Jupyter MicroPython Kernel
Jupyter kernel to interact with a MicroPython ESP8266 or ESP32 over its serial REPL.
Also with capabilities to work through the WEBREPL (available on ESP8266 only), do Ctrl-C, transfer files and esptools flashing (useful for deployment). See https://github.com/goatchurchprime/jupyter_micropython_developer_notebooks for examples.
Then clone this repository to a directory using TortoiseGIT or with the shell command (ie on a command line):
git clone https://github.com/goatchurchprime/jupyter_micropython_kernel.git
Install this library (in editable mode) into Python3 using the shell command:
pip install -e jupyter_micropython_kernel
This creates a small file pointing to this directory in the python/../site-packages directory, and makes it possible to "git update" the library later as it gets improved. (Things can go wrong here, and you might need "pip3" or "sudo pip" if you have numerous different versions of python installed
Install the kernel into jupyter itself using the shell command:
python -m jupyter_micropython_kernel.install
(This creates the small file ".local/share/jupyter/kernels/micropython/kernel.json" that jupyter uses to reference it's kernels
To find out where your kernelspecs are stored, you can type:
jupyter kernelspec list
Now run Jupyter notebooks:
In the notebook click the New notebook button in the upper right, you should see your MicroPython kernel display name listed.
The first cell will need to be something like:
or something that matches the serial port and baudrate that you connect to your MicroPython/ESP8266 with.
You should now be able to execute MicroPython commands by running the cells.
'''On Windows it can sometimes be difficult to find the Serial (COM-port) and the right driver. This is not unique to the jupyter_micropython_kernel and is a function of the USB chip that is on the breakout board containing your ESP32/ESP8266. Find the USB connection in the Device list to see what driver it needs or look for instructions from the supplier of the board.'''
There is a micropythondemo.ipynb file in the directory you could look at with some of the features shown.
If a cell is taking too long to interrupt, it may respond to a "Kernel" -> "Interrupt" command.
Alternatively hit Escape and then 'i' twice.
To upload the contents of a cell to a file, write: %sendtofile yourfilename.py
as the first line of the cell
To do a soft reboot (when you need to clear out the modules and recover some memory) type: %reboot
Note: Restarting the kernel does not actually reboot the device.
Also, pressing the reset button will probably mess things up, because this interface relies on the ctrl-A non-echoing paste mode to do its stuff.
You can list all the functions with: %lsmagic
For reference, the notebooks here might be useful: https://github.com/goatchurchprime/jupyter_micropython_developer_notebooks
The system works by finding and connecting to a serial line and then issuing the enter paste mode command Ctrl-A (hex 0x01)
In this mode blocks of to-be-executed text are ended with a Ctrl-D (hex 0x04).
The response that comes back begins with an "OK" followed by the actual program response, followed by Ctrl-D, followed by any error messages, followed by a second Ctrl-D, followed by a '>'.
You can implement this interface (for debugging purposes) to find out how it's snarling up beginning with: "%serialconnect --raw" and then doing %writebytes -b "sometext" and %readbytes
This had been proposed as an enhancement to webrepl with the idea of a jupyter-like interface to webrepl rather than their faithful emulation of a command line: https://github.com/micropython/webrepl/issues/32
My first implementation operated a spawned-process asyncronous sub-kernel that handled the serial connection.
Ascync technology requires the whole program to work this way, or none of it.
So my next iteration was going to do it using standard python threads to handle the blocking of the serial connections.
However, further review proved that this was unnecessarily complex if you consider the whole kernel itself to be operating asyncronously with the front end notebook UI. In particular, if the notebook can independently issue Ctrl-C KeyboardInterrupt signals into the kernel, there is no longer a need to worry about what happens when it hangs waiting for input from a serial connection.
Other known projects that have implemented a Jupyter Micropython kernel are:
In my defence, this is not an effect of not-invented-here syndrome; I did not discover most of these other projects until I had mostly written this one.
I do think that for robustness it is important to expose the full processes of making connections. For my purposes this is more robust and contains debugging (of the serial connections) capability through its %lsmagic functions.
Other known projects to have made Jupyter-like or secondary interfaces to Micropython:
The general approach of all of these is to make use of the Ctrl-A
paste mode with its Ctrl-D end of message signals.
The problem with this mode is it was actually designed for automatic testing rather than supporting an interactive REPL (Read Execute Print Loop) system (citation required), so there can be reliability issues to do with accidentally escaping from this mode or not being able to detect the state of being in it.
For example, you can't safely do a Ctrl-B to leave the paste mode and then a Ctrl-A to re-enter paste mode cleanly, because a Ctrl-B in the non-paste mode will reboot the device.