What is a good VM for developing a hobby language?

Question

I'm thinking about writing my own little language.

I found a few options, but feel free to suggest more.

• JVM
• Parrot
• OSA

A lot of languages are using the JVM, but unless you write a Java-ish language, all the power the stdlib gives you is going to feel ugly; It's not very good at dynamic stuff either.

Parrot seems a good VM for developing languages, but it has a little abandoned/unfinished/hobby project smell to it.

OSA is what powers Applescript, not a particularly well known VM, but I use Mac, and it offers good system integration.

CLR+Mac doesn't seem a good combination...

My language is going to be an object orientated functional concurrent dataflow language with strong typing and a mix of Python and Lisp syntax. Sounds good, eh?

[edit]
I accepted Python for now, but I'd like to hear more about OSA and Parrot.

Answer 1

• Lua has a small, well-written and fast VM
• Python VM - you can really attach a new language to it if you want. Or write (use?) something like tinypy which is a small and simple implementation of the Python VM.

Both options above have access to useful standard libraries that will save you work, and are coded in relatively clean and modular C, so they shouldn't be hard to connect to.

That said, I disagree that Parrot is abandoned/hobby. It's quite mature, and has some very strong developers working on it. Furthermore, it's specifically a VM designed to be targeted by multiple dynamic languages. Thus, is was designed with flexibility in mind.

Answer 2

Have a look at LLVM. It's not a pure VM as such, more a framework with it's own IR that allows you to build high level VMs. Has nice stuff like static code analysis and JIT support

Answer 3

One approach I've played with is to use the Python ast module to build an abstract syntax tree representing the code to run. The Python compile function can compile an AST into Python bytecode, which exec can then run. This is a bit higher level than directly generating bytecode, but you will have to deal with some quirks of the Python language (for example, the fundamental difference between statements and expressions).

In doing this I've also written a "deparse" module that attempts to convert an AST back to equivalent Python source code, just for debugging. You can find code in the psil repository if you're interested.