showell / bot_computer

little programming language

Overview

This project is inspired by some conversations I had with my friend David, who I met through the Zulip project. I am using only his first name to protect his privacy and to avoid conflating his inspiration for my ideas with my own interpretation/exploration of them.

Zulip is a chat system where humans and bots send messages to each other. David and I are imagining a Zulip ecosystem where humans can easily program bots and see their source code and dynamically create new bots with little fanfare. One prerequisite for that is to have some kind of sandboxed programming language that folks can write in.

This project implements a little programming language called "messie" that is aimed to be an LCD, Turing-complete language that can map to more expressive languages like Python/Ruby/JS/Lisp, but with a little more minimalism and adaptability to a messaging paradigm. The name "messie" refers to messaging, not messiness. The language looks superficially messy, but I am striving to make the underlying architecture simple and clean.

Minimal

The language is intended to be minimal in these regards:

• You cannot mutate data structures, only create new ones.
• There are no variables or assignments.
• There are only a small number of built-in functions.
• The only syntax is JSON (sort of).

Complete

The language is intended to be complete in these regards:

• You will be able to manipulate all the JSON-serializable types, such as integers, floats, strings, lists, and dictionaries.
• You can have functions that work on other functions, such as MAP and APPLY.
• You can create recursive functions like FACTORIAL.
• It's Turing complete, and you can do things like create an HTML table of factorial values with relative ease.

Unique

The language is unique in these respects:

• Functions communicate by passing around JSON data structures as messages.

Extendible

The language is extendible:

• Because "messie" fundamentally works with messages, the VM could be extended to work with multiple processes or multiple computers.
• Because "messie" code is just JSON, functions written in "messie" can be up-ported to faster or more expressive languages like C/Python/Ruby/JS/Lisp.
• Because "messie" code is just JSON, functions written in higher level languages can be down-ported to "messie" if you subset those languages to use only functions and immutable types.

Implementation

There is a script called [computer.py] (https://github.com/showell/bot_computer/blob/master/computer.py) that shows examples of "messie". You run it by simply invoking python computer.py.

The computer runs "messie" using a VM that is all in-process, but which uses a message queue to dispatch calculations. It can handle deeply recursive functions despite Python's recursion limit, because function dispatch is mediated through a queue.

The computer currently calculates a bunch of messages to standard output, and then it also calculates an HTML table of math results that it writes to foo.html.

Here is example HTML output from "messie":

n	double	square	factorial
5	10	25	120
6	12	36	720
7	14	49	5040
8	16	64	40320
9	18	81	362880
10	20	100	3628800
11	22	121	39916800
12	24	144	479001600
13	26	169	6227020800

You can see the "program" that created the above table here:

https://github.com/showell/bot_computer/blob/master/commands.py