Be consistent with command line parsing by standardizing on clap or argh

[stmcginnis]

What I'd like:

Most of our CLIs use the argh crate for parsing command line input. But some are using clap instead.

It would be good to convert these over to argh so we are consistent across our code and avoid importing two different crates that provide the same functionality.

Any alternatives you've considered:

Could keep them as is since there is no conflict between using the mix of the two, but not a real reason not to just use one or the other.

[alok-sin]

I would like to take this. Does this concern with the testsys repo as well? bottlerocket-os/bottlerocket-test-system#535

[stmcginnis]

Awesome, that would be great @alok-sin!

I'm not 100% on the bottlerocket-test-system changes yet. There has been some talk of removing some of that code, so I don't want to suggest you make changes there only to have it all be removed.

If you could start in this repo that would be ideal. Then we can hopefully have a little better understanding if anything should be updated in the other repo once that is done.

Thanks a lot!

[webern]

How did you arrive at standardizing on argh instead of clap? I think clap is becoming the de-facto standard crate for this.

[stmcginnis]

@bcressey had expressed a desire to go with argh over clap. If clap is becoming the standard for this, then perhaps we should reevaluate that position. I am fine either way - I would just like to see us using one or the other.

[bcressey]

I don't feel as strongly about "only one" but I definitely don't want to add clap to the binaries shipped in the OS image, and would prefer that the mix of custom / structopt / argh that is currently shipped get trimmed down to just argh.

That's primarily for reasons of code size. It's not a huge factor for existing builds, but there are some use cases (like a microVM guest) where size is a concern. I don't like the idea of spending 10+ MiB of space for arg parsing that no one will ever see.

[webern]

I don't feel as strongly about "only one"

I also don't feel strongly about "only one" and would hate to see us go from clap -> argh in places where we are using clap now.

Have you checked the size of the compiled code of clap vs argh? Just curious if it's a big contributor to binary size.

[stmcginnis]

structopt is deprecated, so we should review what we're using and migrate to whichever makes the most sense.

[jpculp]

I looked into this last time it was brought up and while argh is smaller it is also more limited in terms of features and flexibility, requiring some refactor. I agree we should get off structopt, but I think we'd have an easier time going the clap route for things outside of the sources workspace.

[stmcginnis]

Sorry, I noticed an older, closed issue and started poking around. I ended up submitting a couple PRs to address this before I realized we had this open. I'm going to assign back to myself and close this out if/when we merge #3228 and #3229.

[stmcginnis]

Marking complete as #3228 and #3229 have merged and should address things.