Adding packages or a custom filesystem to the jupyterlite kernel for a self-contained distribution

[psychemedia]

I'm exploring the possibility of using JupyeterLite / jupyterlite-webr-kernel in a secure (prison) environment. This requires having a self-contained jupyterlite environment running on a secured server.

Loading packages into webR jupyterlite notebooks seems to involve downloading the wasm compiled packages from a public repository on demand. Is there a way of bundling the packages into a custom built JupyterLite webr kernel?

[psychemedia]

Related to this, I wonder: is there a way of mounting a custom filesystem built with the https://github.com/r-wasm/actions/tree/main/.github/workflows#r-wasmactions-reusable-workflows Github Action with the JupyterLite kernel?

I would like to be able to ship a JupyterLite environment with custom packages bundled into it.

[georgestagg]

Yes, in principle this should be possible. The idea would be to build an R repository of the packages you'd like using {rwasm} and then including this in your jupyterlite static output, so that the packages are served in the same way as jupyterlite. Probably the Wasm package repository can just be added to jupyterlite's files in a subdirectory.

You can serve either a package repository or a filesystem image, but I'd personally recommend a repo.

Then, once you have started your kernel under jupyterlite and are loading packages you can set the repos= argument to point to your repository relative to the served jupyterlite directory:

install.packages("cli", repos = "./repo")

or, mount your filesystem image and setup .libPaths() to load packages from the mount location:

webr::mount(
  mountpoint = "/packages",
  source = "./some/path/to/image/packages.data"
)
.libPaths(c("/packages", .libPaths()))

---

Ideally, we'd be able to set the default webR package repository from the jupyterlite configuration, avoiding the need for the repos = argument. This is not possible currently, but something that would be good to have. I'll leave this issue open for tracking that.

[psychemedia]

Thanks — will give that a try...

[georgestagg]

Its also possible to mount each package one-by-one, if you have difficulty building a combined packages.data file using the rwasm package. Perhaps it would be best to attempt that first - then move onto a combined VFS image once that is confirmed working.

See my comment on r-wasm/webr#465 (comment) for more info on that. The comment is targeting running webR on iOS, but the same idea should apply with Jupyterlite: serving the package data in some subdirectory, then using webr:mount() to make it availabe on the virtual filesystem.

[psychemedia]

I can get the packages to load from a relative path (not sure if there is a jupyterlite settling that also needs fettling?) but if I put the full repo url (eg repos = "https://ouseful-demos.github.io/jupyterlite-m348-demo/repo/") then that seems to work.

(In the above case I have gone "out of path"; I think in that build there is currently also a copy of the files at https://ouseful-demos.github.io/jupyterlite-m348-demo/notebooks/lab/repo/ to test the "./repo" path in lab; (that didn't seem to work; neither did putting the repo into ../files)

I wonder, can webR introspect and return the jupyterlite installation URL?

I also note there are still a few things arriving that are not bundled (my requirement is to be completely self-contained):

The requested files seem to be on the path [webr.r-wasm.org](https://webr.r-wasm.org/v0.3.0). E..g.:

• https://webr.r-wasm.org/v0.3.0/R.bin.wasm
• https://webr.r-wasm.org/v0.3.0/libRlapack.so
• etc

Checking my own distribution, the path webr.r-wasm.org/v0.3.0 seems to be baked into:

43.fadaec40722488dbf52a.js
libRlapack.so
R.bin.data

I note that WebR.WebROptions allows the setting of baseurl (for the webr distro? i.e. the home for the files I am seeing above?) and repourl (for the package repo?) so I guess I need to either copy those files and serve them myself or build my own webR distro? The webR build/deploy action seems to set the base URL here:

https://github.com/r-wasm/webr/blob/d455321cd377fab530ced70ad81279c24751d570/.github/workflows/deploy.yml#L41

But can jupyterlite server be configured to point to that? Or do I need a custom JupyterLite build?