For those of you not subscribed for #science channel, a little trick for using Jupyter with locally built libraries and for distribution non-published buncles. The video is . The short recepie is the following: 1. Create a project in the idea and do what you want, exposing APIs that you want to use from the jupyter. 2. Use shadow plugin to create fat-jar distribution 3. Copy resulting jar from

build/libs

to the location of your notebook or send it to the consumer alongside the notebook 4. Load it using

@file:dependsOn("yourJar.jar")

5. OR create a module json file which references local jar file in the same way. In a module you can also define default imports and startup scripts if you want. 6. The resulting notebook + jar + [optional] module file form a stand-alone bundle that will work for anybody you send it to without the requirement to install environment or build anything.

if there is a version of maven's dependency plugin for gradle to dump all the deps to a lib/ directory you can avoid the problems with shade when two manifests conflict. the same basic simplicity applies for jvm java -cp $dist/lib/*

why not just using gradle publishToMavenLocal and then @file:Repository("*mavenLocal") @file:DependsOn("com.github.holgerbrandl:kravis:0.8-SNAPSHOT") Works great for me. However since I don't trust the ivy cache, I often clean it from within the notebook before loading my library

Because. You can't send your maven local to a friend. Also, as you said, you need to clean up ivy cache on each build. Otherwise, you will see the old jar with the same version stuck in the cache.

My friends would kill me if I start sending around fatjars. 🙂 I got used to inlining ivycleanup as preamble in jupyter. E.g. see https://nbviewer.jupyter.org/github/holgerbrandl/kalasim/blob/master/simulations/notebooks/simu-letsplot.ipynb

It still won't help you if you use kind of quick-fix bundled from unreleased versions.