Datasette based notebook search extension, originally inspired by Simon Willison's Fast Autocomplete Search for Your Website.
Related: Sketching a datasette powered Jupyter Notebook Search Engine: nbsearch
Released: pip3 install --upgrade nbsearch
Latest: pip3 install --upgrade git+https://github.com/ouseful-testing/nbsearch.git
This is still very much a work in progress. I think it works in Binder (erm, maybe...)
Installing the extension and restarting the Jupyter server should make a Jupyter proxy served version of the search form available from the New menu.
At the moment, I think you need to create the index explicitly. From the command line:
nbsearch index -p PATH
(From a notebook command cell, prefix with !. Default path is .).
The sqlite db that stores the results is in ~/.nbsearch/notebooks.sqlite.
All notebooks down from PATH will be indexed.
When you have generated an index, launch the nbsearch panel from the notebook homepage New menu.
- create a database by passing a path to some notebook files, eg:
nbsearch index -p "/Users/myuser/Documents/content/notebooks"
- run the server, eg:
nbsearch serve
datasette should start up and display a server port number. (To kill it, I look for process IDs: ps -al |grep datasette; there is probably a better way... It might be nice if CLI kept track of process IDs and let you kill from a selection?)
Old screenshot
A copy button on a code cell lets you copy code from the code cell.
Results are limited in length; the Show all cell button expands the result cell to its full length. The Show previous cell and Show next cell buttons display the full previous / next cell (repeatedly clicking these buttons grab the next next and previous previous cells etc.)
Clicking on the notebook structure visualisation graphic (the pink/blue image: the colours representent cell type and relative length) will collapse / reveal the display of the result block.
The links to notebooks may well be broken in the search results: I need to think about how to index and handle paths in links, particular in proxy server case.
The app requires the latest version of datasette from the repo, not pypi.
The index is not updated ever unless you rerun the indexer, although I've started trying to ponder a filesystem watchdog here. Another possibility my be a Jupyter notebook content manager or post-save hook to update records as the notebook server saves them but this would not catch filesystem operations (dragging new notebooks to a folder, deleting notebooks etc?)
If the sqlite db is updated, I assume by some magic that the datasette server queries over the updated content?
Quick way to kill datasette processes: ps aux | grep datasette | grep -v grep | awk '{print $2}' | xargs kill