Hello.

I'd like to embed Kawa into a Java program, using it as a base for a
custom Scheme-like (but almost certainly not R*RS compatible) language.
Leaving aside resource handling issues (such as scripts exhausting all
available memory, spinning at 100% cpu usage, etc), I'm trying to work
out how I can expose an utterly spartan bare-minimum interpreter to the
host program that can only call a few functions that I expose to it.
Anyone familiar with embedding Lua into a C program (or even into a
Java program via something like Rembulan [0]) will probably be
familiar with the idea: The language is used just to provide the basic
syntax and evaluation semantics, but the standard library is more or
less completely removed and replaced with a bare minimum API relevant
to the domain in question. Doing this provides a relatively safe
sandbox, because the sandboxed code simply doesn't have access to any
functions that can do anything dangerous.

I'd like to state beforehand that I'm trying to avoid using the Java
SecurityManager unless it's utterly unavoidable (due to performance and
administrative concerns, along with the fact that Oracle might be
deprecating it eventually).

I have the following questions after playing with the Kawa API a bit:

1. Is it possible to restrict the initially available symbols in a
kawa.standard.Scheme instance to a tiny core subset (such as lambda,
if, define, begin, etc)? A default Scheme instance in Kawa has 807
symbols in the environment.

2. Is it possible to restrict the interpreter to only working with a
single java.nio.file.FileSystem? I'd like it if any attempt to do I/O
went through a given filesystem instance. I don't mind if I have to
implement my own I/O library to do this.

3. Is it possible to restrict the classes that the interpreter is
allowed to access or import? For example, right now nothing stops the
someone from writing (java.lang.System:exit 0).

[0] https://github.com/mjanicek/rembulan

-- 
Mark Raynsford | http://www.io7m.com