Stimsym

Rewrite system designed for symbolic manipulations and maximal expressiveness.

Why?! Well, I see that Mathematica as a symbolic language has a lot of nice features, and is rooted in term rewriting. However, because of its price and closedness, I started this small project. The goal is certainly not to reimplement all the complex mathematical primitives of Mathematica; instead, I just want a terse, powerful language in a sandbox (here, a Jupyter notebook or a CLI interface).

I drew a lot of inspiration from https://github.com/jyh1/mmaclone/, which was also my reference for operator priorities for the expression parser. Stimsym is not yet feature complete (it lacks AC matching, among others, for now), but it has a Jupyter interface and I believe the architecture might scale a bit better if lots of primitives were added. Most of the code for the Jupyter frontend is adapted from IOCaml so as to use lwt-zmq and work with Jupyter 4.

Build

I recommend using opam. The following will clone this repository, and use opam to install the dependencies and the program itself

git clone https://github.com/c-cube/stimsym.git
cd stimsym
opam pin add -k git -n stimsym
opam install stimsym

For the jupyter frontend, some more dependencies are needed (see 'opam' to see which one exactly, but it should be something along: opam install zmq atdgen yojson uuidm lwt lwt-zmq).

Usage

Once installed, you can either use the simple command line interface:

$ stimsym_cli
> 1+1
2
> f[1,2,g[3],4,5] //. f[l1__,l2__,g[x_],r__] :> h[r,x,l2,l1]
h[4,5,3,2,1]
> {f[a],f[b],f[c],f[d]} //. f[x_] /; ((x===a)||(x===c)) :> g[x]
List[g[a],f[b],g[c],f[d]]

Some Examples

See the list of commented examples to get an idea of how to use Stimsym. The file 'tests/run_tests.ml' also contains a lot of small test cases.

Some notebooks can be found on the gallery branch, including an improved version of the examples.

Jupyter Interface

Use the (experimental) library jupyter-kernel. Recommended way is jupyter kernelspec install data/ --name=stimsym --user --replace You can also copy manually 'data/kernel.json' into the directory ~/.local/share/jupyter/kernels/stimsym.

mkdir -p ~/.local/share/jupyter/kernels/stimsym
cp data/kernel.json ~/.local/share/jupyter/kernels/stimsym/

opam pin add -k git jupyter-kernel https://github.com/ocaml-jupyter/jupyter-kernel.git

Start the jupyter notebook with jupyter-notebook; you can find a sample notebook in 'data/notebook1.ipynb'. The make jupyter target will open the sample notebook.

License

BSD license, you can modify as you like. Contributions are welcome.

Hacking

Almost everything in the language itself can be found in src/core/Expr.ml, src/core/Eval.ml{,i} and src/core/Pattern.ml{,i}; the primitives are in src/core/Builtins.ml. The rest is about parsing, CLI, jupyter frontend, etc. The code is relatively naive and will certainly not perform well, and there is a lot of room for algorithmic improvement.

If (who knows?) you are interested in hacking on this in any way, do not hesitate to contact me or just say "hi" on IRC. I'd be interested in discussing or helping.

Why the name?

It is a reference to Dan Simmons' Hyperion series of books, with the suffix "sym" altered because, hey, we're doing symbolic stuff around here!