bug#63336: [PATCH] package-vc: Process :make and :shell-command spec args

[Philip Kaludercic <philipk@posteo.net>]

Joseph Turner <joseph@breatheoutbreathe.in> writes:

> Philip Kaludercic <philipk@posteo.net> writes:
>
>> Joseph Turner <joseph@breatheoutbreathe.in> writes:
>>
>>>>> +(defun package-vc--make (pkg-spec dir)
>>>>> +  "Process :make and :shell-command spec arguments."
>>>>> +  (let ((target (plist-get pkg-spec :make))
>>>>> +        (cmd (plist-get pkg-spec :shell-command)))
>>>>> +    (when (or cmd target)
>>>>> +      (with-current-buffer (get-buffer-create " *package-vc make*")
>>>>                                                  ^
>>>>                                                  should the package name
>>>>                                                  be mentioned here?
>>>
>>> I like this idea, but IIUC package-vc--make would then need to take an
>>> extra arg, since pkg-spec doesn't contain the :name of the package. We
>>> could also add :name to the pkg-spec plist?
>>
>> I wouldn't be in favour of that, I think that passing the name as a
>> separate argument would be a better solution.
>
> I agree.
>
>>> For comparison, package-vc--build-documentation creates a buffer called
>>> " *package-vc doc*" without the package name.
>>
>> The difference I see here is that documentation usually builds fine,
>> while :make or :shell-command have a higher chance of failing because
>> some software is missing, especially if people don't use :make the way
>> it is used on the ELPA server but to build external dependencies (I'm
>> thinking of mail clients like notmuch)
>
> That makes sense to me. In the attached patch, I pass pkg-desc to
> package-vc--make instead just name.
>
> Want me to submit a separate patch which adds the package name to the
> " *package-vc doc*" buffer name?

No, I don't think it is necessary.  But thanks.

>>>>> +                   target (buffer-name)))))))
>>>>> +
>>>>>  (declare-function org-export-to-file "ox" (backend file))
>>>>>
>>>>>  (defun package-vc--build-documentation (pkg-desc file)
>>>>> @@ -486,6 +515,9 @@ documentation and marking the package as installed."
>>>>>        ;; Generate package file
>>>>>        (package-vc--generate-description-file pkg-desc pkg-file)
>>>>>
>>>>> +      ;; Process :make and :shell-command arguments before building documentation
>>>>> +      (when package-vc-process-make (package-vc--make pkg-spec pkg-dir))
>>>>
>>>> Wasn't the plan to allow `package-vc-process-make' to either be a
>>>> generic "build-anything" or a selective listing of packages where we
>>>> allow :make and :shell-command to be executed?
>>>
>>> Let me know if the attached commit accomplishes what you had in mind.
>>
>> Yes, that (or rather the newest version from a different message) looks good.
>>
>>>>> +
>>>>>        ;; Detect a manual
>>>>>        (when (executable-find "install-info")
>>>>>          (dolist (doc-file (ensure-list (plist-get pkg-spec :doc)))
>>>>
>>>> Otherwise this looks good, but I haven't tried it out yet.
>>>
>>> I fixed up a couple other issues:
>>>
>>> - removed unnecessary dir arg to package-vc--make
>>> - added function arg to the docstring for package-vc--make
>>>
>>> I'm not sure if the customization type for package-vc-process-make is
>>> correct. Please double check that.
>>>
>>> Also, should users be able to run :make and :shell-command args defined
>>> in a spec passed into package-vc-install?
>>
>> Yes, is that currently not supported?
>
> Nevermind! It is supported. I didn't notice that package-vc--unpack adds
> the user-defined pkg-spec to package-vc-selected-packages just before
> calling package-vc--unpack-1.

1+

> Best,
>
> Joseph
>
> From b27724197acd4ee72f9d336843f0e6ed9fcee87b Mon Sep 17 00:00:00 2001
> From: Joseph Turner <joseph@breatheoutbreathe.in>
> Date: Sat, 13 May 2023 10:05:04 -0700
> Subject: [PATCH] package-vc: Process :make and :shell-command spec args
>
> ---
>  lisp/emacs-lisp/package-vc.el | 37 +++++++++++++++++++++++++++++++++++
>  1 file changed, 37 insertions(+)
>
> diff --git a/lisp/emacs-lisp/package-vc.el b/lisp/emacs-lisp/package-vc.el
> index beca0bd00e2..8529d1dad5c 100644
> --- a/lisp/emacs-lisp/package-vc.el
> +++ b/lisp/emacs-lisp/package-vc.el
> @@ -344,6 +344,33 @@ asynchronously."
>          "\n")
>         nil pkg-file nil 'silent))))
>  
> +(defcustom package-vc-process-make nil

Have we discussed the name of this user option?  I feel it is too
immediate, and therefore not intuitively obvious what purpose it serves.
I would imagine something along the lines of
"package-vc-allow-side-effects" or "package-vc-permit-building" could be
better?  WDYT?

> +  "Whether to process :make and :shell-command spec arguments.

I guess here too an explanation would be warranted (and in the manual).
Explaining what the issue is, and why one might be wary to enable the option.

> +When set to a list of symbols (packages), run commands for only
> +packages in the list. When `nil', never run commands. Otherwise
> +when non-`nil', run commands for any package with :make or
> +:shell-command specified.
> +
> +Package specs are loaded from trusted package archives."
> +  :type '(choice (const :tag "Run for all packages" t)
> +                 (repeat :tag "Run only for selected packages" (symbol :tag "Package name"))
> +                 (const :tag "Never run" nil))
> +  :version "30.1")
> +
> +(defun package-vc--make (pkg-spec pkg-desc)
> +  "Process :make and :shell-command PKG-SPEC arguments for PKG-DESC."
> +  (let ((target (plist-get pkg-spec :make))
> +        (cmd (plist-get pkg-spec :shell-command)))
> +    (when (or cmd target)
> +      (with-current-buffer (get-buffer-create

I'd format the buffer name in the top let to prevent this line-break here.

> +                            (format " *package-vc make %s*" (package-desc-name pkg-desc)))
> +        (erase-buffer)
> +        (when (and cmd (/= 0 (call-process shell-file-name nil t nil shell-command-switch cmd)))
> +          (warn "Failed to run %s, see buffer %S" cmd (buffer-name)))
> +        (when (and target (/= 0 (apply #'call-process "make" nil t nil (if (consp target) target (list target)))))
> +          (warn "Failed to make %s, see buffer %S" target (buffer-name)))))))

If :shell-command fails, do we really want to proceed to :make?

>  (declare-function org-export-to-file "ox" (backend file))
>  
>  (defun package-vc--build-documentation (pkg-desc file)
> @@ -486,6 +513,16 @@ documentation and marking the package as installed."
>        ;; Generate package file
>        (package-vc--generate-description-file pkg-desc pkg-file)
>  
> +      ;; Process :make and :shell-command arguments before building documentation
> +      (pcase package-vc-process-make
> +        ((pred consp) ; When non-`nil' list, check if package is on the list.
> +         (when (memq (package-desc-name pkg-desc) package-vc-process-make)
> +           (package-vc--make pkg-spec pkg-desc)))
> +        ('nil         ; When `nil', do nothing.
> +         nil)

Perhaps swap the two conditions, first checking nil then listp which I
think reads more natural.  Then again, is pcase actually serving
anything here?

> +        (_            ; When otherwise non-`nil', run commands.
> +         (package-vc--make pkg-spec pkg-desc)))
> +
>        ;; Detect a manual
>        (when (executable-find "install-info")
>          (dolist (doc-file (ensure-list (plist-get pkg-spec :doc)))