wolfram language kernel for jupyter notebook

Installation

sed -i "s|init_file|"$(realpath init.wl)"|g" wolframforjupyter/kernel.json
jupyter kernelspec install wolframforjupyter/ --user

No explicit version implied, it depends on wolfram execute version in your $PATH.

For code completion of user defined symbols. (change Wolfram to Mathematica if version under v14.0)

ln -s $(realpath stream.wl) ~/.Wolfram/Applications/stream.wl

• [ ] If not, system symbols completion still work.?

Kernel

Jupyter Client 8.3 — jupyter_client 8.3.1 documentation

Language for kernel

• Making kernels for Jupyter — jupyter_client 8.3.1 documentation

Obviously we should choose option 2 (code execute done in WTSP)

Debug

colored debug message

• bash - How to change the output color of echo in Linux - Stack Overflow

Examples

• Making simple Python wrapper kernels — jupyter_client 8.3.1 documentation
• GitHub - jupyter/echo_kernel: A simple example kernel for Jupyter
• GitHub - takluyver/bash_kernel: A bash kernel for IPython
• GitHub - dsblank/simple_kernel: A small, simple kernel example for IPython

Test

GitHub - jupyter/jupyter_kernel_test: A tool for testing Jupyter kernels

ZeroMQ Socket

ZeroMQ | Get started

zmq identity

• 3. Advanced Request-Reply Patterns | ØMQ - The Guide

UTF-8 encode string

• Reading an UTF-8 encoded text file in Mathematica - Stack Overflow

Evaluation

WSTP Wolfram Language Functions—Wolfram Language Documentation

currently, interrupt in jupyter is implemented by LinkInterrupt. For signal interrupt see Linux Signals Help

Notation:

client

jupyter front end (i.e. browser tab / vscode / emacs-jupyter)

kernel

the monitor process client communicate with

link

the wolfram language evaluation take place, monitored by kernel

Code input

Currently, codes are input as EnterTextPacket to link. That makes output render easily, since output being wrapped inside ReturnTextPacket. But have following downside:

1. unicode garbled
2. further handler difficult (e.g. Short)
3. code completion hard implement (see below)

Asynchronous Execution

LocalSubmit failed inside WSTP link, then there’s no “caveat for asynchronous output” (mentioned in jupyter client documentation). Anyway, why not launch a new link (jupyter session)?

Completion

Where should completion occurs, in kernel or link? Jupyter’s answer is surely the later, since the complete_request being send from shell channel. And the symbols in link is non-awareable to kernel. (for simple, currently front)

Then, the problem emerge when completion request arrive at link busy status. But like in jupyter-python, is ok for link to ignore those requests (check it’s the case). And for completion inside link, we need ReturnPacket or ReturnExpressionPacket to store matches in List (not string of lengthy list), which is not implemented.

Other problem is that execute_count get increase when handle completion request, see $Line variable for maintain it.

Features [3/8]

• [ ] check completion request after link go from busy to idle
• [ ] code completion
• [ ] Short with too long expression
• [ ] log evaluation time
• [ ] export
  • Customizing exporters — nbconvert 7.8.0 documentation
• [X] interrupt
• [X] % no content
• [X] message stream output

Similar

WolframLanguageForJupyter

• GitHub - WolframResearch/WolframLanguageForJupyter: Wolfram Language kernel f…

  Jupyter kernel from WolframResearch official. Very nice structured code, this project borrows many ideas from there.

wolfram-language-notebook

• GitHub - njpipeorgan/wolfram-language-notebook: Notebook support for Wolfram …

  Functional wolfram notebook for vscode.

IWolfram

• GitHub - mmatera/iwolfram: A jupyter kernel for Wolfram Mathematica

  Kernel written in python and

  So far this is a proof of concept.