Consider a different CLI framework?

[jotaen]

The klog CLI is currently implemented with kong. Back when I started the project, I didn’t have a strong reason to choose kong over something else, and apart from a few minor shortcomings it turned out to work well.

I’m not generally opposed to migrating from kong to something else (e.g. Cobra), but I have to admit that it wouldn’t be super important to me at this point. The shortcomings are currently manageable, and my overall experience with kong has been very positive.

Advantages of kong

• It’s simple and lightweight to setup, e.g.
  • Declaring args/flags as struct’s and re-using them through regular composition is both straightforward to work with
  • Bootstrapping and wiring is <100 LOC in total. There is nothing kong-specific outside the main package.
  • The implementation of the actual sub commands is agnostic, so the commands themselves don’t know anything about the CLI framework. They are just a plain functions. (See the klog total command, for example.)
• The maintainer is quite responsive. I’ve opened multiple (smaller) issues in the past, and they helped out or fixed things quite quickly.
• It let’s you write custom type decoders, for validating and unmarshalling flag input to custom data structures.
• The CLI app as a whole can be end-to-end tested.

Shortcomings of kong

• No out-of-the-box support for shell completions. The kongplete library is missing a few things, so I had to fork it.
• It would be cool to have slightly more control over the help output. E.g. for some reason, kong ignores single line breaks in the help output. Flag descriptions also can’t be multi-line.

[jotaen]

What do you think about reimplementing the command-line interface with Cobra? I understand that forking kongplete is easier and a more immediate solution, but Cobra is already maintained and has more features than kongplete. Would you accept a switch to Cobra if it preserves the current behavior of the CLI? If you are okay with the idea, I will explore it and report what I learn.

(Posted by @chairmank, https://github.com/jotaen/klog/issues/177#issuecomment-1196670639)

I have to say that I’m currently rather hesitant to consider switching away from kong, also since it’s a “low-leveled” change that would require extensive testing. (Not just all the features, behaviour, and edge-cases, but also compatibility of different shells and operating systems.) The whole kongplete/completions topic is a bit unfortunate for sure, but it’s still a manageable shortcoming, and the rest of kong works pretty well overall.

On the other hand, it would be interesting to see how the setup would look with Cobra. So if you (@chairmank, or anyone else) are motivated to explore this, feel free to go ahead and create a spike/POC. But there still is the risk that I have to put this topic on hold, or suspend it altogether.