To quick start, see the very simple tutorial! And here live version in Google's Colab that one can interact with (make a copy first) -- if link doesn't work (Google Drive sharing publicly is odd), download it from github and upload it to Google's Colab.
Go is a compiled language, but with very fast compilation, that allows one to use it in a REPL (Read-Eval-Print-Loop) fashion, by inserting a "Compile" step in the middle of the loop -- so it's a Read-Compile-Run-Print-Loop -- while still feeling very interactive.
GoNB leverages that compilation speed to implement a full-featured (at least it's getting there) Jupyter notebook kernel.
It's still experimental. This is very fresh from the oven, and likely there are many nuanced (or not so nuanced) situations where it may not work as expected. Reports of issues and even better fixes are very welcome.
The tutorial explains, but in short:
$ go install github.com/janpfeifer/gonb@latest
$ go install golang.org/x/tools/cmd/goimports@latest
$ go install golang.org/x/tools/gopls@latest
$ gonb --install
And then (re-)start Jupyter.
GoNB opens a named pipe (set in environment variable GONB_PIPE) that a program can use to directly
display any type of HTML content.
For the most cases, one can simply import
github.com/janpfeifer/gonb/gonbui:
the library offers and convenient API to everything available. Examples of use in the
tutorial.
If implementing some new mime type (or some other form of interaction), see kernel/display.go for the protocol
details.
Many! Contributions are welcome. Some from the top of my head:
- Mac and Windows:
- Installation.
- Named-pipe implementation in
kernel/pipeexec.go.
- Tracking of lines on generated Go files back to cell, so reported errors are easy to follow. In the meantime the errors can be moused over and will display the lines surrounding them.
- Run
goplsas a service (as opposed to invoking it every time -- super slow). - Add auto-complete with
gopls. - Library to easily store/retrieve calculated content. When doing data analysis so
one doesn't need to re-generate some result at a next cell execution. Something
like
func CacheResult[T any](id string, fn func() (T, error)) T, errorthat will either loadTfrom some storage, and if not stored, callfn()to generate the result and save it again. This way results can easily be reused on different cells.
The Jupyter kernel started from gophernotes implementation, but was heavily modified. Also, the execution loop and mechanisms are completely different and new.