clang-repl

clang-repl based kernel for Jupyter notebooks

Overview

• This is a kernel enabling using C++ in a Jupyter Notebook
• It bases on a minimalistic instrumentation of the interactive clang-repl prompt using
  • this showcase for ipykernel.kernelbase.kernel as blueprint for the overall project structure, and
  • python-pexpect for instrumenting the clang-repl> prompt.

Motivation

• The motivation is to enable notebook-based learning material for C++.
• The focus is not on language interoperability between C++ and Python.
• Exploring the latest capabilities of clang-repl using different C++ language standards/settings in a convenient way.
• Existing projects providing a form of interactive C++ in notebooks do currently not support C++20/C++23 and use fixed/patched LLVM versions.

Details

• It is required that clang-repl is installed on the (backend) system
• On launch, the kernel starts an interactive clang-repl session.
• The default settings and initial includes/libs can be configured by placing a .clang-repl file in the users home directory, the defaults (if not .clang-repl file is present) are listed below

  [defaults]
  repl = "clang-repl"
  args = ["-std=c++20", "-ferror-limit=3", "-O1"]
  includes = ["vector", "iostream"]
  libs = []
  timeout = 10
  debug = false
  

• The kernel performs the following steps for each source cell 0. Check if the clang-repl session is alive
  1. Inspect first line of the cell if starting with these magic commands
     • %status: print kernel status
     • %lib: forward first line of cell directly to clang-repl
  2. Comment the first line (by prepending //) if it starts with %
  3. Transform the cell content if the first line contains a %main: the cell content is wrapped and run via a unique global function, e.g. void mainUUID(){ ... }; mainUUID();
  4. The (transformed) cell content is forwarded to clang-repl by always using a single line command realized via a indirection of, e.g. this form: #include /tmp/cell-e3tp24ne.repl
  5. The result of the interactive session (i.e. incremental compile + execute) is awaited (using a timeout) and printed as output of the cell.
  6. If the cell additionally contained a %undo in the first line (and the incremental compile + execute was successful) the cell is "undone" via sending a subsequent %undo directly to clang-repl

Installation

python -m venv .venv
source .venv/bin/activate
python -m pip install clang-repl 

# demo notebook
jupyter notebook demo.ipynb

# new empty notebook
echo '{ "cells": [], "nbformat": 4, "metadata": {} }' > empty.ipynb
jupyter notebook --MultiKernelManager.default_kernel_name=clang_repl empty.ipynb

Installation (for development)

git clone https://github.com/pmanstet/clang_repl_kernel.git
cd clang_repl
python -m venv .venv
source .venv/bin/activate
python -m pip install -e .
jupyter kernelspec list 

# interactive console
jupyter console --kernel clang_repl

# demo notebook
jupyter notebook demo.ipynb