[<<][libprim][>>][..]
Fri Aug 7 19:14:57 CEST 2009

Bootstrapping

Bootstrapping is fun.  It's the first time I did this for a Scheme
interpreter.  I've added two mzscheme .ss files.  One to convert .scm
files to .c_ files as long as there is no reader, and one to lift the
primitives from the .c file and write boilerplate registration code.

Then bootstrapping the rest of the basic scheme functionality on top
of the bare words is quite interesting.

One thing though: I don't immediately see how to make apply and eval
work.  Eval is probably iterating the step() function.  But apply?
Similarly maybe?  Constructing a continuation frame and iterating the
interpreter?

There is probably a simpler though non-obvious way to do this..
Manipulating the machine state from inside a primitive would probably
do the trick.

The trick is that the function modifies the continuation.  This could
be done using mutation, but that won't play nice with running the
interpreter manually (as part of debugging for instance).  It's
probably best to make it possible for primitives to return a
continuation.  Or better: to allow for explicit "state transforming"
primitives.

Let's thinkg about this first..

Before the step() function can really be a pure function, it has to
isolate all side effects, and should not rely on a main loop's
actions.

Maybe it's best to first try to implement call/cc.

Applying a continuation is really simple.  But how to capture it?
Looks like that's similar to apply.

Then about errors.  Maybe it is better to keep errors local to the
step() function that generated them, and return them as values
containing the continuation.  Except for GC restarts: they need to go
all the way up to make sure data is not corrupted.

Ok. errors are now local to step() allowing for recursive step() but
GC restart goes all the way up the chain.

Then a surprise: eval is (make-ast '(post 'foo-foo))

If a primitive returns a reducable term (represented by the ast
wrapper type), it will be evaluated.

Maybe apply can be made in a similar way?  Well.. I found something
that uses eval and map and quote.  Not pretty..

Looks like manipulating the continuation is best.



[Reply][About]
[<<][libprim][>>][..]