[<<][compsci][>>][..]
Mon Aug 3 14:36:10 CEST 2009

Task-based C interface

Here's the basic idea I'm trying out for writing reusable C primitives
for different kinds of scripting languages (operating systems, in
essence Scheme and PF, a linear concatenative language).

                     C code shouldn't CONS.

C code should only communicate with the outside world using _channels_
which have a limited number of primitive types, but do not allow for
aggregates.

All aggregate data types should be transferred using _protocols_ :
explicit sequencing of primitive types to represent data structure.

Doing it this way makes it possible to write C code that doesn't
perform any memory management, except for allocation of local
variables.  This makes automatic wrapping very simple, and allows a
single abstract object: the task (zipper).  Moreover: the code itself
can be incorporated in a static scheduling policy (whole program
optimization: compile time weaving).

This technique is essentialy premature deforestation: eliminating
intermediate data structures using compile-time transformation.  The
slogan is something like this:

               Replace data structures with protocols.

From the outside, it doesn't matter that C code is stateful, as long
as _all_ the state is contained in the continuation.  The essential
insight is that all memory and control management can be abstracted.

Funny how I got here.  I've rediscovered concurrency-oriented
programming by looking for the simplest way to interface C code with
Scheme.



[Reply][About]
[<<][compsci][>>][..]