Installation

local

• See the Dockerfile for a recipe on how to build the tclkernel.

Docker

The docker folder contains the Dockerfile for building a tcljupyter container based on alpine. An example docker-compose.yml file is added.

Prebuilt images are on Docker Hub.

Binder

If you want to try it out without installation you can use the Binder version in the cloud.

• (with binary zmq library)

Dependencies

The Tcl used to run the init.tcl script should have the following available:

• Tcl 9,0 with threads
• tclzmq (Not for the tclonly version)
• rl_json custom built version for Tcl 9.0
• tcllib uuid
• tcllib sha256

Build tclzmq on windows

Easiest way to build tclzmq on Windows is to use mingw.

• Install tcllib
• Install critcl
• Install the msys zmq packages

pacman -Syu mingw64/mingw-w64-x86_64-gcc mingw64/mingw-w64-x86_64-zeromq

• Run the build.tcl installer from the tclzmq distro. NB: Static builds don't work in the sense that they are the same as shared builds.

Supported

Most web client commands are supported. Only thing missing is reading from stdin with for example gets.

Commands

• jupyter::display mimetype body: Display body in the cell. Returns the display id for use in updatedisplay.
• jupyter::html body: Display body as html in the cell. Returns the display id for use in updatedisplay.
• jupyter::updatedisplay id mimetype body: Updates the display with id id with then new body.
• jupyter::updatehtml id body: Updates the html display with id id with the new body.

Magic commands

Magics are lines starting with ## at the top of the cell.

• ##timeit ?count?: Time the execution of the cell and display the results. Changing count will execute the cell multiple times and show the average time.
• ##noresult: Do not show the last result. This is convenient if the last result is very large. The same effect can be achieved by postfixing a ; as the last character in the cell.

Test suite

A prequisite is the jupyter_kernel_test (JKT) module:

pip install jupyter_kernel_test

To run the test suite, execute (from within the checkout directory):

python -m tests.test_tcljupyter

Design

For communication from kernel to session thread [thread::send -async] is being used. stdout and stderr are being intercepted by chan push