clean-ns: Introduce ^:keep metadata on libspecs

[plexus]

This allows marking up required namespaces that should not be cleaned by clean-ns.

This is useful to require namespaces only so they get loaded for side-effects. It can also be used in (ns user) to make functions available for the REPL.

Example

    (ns user
      (:require [com.stuartsierra.component :as component]
                ^:keep [reloaded.repl :refer [system init start stop go reset]]])

This PR is not ready yet, I'm just at the point where I got it to work, and I'd appreciate feedback.

Questions:

• Is ^:keep the best key to use? (bikeshedding time!)
• Right now it's only supported in :require, does it make sense to also have it for :import or :use. (Probably not since you don't usually :import for side effects, and :use is deprecated, but maybe there's an argument to be made).
• Should it also work on individuals function names in a :refer macro?
• Should it work on nested vectors when grouping by prefix?

Todo for me:

• Check if it also works on CLJS/CLJC
• Update CHANGELOG
• Update README

---

Before submitting a PR make sure the following things have been done (and denote this
by checking the relevant checkboxes):

• The commits are consistent with our contribution guidelines
• You've added tests (if possible) to cover your change(s)
• All tests are passing (run lein do clean, test)
• Code inlining with mranderson works and tests pass with inlined code (run ./build.sh install -- takes a long time)
• You've updated the changelog (if adding/changing user-visible functionality)
• You've updated the readme (if adding/changing user-visible functionality)

Thanks!

[plexus]

This code was too eager, it would transform

(ns user (:require ^:keep [,,,]))

into

(ns ^:keep user)

[expez]

Is ^:keep the best key to use? (bikeshedding time!)

I'd prefer refactor-nrepl/keep, that way it's possible for someone to find out what that metadata means even if they're not familiar with the tool.

Right now it's only supported in :require, does it make sense to also have it for :import or :use. (Probably not since you don't usually :import for side effects, and :use is deprecated, but maybe there's an argument to be made).

Agree with you on this. We can always revisit it later.

Should it also work on individuals function names in a :refer macro?

I don't think that's needed, when you mark it as keep I think it's OK to interpret that as "I'll manage this one manually". I might be biased because I almost never refer stuff.

Should it work on nested vectors when grouping by prefix?

We do prefix re-writing today so we need at least partial support.

(ns my-ns 
  (:require 
    [foo.bar :as bar]
    ^:refactor-nrepl/keep 
    [foo.baz :as baz]))

With prefix rewriting this would look like:

(ns my-ns 
  (:require 
    [foo
      [bar :as bar]
      ^:refactor-nrepl/keep 
      [baz :as baz]]))

But I don't think we need to worry about:

(ns my-ns 
  (:require 
    ^:refactor-nrepl/keep 
    [foo
      [bar :as bar]
      [baz :as baz]]))

If it's easier to just support both, then that's fine I think, but I wouldn't bother adding complexity to support this final variant.

[plexus]

I'd prefer :refactor-nrepl/keep, that way it's possible for someone to find out what that metadata means even if they're not familiar with the tool.

That makes sense. I was thinking it could be a convention that other tools can also pick up, but that's probably not too likely anyway.

[expez]

Ping @plexus :)

[plexus]

Hey, sorry for letting this sit. I've been completely swamped with other stuff, and given that it's still a bit of work to do this feature properly I won't get to it very soon.

[bbatsov]

I'm closing this PR due to no recent activity. Feel free to re-open it if you ever come back to it.

[aiba]

I would find this really useful. A common challenge for me with clean-ns is when I require a clojure namespace that contains a defrecord, and then also import that record. As in:

(ns foo.core
  (:require [bar.core])
  (:import [bar.core.SomeRecord]))

The require is needed for the runtime generation of SomeRecord, but gets removed by clean-ns.

[slipset]

Here is the solution that was chosen for clj-kondo, which I think is a very wise choice.

Any :require that doesn't either use :refer or :as is assumed to be there for sideeffecting reasons and thus cannot be removed.