Let me introduce the context before of my questions.

In order to support some new kind of hardware in AtomVM, a driver should be written first.

It basically means writing an Elixir/Erlang module for managing that device.
E.g. In order to support lora radio chip “sx126x” I did a module called Sx126XDriver, same thing for “bq25896” power management chip and so on.

Most of the times I end having 1 driver = 1 module.

Then I would like to make some packages, it would be nice having a package for each single driver, so when I build a new project I don’t have to pull drivers for devices I don’t have.

How can I avoid having 100 git repos for 100 packages?

What I was thinking about is:

a) When dependency is on git:

• mix.exs deps: use sparse option when package is on git, e.g.: sparse: “apps/sx126x_driver” (should I use apps as a container directory?)
• rebar3 deps: use git_subdir

b) Otherwise on hex: just use regular hex packages, use CI for publishing in a automated way packages to hex.

• Are you aware of any relevant issue with such a similar workflow?
• Do you have any suggestion for doing this in the best way?
• Do you have any suggestion for providing the best possible developer experience to package users?
• Should I prefix/suffix each package with some common prefix/suffix to indicate they are comming from the same megarepo? Such as in my case: avm_drivers_sx126x, avm_drivers_bq25896, etc…

Otherwise do you have any opinion against a similar pattern?

@Nerves folks: did you have any similar experience?

I’m looking forward opinions about this approach.