PORTING

		Porting Caml Light to another system
		------------------------------------

This note contains some hints on how to port Caml Light to a machine
that does not run Unix.

ARCHITECTURAL REQUIREMENTS:

Caml Light has been designed for 32-bit architectures, with
a flat address space. Porting to segmented address spaces is very
difficult (the 8086 port required a huge amount of work). 
The code contains support for 64-bit architectures, but has only been
tested on a Dec Alpha.

COMPILER REQUIREMENTS:

You'll need a robust, but not too fussy ANSI-C compiler. Robust,
because the code for the runtime system is pretty complex (one huge
function, a few large macros), and has been reported to break some
compilers. Not too fussy, because some parts are in old-style C, and
will probably be rejected by compilers that insist on doing inter-file
type checking.

OPERATING SYSTEM DEPENDENCIES:

Caml Light has been developed under Unix, but actually uses very few
Unix-specific features. All system functions used are in the standard
ANSI C library, except open, read, write, close and lseek. Many
compiler libraries provide emulations for these. Otherwise, you can
emulate them on top of the standard IO library (fopen, fread, ...).
That's kludgy, but it should work.

HOW TO PROCEED:

First port the runtime system (directory src/runtime) to the new
machine. You'll have to write the configuration files (directory
config/) by hand.

Test the runtime system by compiling some Caml Light sample programs
on a Unix machine, then transferring them (in binary mode) to the
target machine, and run them through the runtime system. (Bytecode
files are portable across architectures.)

You can't just transfer camlcomp, camllink, camltop on the target
machine and execute them as is. Well, you can execute them, but they
won't work. That's because they contain some Unix-specific ML code,
e.g. for filename handling. Instead:

- Adapt src/lib/filename.ml and src/compiler/config.ml to the target
operating system. filename.ml contains functions to handle file names.
config.ml contains various constants, default names, prompts, ... In
the current setting, they are derived from .mlp files, that contain
the code for Unix, MS-DOS and Mac-OS, inside #ifdef directives. You
can either add your code to the .mlp files, or write the .ml files by
hand.

- Compile all the ML code on a Unix machine.

- Transfer the files to install (bytecode executables + standard
library) to the host machine.

- You should now be able to bootstrap the system on the host machine
if you wish.

Finally, you will have to write the driver programs camlc, camllight
and camlmktop. You can either translate the shell scripts from the
Unix version, or start with the C implementations found in the PC
version.