Re: feature/package-vc has been merged

[Philip Kaludercic]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Philip Kaludercic <philipk@posteo.net>
>> Cc: monnier@iro.umontreal.ca,  rms@gnu.org,  emacs-devel@gnu.org
>> Date: Mon, 07 Nov 2022 08:42:18 +0000
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>> >> From: Philip Kaludercic <philipk@posteo.net>
>> >> Cc: monnier@iro.umontreal.ca,  rms@gnu.org,  emacs-devel@gnu.org
>> >> Date: Sun, 06 Nov 2022 18:35:10 +0000
>> >> 
>> >> Eli Zaretskii <eliz@gnu.org> writes:
>> >> 
>> >> > And if you think many do, why not clone the repository directly into
>> >> > ~/.emacs.d/elpa/?
>> >> 
>> >> Because that won't take care of scraping for autoloads, byte
>> >> compilation and installing missing dependencies.
>> >
>> > I don't see why.  Please elaborate how having the repository inside
>> > ~/.emacs.d gets in the way of these activities.
>> 
>> It doesn't get in the way, the issue just is that if you were to just
>> clone a package right into .emacs.d, you would still have to do all
>> these steps individually and manually.
>
> Which steps are those, and why do we have to do it manually?

Let us assume `default-directory' is (locate-user-emacs-file "elpa").
If I run "M-! git clone https://some.git.host.com/path/to/repo/foo.git";,
then I'll just have a directory called "foo", right?  If I want to byte
compile the files I'd e.g. have to open foo in Dired, mark all Emacs
Lisp files, byte compile them, then run something like
`make-directory-autoloads' myself.  Then I'd have to find the main file,
check the dependency list and run M-x package-install on every one that
is missing, again one-by-one.

You don't have to do this for `package-install', because it invokes
`package-unpack' that takes care of those details.  As
package-vc-install doesn't use prepared tarballs, the equivalent process
is a bit different (thus we have `package-vc-unpack'), but the intention
is the same.  Bundle all the repetitive task into a single command.

>> >> You mean as in only allowing for packages to distribute lisp code in the
>> >> root directory of the repository?  That would pointlessly break too many
>> >> packages that decide to structure their file hierarchy for whatever
>> >> reason.
>> >
>> > Is that what :lisp-dir is about? then the doc strings in package-vc.el
>> > doesn't even hint about that.  In particular, there's nothing there
>> > about the root directory of the repository.  
>> 
>> The docstring for `package-vc--archive-spec-alist' has the following
>> (I'm still looking for a better way to document this):
>> 
>>  `:lisp-dir' (string)
>>     The repository-relative name of the directory to use for loading the Lisp
>>     sources.  If not given, the value defaults to the root directory
>>     of the repository.
>
> Ah, it's a misunderstanding.  See below.
>
>> >                                              (Not that I understand
>> > why having Lisp files in a subdirectory of the repository would be a
>> > problem that needs an explicit configuration of the package, probably
>> > missing something else again.)
>> 
>> Maybe I am mistaken, but having a directory in `load-path' doesn't mean
>> all sub-directories are automatically indexed, right?
>
> We have a standard solution for that:
> normal-top-level-add-subdirs-to-load-path.  We install in any
> directory that needs this a file called subdirs.el with the following
> contents:
>
>   (if (fboundp 'normal-top-level-add-subdirs-to-load-path)
>       (normal-top-level-add-subdirs-to-load-path))
>
> Example of directories which need this is the site-lisp directory.

I was not familiar with this function.

> Why cannot we do something like this in this case?

I would have to try this out, but my worry is that in some cases this
could add too many non-lisp directories.

It still seems more elegant to encode what the lisp directory is in the
package description.