bug#66114: [PATCH] Fix interactive prompt for selecting checkout directory

[Joseph Turner via "Bug reports for GNU Emacs, the Swiss army knife of text editors" <bug-gnu-emacs@gnu.org>]

Philip Kaludercic <philipk@posteo.net> writes:

> Joseph Turner <joseph@breatheoutbreathe.in> writes:
>
>> Tags: patch
>>
>> This patch fixes the interactive prompt in package-vc-checkout. See
>> commit message for details.
>>
>> From 70061d76542968575555f247d4715d18ed3c4e75 Mon Sep 17 00:00:00 2001
>> From: Joseph Turner <joseph@breatheoutbreathe.in>
>> Date: Tue, 19 Sep 2023 23:58:43 -0700
>> Subject: [PATCH] Fix interactive prompt for selecting checkout directory
>>
>> * lisp/emacs-lisp/package-vc.el (package-vc-checkout): Use
>> read-directory-name instead of read-file-name with
>> predicate.  Previously, it was impossible to interactively navigate to
>> a nested subdirectory.
>
> I can "navigate" into whatever directory I want, but I don't get completion.

You're right.

>> ---
>>  lisp/emacs-lisp/package-vc.el | 17 +++++++++++------
>>  1 file changed, 11 insertions(+), 6 deletions(-)
>>
>> diff --git a/lisp/emacs-lisp/package-vc.el b/lisp/emacs-lisp/package-vc.el
>> index c7a30736e32..29b540d86b8 100644
>> --- a/lisp/emacs-lisp/package-vc.el
>> +++ b/lisp/emacs-lisp/package-vc.el
>> @@ -823,12 +823,17 @@ repository.  If REV has the special value
>>  `:last-release' (interactively, the prefix argument), that stands
>>  for the last released version of the package."
>>    (interactive
>> -   (let* ((name (package-vc--read-package-name "Fetch package source: ")))
>> -     (list (cadr (assoc name package-archive-contents #'string=))
>> -           (read-file-name "Clone into new or empty directory: " nil nil t nil
>> -                           (lambda (dir) (or (not (file-exists-p dir))
>> -                                             (directory-empty-p dir))))
>> -           (and current-prefix-arg :last-release))))
>> +   (let* ((name (package-vc--read-package-name "Fetch package source: "))
>> +          (desc (cadr (assoc name package-archive-contents #'string=)))
>> +          (dir (read-directory-name "Clone into new or empty directory: ")))
>> +     (unless (or (not (file-exists-p dir))
>> +                 (directory-empty-p dir))
>> +       (let ((subdir (expand-file-name (package-desc-full-name desc) dir)))
>> +         (if (and (not (file-exists-p subdir))
>> +                  (y-or-n-p (format "Create new directory %s ?" subdir)))
>> +             (setf dir subdir)
>
> I'd prefer to use a setq here.
>
>> +           (user-error "Directory not empty: %S" (expand-file-name dir)))))
>
> This seems like an anti-feature to me, because you need to know what
> directory is empty before confirming your choice, and if it is not
> empty, you have to restart the entire command again.

Yes, I agree.

IIUC, we want to read from the user a directory which is either
nonexistent or empty.  However, we also want the completions to include
non-empty directories so that the user can easily select a deeply nested
directory.  The current solution does not offer non-empty directories for
completions, meaning that users must manually type in a potentially long
path to a nested empty or nonexistent directory.

Here's a different idea I tried:

(read-file-name "Clone into new or empty directory: " nil nil
                ;; Must match a nonexistent or empty directory
                (lambda (dir) (or (not (file-exists-p dir))
                                  (directory-empty-p dir)))
                nil
                ;; `read-directory-name' accepts no PREDICATE
                ;; argument: hack `read-file-name' instead.
                (lambda (file-name) (file-directory-p file-name)))

but this erroneously returns the default filename, which might be
buffer-file-name or whatever string is currently in the minibuffer.

I would have expected that when MUSTMATCH is a function and it returns
nil, a "[Match required]" message would appear.  However, the behavior of
read-file-name is unspecified when a MUSTMATCH function returns nil:

- a function, which will be called with the input as the
  argument.  If the function returns a non-nil value, the
  minibuffer is exited with that argument as the value.

Is this a bug, a case of under-documentation, or do I misunderstand something?

I'm happy to post this in a separate bug thread if you think it's worth it.

Joseph

>> +     (list desc dir (and current-prefix-arg :last-release))))
>>    (package-vc--archives-initialize)
>>    (let ((pkg-spec (or (package-vc--desc->spec pkg-desc)
>>                        (and-let* ((extras (package-desc-extras pkg-desc))