tidy-requires: module paths w/in require subforms should sort

[Bogdanp]

This suggestion conflicts with this racket-review rule which ensures that uses of require subforms sort according to their module path. Another example can be found here.

[jackfirth]

I agree with this, and others have mentioned the possibility to me too. But I'd also like to keep Resyntax and racket-mode in sync when it comes to require tidying. I'll reach out to Greg to discuss this.

[greghendershott]

This sounds good in principle. I'd need to reload my brain with details about multi-in; the other clauses seem straightforward. I'll comment back here later...

[greghendershott]

So actually. Having finished my first coffee. I realized that racket-tidy-requires already works the way you described in your email, as in your example:

(require a
         b
         c
         (only-in a/extra x))

becomes

(require a
         (only-in a/extra x)
         b
         c)

Was there maybe some other example where it does not? Maybe I have a bug in some cases. If so, I'll fix. But the intent is definitely to sort by module (within each meta level).

TL;DR: I think you could go ahead and change resyntax to sort that way, and if some bug turns up, then it's on me to fix that.

[jackfirth]

Ah, I was going off the description here rather than actually installing racket-mode and seeing what it does, since I'm not an emacs user. What does it do if the require subform doesn't contain exactly one collection in it? Like (prefix-in foo: (combine-in a b c)), for example.

[Bogdanp]

Review didn't support combine-in at all (because I never use it), so I just added it:

Bogdanp/racket-review@6ce349f#diff-dd9e24dda16c8ecc44babd298601cf79cbe4cd068711b7e6f8b8135691492e2a

Running tidy requires on that same file results in:

#lang racket/base
(require (prefix-in combined: (combine-in threading))
         (combine-in)
         racket/string
         (prefix-in combined: (combine-in threading racket/base)))

[greghendershott]

Ah, I was going off the description here rather than actually installing racket-mode and seeing what it does, since I'm not an emacs user.

Although that's intended as user docs, not a full spec, it is ambiguous; I can improve that.

What does it do if the require subform doesn't contain exactly one collection in it? Like (prefix-in foo: (combine-in a b c)), for example.

It's not even attempting to handle combine-in, yet.

I started a branch last week to refactor some of this (very old) code of mine, looking at the require grammar.

Among other things, I did add combine-in -- but I didn't consider the empty (combine-in) case. Thanks for pointing it out!

My instinct is to prune empty (combine-in) clauses. Although "tidy" shouldn't try to prune unused imports -- that's the job of a "stronger" command in Racket Mode, that runs the unused requires analysis -- pruning no-meaning, no-op "import nothing" clauses like (combine-in) is probably OK or in fact desirable. What do you think?

In any case it begs the question, if we're sorting by module, but (combine-in) has zero modules, where to sort it -- which I think was why you asked, @jackfirth.

EDIT: I saw the empty (combine-in) from @Bogdanp 's example, and overlooked your actual question, @jackfirth . I'm proposing to sort by the module of the first require-spec (if any) in a combine-in form. So a for (combine-in a b c).

---

Generally: It looks like I should re-read the grammar carefully, looking for any more edge cases or bad assumptions.

[jackfirth]

Perhaps the three of us ought to standardize on a shared implementation? A minimal-dependency package that we all reuse would get rid of any inconsistency problems.