Experiment: node-sea

[arcanis]

Since the action item of the previous meeting was to build SEA experiments, I built mine: node-sea.

It generates JS files that, when required or executed from the CLI, either re-export a module or run a binary. It uses the Yarn libraries in order to generate the dataset that can be stored, and applied at runtime as a layer on top of the filesystem.

Note that you'll currently need to build the node binary from sources (nodejs/node#44537 is required), and some binaries such as Webpack will require nodejs/node#44732 for workers to be supported.

npx node-sea -c 'npm add prettier' -b prettier -m prettier -o prettier
~/node/node ./prettier.cjs --help

[jviotti]

Amazing stuff @arcanis ! I'm making a note to take a deeper look into it soon. How are you embedding the generated JS code into the binary?

[arcanis]

It's stored as a base64 string, and unpacked on the fly with the native Buffer.from(data, 'base64') method

[jviotti]

Also, let me convert this issue into a GitHub Discussion on the "Show and Tell" category!

[jviotti]

@arcanis How would you feel about sending a PR to this repo including a document (maybe inside an implementations/) folder that describes in more detail what you built, the decisions you make (i.e. what did you use as a VFS, how did you taught Node.js about it, how did you inject the data on the binary, etc), the reasoning behind them, etc. Some sort of architecture review document.

I tried to do this myself while going through the node-sea code, but I think you are much better positioned to do this given you wrote the code!

Once the document is in place, we should add it to a future meeting agenda to discuss the results of the experiment and see what conclusions and decisions we can draw out of it.