Suggested workflow for handling c src dependencies in packages

General Usage
1

I have a private package (lets call it MyPkg.jl) that requires wrapping a very minimal c library. This c library provides a small subset of the package’s total capabilities. I have been playing w/ BinaryBuilder and I’ve been able to deploy a _jll package locally. However, this seems overkill for my needs. If I understand the BinaryBuilder docs correct, it seems like I would need to maintain a URL for the c src and a _jll repo in addition MyPkg.jl itself. Also, it is not clear of how these all interact. E.g. it seems to assume some upstream CI for building. I would prefer to keep all non-general repository dependencies self-contained in the MyPkg.jl repository.

Is there a recommended workflow for shipping the c source w/ the julia package and automatically building for the specific system?

If using the BinaryBuilder _jll is recommended, is there a way to deploy to a non GitHub repo, e.g. a private GitLab?

2

Is GitHub - JuliaPackaging/Yggdrasil: Collection of builder repositories for BinaryBuilder.jl not an option?

3

No, the source code is private

4

You can just keep the source with the package itself. When you add the package, Julia calls the deps/build.jl script, which can just call your makefile. You have to actually be able to build the source on the machines where you add the package, of course.

5

Would it make any sense to use deps/build.jl to call a BinraryBuilder build_tarballs.jl targeting the host platform with something like platform = [parse(Platform, Base.BinaryPlatforms.host_triplet())] so to not require build tools explicitly on the host machine?

1 Like