[dev] Re: JIT & VM discussion

From: Connor Lane Smith <cls_AT_lubutu.com>
Date: Mon, 20 Jun 2016 13:15:34 +0100

Thanks for the suggestions, everyone.

The motive for this is essentially that I'd been playing around with
designing interpreters for various Turing-incomplete programming
languages, and I kept having to design and write new abstract
machines, which seemed like a lot of wasted effort. And JIT seemed
especially nice to have, since it would reduce the interpretation
overhead, although it isn't strictly necessary. As Kamil suggests, I
think that being able to have a portable bytecode interpreter may be
key, since then the JIT can simply act as an optimisation for certain
supported platforms, and then it won't tie the program to any
particular architecture.

So I think the closest to what I'm looking for might be Dis and QBE.

My problems with Dis are that it has a lot of higher-level features
(due to Limbo), and that it assumes garbage collection. A subset might
work, but since it's unsupported now anyway it might be worth just
learning from it, instead of actually adopting it. Lua has similar
problems with high-level language features and garbage collection,
neither of which fit in C without an additional object model on top,
which I'd like to avoid.

QBE seems like it could be good, although I couldn't tell when I
looked whether the intermediate language is structured or a bytecode.
In any case, I think the way it's designed rules out a portable C
interpreter, since it has no pointer type, for instance. It might be
relatively easy to adapt to the task though.

I'll have to give it some more thought.

Thanks again,
cls
Received on Mon Jun 20 2016 - 14:15:34 CEST