all-defined-out doesn't seem to work well with macro

[sorawee]

When running the following a.rkt, I get the expected result:

#lang racket

(provide module++)
(require syntax/parse/define)

(define-simple-macro (module++ name base xs ...)
  (module name base
    (provide (all-defined-out))
    xs ...))

(module++ bar racket/base
          (define a 2))

(require 'bar)
a ;=> 2

However, running b.rkt, which uses module++ in a.rkt, fails.

#lang racket

(require my/a)

(module++ foo racket/base
          1)

provide: identifier's binding is ambiguous
  context...:
  matching binding...:
  matching binding...:
  matching binding...: in: provide

[sorawee]

Possibly related to #2469?

[sorawee]

Hmm. Seems to be a problem with provide in general.

#lang racket

(provide module++2)

(define-syntax-rule (module++2 name base x xs ...)
  (module name base
    (provide x)
    xs ...))

(module++2 bar racket/base a
          (define a 2))

(require 'bar)

works fine, but

#lang racket

(require my/a)

(module++2 foo racket/base a
          (define a 1))

does not.

[bennn]

I think this is why provide (pre-)transformers exist:

https://docs.racket-lang.org/reference/stxtrans.html#(part._provide-trans)

[sorawee]

@bennn could you give me a concrete code? I tried

(define-simple-macro (module++ name base xs ...)
  #:with PROVIDE (pre-expand-export #'(all-defined-out) '())
  (module name base
    (provide PROVIDE)
    xs ...))

and it doesn't seem to work.

This does work, however.

(define-simple-macro (module++ name base xs ...)
  #:with PROVIDE (strip-context #'(provide (all-defined-out)))
  (module name base
    PROVIDE
    xs ...))

So my problem at hand is resolved. The question is, is the version in the original post expected to work? If so, this would be considered a bug.

[bennn]

My mistake! The real issue here is that a macro can't easily make a submodule that uses a non-#f language.

strip-context is the right thing to do for now, I think. You may need to strip the xs ... too.

@LeifAndersen discovered the same problem a few months ago (but hasn't written anything about it online). Leif's been talking with @michaelballantyne about solutions. They might have more to say.