Proposal: Fold back in client-pkg to Go submodule about client

[rhuss]

I'm ok with this if other are ok, too.

[dsimansk]

Per several talks with @cardil on the topic. I'm definitely +1 to move forward.

[lkingland]

This seems fine to me. ACKing because it may require a small update to Functions.

[rhuss]

I'm not generally opposed to the proposal (aside from the fact that I constantly had issues with git submodules in various toolings like IDEA, so I really try to avoid them, as they are also very confusing for Git newbies), but I think we should also revisit the original motivation.

The main reason for introducing client-pkg is to break cyclic dependencies between client and plugins on a repository level. Since Knative still sticks to a multi-repository approach backed by tons of automation (instead of a mono-repo), this repository separation makes it easy to understand the dependencies.

The main problem is this:

... but not removing the code in the client, we end up with duplicated code both here and in the client-pkg

IMO, it should be mandatory to remove the code in client when it has been moved to client-pkg. I know this will break older plugins, but it is also a good way to determine which plugins are still maintained/used. We can pin their dependencies to an older version of the client.

From my feeling, Git submodules make everything even more complex than it already is and even harder to understand for people new to the project or people coming back to the project after some time.

But if we decide to go this route, I would only accept this change with an extensive developer documentation update explaining how to set up a project like this.

@cardil, do you have some samples of other open-source projects that use a similar setup to keep cyclic dependencies on the repository level but logically break them with git submodules? Or do we invent something new here? I would be eager to learn from some past experiences with this usage of git submodules first.

[rhuss]

BTW, I'm not alone in my suspicion against git submodules. There are quite some strong opinions out there, too:

• https://news.ycombinator.com/item?id=31792303
• https://codingkilledthecat.wordpress.com/2012/04/28/why-your-company-shouldnt-use-git-submodules/
• https://diziet.dreamwidth.org/14666.html
• https://www.feoh.org/posts/git-submodules-are-awful-but-occasionally-necessary.html
• ....

[cardil]

@rhuss: BTW, I'm not alone in my suspicion against git submodules. There are quite some strong opinions out there, too: [..]

Roland, this has nothing to do with GIT submodules. It's about having multiple GO modules (and a GO workspace file) in one GIT repository.

[rhuss]

🤦
Yeah, I totally messed this up, shame on me .... Forget my last comment.

Still, my question would be if there has been some prior art so that we can learn from their experiences.

[cardil]

Kubernetes has recently adapted go workspace with multiple go modules: https://github.com/kubernetes/kubernetes/blob/d83cd48e5ebbb1b073164a574ef9aa5a68569d9c/go.work

[cardil]

Okay. I'll proceed with the PRs...