>>>>> "Tom" == Tom Lord <firstname.lastname@example.org> writes:
Tom> 1) Don't bother trying to make a new kernel. Use an existing
Tom> one. Otherwise, at _best_, you'll spend a decade fretting
Tom> over device drivers and the tcp/ip stack.
The TCP/IP stack is a one time, simple driver. The problem with other
device drivers is that they become outdated so quickly. I buy a new
digital camera and it doesn't have any support under Linux, the HW
manufacturers won't provide specs, and Linux+C is a terrible platform
for reverse engineering and experimental driver development. It would
be much better to have a nice Lisp/Scheme repl where you can
interactively query the HW and write newer drivers faster. It would be
worth the initial time investment.
Tom> 2) Don't bother trying to make a "pure lisp" system.
Tom> Otherwise, at _best_, you'll spend half a decade fretting
Tom> about replacing X11 and writing a web browser.
X11 is a monster, better not to replace. All functional web browsers
are likewise. GUI's are not difficult, and easy to improve on.
Tom> 4) An interesting starting place -- especially appropriate if
Tom> you're interested in SCSH: work on replacing the boot
Tom> scripts of your system; then on replacing the inetd
Tom> services. Can you get to a state where the first invocation
Tom> of the (traditional) shell is after you log in? From my
Tom> personal experience, if you're a pretty good scheme/lisp
Tom> hacker, you can get a bootable system (missing a bunch of
Tom> services, of course) in a couple of weeks.
Using a C-compiled Lisp/Scheme? If you want to up the ante a few more
years, drop C and write a native compiler. Are there any free Lisp
assemblers out there? I have a Scheme assembler if there's nothing
better, but it's x86-only right now and the ELF linker is primitive and
doesn't handle shared libraries.