouldn't be feasible in a
dynamically linked approach.  I've been _amazed_ at the things the
SPARCompiler C compiler will do when it is given a whole program in one
file.

Make no mistake:  with today's machines and today's top end compiler
technqiues, dynamically linked code can be somewhat to much slower than
statically built code.

Someone was doing type inference for Squeak.  Whatever happened to that?

Of course code can be modular without being dynamically linked, and
it can be dynamically linked without being modular.