Disclaimer: Just for fun, as an self-educational experiment. Not fit for any purpose whatsoever. Inspired by Chapter 6 of LiSP, which in turn was based upon the paper "Using closures for code generation".
This is a JavaScript metacircular "fast interpreter" which removes some of the interpretive overhead by doing a direct translation of the tree of AST nodes into a tree of closures. That is, the following (SpiderMonkey- / Esprima-style) AST snippet
{
type: "BinaryExpression"
operator: "+"
left: { type: "Identifier", name: "a" }
right: { type: "Identifier", name: "b" }
}
gets converted to something like
function() {
return (function() {
return stack[0];
})() +
(function() {
return stack[1];
})();
}
This means that we do not need to decode the AST at runtime. Additionally, where a simpler interpreter would implement variable lookup by checking a series of hash tables, we speed things up by converting the hash lookups into array references.
Compared to a full compiler, we still incur a lot of overhead due to the cost of function calls. Perhaps more importantly, since we do not linearize the AST, we are still forced to use exceptions for break/continue/return. JS engines do major deoptimization when encountering exceptions, so this is not great. Overall, it makes for an approximate 65x slowdown on SunSpider.
On the upside, this "compilation" process is fairly straightforward, and the resulting code still looks very much like a plain old eval-apply interpreter.
"Tricky" language bits that have been implemented:
- Hoisting of
varandfunctiondeclarations - Direct and indirect
eval(but without strict mode) - Immutable bindings (for the identifier of a function expression)