Re: [PATCH] gnu: magit: Update to 2.2.0.

[Alex Kost]

Mark H Weaver (2015-08-22 20:31 +0300) wrote:

> Alex Kost <address@hidden> writes:
>
>> From 55e67d4f3867f11a77b9a91d4c3c54599846cc83 Mon Sep 17 00:00:00 2001
>> From: Alex Kost <address@hidden>
>> Date: Mon, 17 Aug 2015 21:44:40 +0300
>> Subject: [PATCH] gnu: magit: Update to 2.2.0.
>>
>> * gnu/packages/emacs.scm (magit): Update to 2.2.0.
>>   [inputs]: Remove 'git:gui'.
>>   [propagated-inputs]: Add 'dash', remove 'git-modes'.
>>   [arguments]: Enable tests.  Add VERSION, PREFIX and DASH_DIR to
>>   'make-flags'.  Remove 'configure', 'augment-load-path' and
>>   'post-install' phases.
>
> [...]
>
>> +       #:make-flags
>> +       (list (string-append "VERSION=" ,version)
>
> Interesting, you have to tell it what it's own version number is?

Yes, it tries to define the version using "git describe --tags ..." and
then it generates "magit-version.el" which is used only to set 'magit-version'
variable.

I think the main problem with this package (and many other packages,
sigh) is that there is no configure script, so there are some
workarounds to reinvent the wheel.

>>           (add-before
>>            'build 'patch-exec-paths
>>            (lambda* (#:key inputs #:allow-other-keys)
>> -            (let ((git (assoc-ref inputs "git"))
>> -                  (git:gui (assoc-ref inputs "git:gui")))
>> -              (emacs-substitute-variables "magit.el"
>> -                ("magit-git-executable" (string-append git "/bin/git"))
>> -                ("magit-gitk-executable" (string-append git:gui
>> -                                                        "/bin/gitk"))))))
>
> Does magit 2 no longer use gitk?

I don't see any sign of gitk (I don't understand why it was used before).

>> -         (add-before
>> -          'build 'augment-load-path
>> -          (lambda* (#:key inputs #:allow-other-keys)
>> -            ;; Allow git-commit-mode.el & co. to be found.
>> -            (let ((git-modes (assoc-ref inputs "git-modes")))
>> -              (setenv "EMACSLOADPATH"
>> -                      (string-append ":" git-modes 
>> "/share/emacs/site-lisp"))
>> -              #t)))
>> -         (add-after
>> -          'install 'post-install
>> -          (lambda* (#:key outputs #:allow-other-keys)
>> -            (emacs-generate-autoloads
>> -             ,name (string-append (assoc-ref outputs "out")
>> -                                  "/share/emacs/site-lisp/")))))))
>
> Are the autoloads installed by magit 2?

Yes.

>> +            (let ((git (assoc-ref inputs "git")))
>> +              (emacs-substitute-variables "lisp/magit-git.el"
>> +                ("magit-git-executable" (string-append git 
>> "/bin/git")))))))))
>
> Return #t from this phase procedure.

Ah, OK.  At first I blindly thought if 'emacs-substitute-variables'
succeeds, it returns #t, but now I see that it's not the case.  I see
that other emacs packages (except 'emacs-w3m') don't return #t after
this "substitute vars" phase.  So perhaps it would be better to modify
'emacs-batch-edit-file' (and 'emacs-batch-eval') from (guix build
emacs-utils) to return #t if everything was OK.  WDYT?

> Otherwise, looks good to me!  I haven't yet tried magit 2, but for now I
> will take your word for it that we don't need magit 1 anymore.  We can
> always add it later if there's demand for it.

OK, but I don't think it should be done, because it is obsolete and
there will be no Magit 1.4.3 or 1.5, only 2.X.X.

But switching from Magit 1.4 to 2 requires some… I don't know how to
say… attention, especially if you have many settings for the old Magit.
The new magit was mostly rewritten, so its settings differ a lot.

-- 
Alex