"Remove Unused Requires" refactoring doesn't handle multi-in forms correctly

[dfeltey]

Running the "Remove Unused Requires" refactoring on this program:

#lang racket/base

(require racket/require)
(require (multi-in racket (list dict)))
first

produces:

#lang racket/base

(require racket/require)
(require)
first

Rather than just removing the dict part of the multi-in form.

[greghendershott]

Related: Dr Racket will color the entire (multi-in racket (list dict)) form red as "unused" -- even though list is used. In an ideal universe only the dict portion would correctly be marked red as unused.

I actually first noticed this yesterday while consuming syncheck:add-unused-require items from drracket/check-syntax for Racket Mode, then went to DrR to cross-check its behavior.

---

I wonder if it would be pragmatic and reasonable to change multi-in to expand to one (#%require) form per module? That's what (require racket/list racket/dict) would expand to, I think?

One the one hand drracket/check-syntax should ideally handle "everything". On the other hand if multi-in were from a third-party package, and its author learned there is a de facto, soft, unwritten protocol like this for tooling? They would probably adapt multi-in to use it. At least as a mitigation until/unless the tooling improves and/or the protocol becomes formal.

[greghendershott]

It looks like multi-in currently expands to combine-in.

Using combine-in directly in the original source, also exhibits this bug.

(combine-in racket/list racket/dict) results in one big $%require form like this:

(#%require
 (just-meta 0 (rename racket/list filter-not filter-not))
 (just-meta 0 (rename racket/list list-prefix? list-prefix?))
 ...
 (just-meta 0 (rename racket/dict dict-set dict-set))
 (just-meta 0 (rename racket/dict dict-value-contract dict-value-contract))
 ... )

My guess is that any unused module path, like racket/dict in this case, will cause drracket/check-syntax to say that the entire #%require is unused.

I bet it would not happen with combine-in expanding instead to:

(#%require
 (just-meta 0 (rename racket/list filter-not filter-not))
 (just-meta 0 (rename racket/list list-prefix? list-prefix?))
 ...)

(#%require
 (just-meta 0 (rename racket/dict dict-set dict-set))
 (just-meta 0 (rename racket/dict dict-value-contract dict-value-contract))
 ... )

[greghendershott]

Or maybe combine-in should be using syntax/loc on the pieces?

[greghendershott]

I opened PR #362 with an effort to fix this within drracket/check-syntax.