Submodules cause false positive cyclic dependencies errors

Given two files with the following contents:

A.rkt:

(module+ foo
  (require "B.rkt"))

B.rkt:

(require "A.rkt")

On Racket 6.2.1 I get the following error when running racket A.rkt:

standard-module-name-resolver: cycle in loading
  at path: .../A.rkt
  paths:
   .../A.rkt
   .../B.rkt
  context...:
   standard-module-name-resolver
   standard-module-name-resolver
   standard-module-name-resolver

The module loader complains of a module dependency cycle, but because of the semantics of submodules there really isn't one. Specifically, B depends on A and A-foo depends on B and A. Therefore it should be possible to load first A and not A-foo, then B, then A-foo. This is not the case.

I've run into this problem several times in circumstances where A is a module in the core of a library I'm writing, B is a module defining some extra functionality in the library, and A-foo is a module testing A using the functionality in B to make the tests easier to write and avoid duplication.

As best as I can understand it, this shouldn't be an error.

Start by reproducing the A.rkt and B.rkt example with racket A.rkt, then inspect the standard-module-name-resolver entry point named in the error. Trace how the loader handles A, B, and the foo submodule. Done means this dependency arrangement loads without reporting a false cycle while preserving genuine cycle detection.