(run-at-time 0.0 nil (lambda () (goto-char (point-min))))

--8<---------------cut here---------------start------------->8---
  (defun my/diff-refine-all-hunks ()
    (interactive)
    (condition-case nil
        (save-excursion
          (goto-char (point-min))
          (while (not (eobp))
            (diff-hunk-next)
            (diff-refine-hunk)))
      (error nil))
    (run-at-time 0.0 nil (lambda () (goto-char (point-min)))))
--8<---------------cut here---------------end--------------->8---

Yep!  This does work perfectly: all hunks refined, and pointer at beginning of
buffer...

Thanks a lot for your more than precious help!

Best regards,
  Seb

--
Sebastien Vauban

BTW, highlighting within the lines is helpful in most of the cases, but
sometimes when the diff is very convoluted it can get in the way. To remedy
this I added a key (f7) which when pressed in the diff buffer toggles
the in-line highlight on and off.

I put it here in case someone else also finds it useful:

(add-hook 'diff-mode-hook 'my-diff-stuff)

(setq my-diff-refine-change-color "yellow1")

(defun my-toggle-diff-refine-change-color()
  (interactive)
  (if (face-background 'diff-refine-change)
      (set-face-background 'diff-refine-change nil)
    (set-face-background 'diff-refine-change my-diff-refine-change-color)))

(defun my-diff-stuff ()
  (set-face-background 'diff-refine-change my-diff-refine-change-color)
  (local-set-key (kbd "<f7>") 'my-toggle-diff-refine-change-color)

  (unless (or (eq this-command 'dvc-diff)
              (eq this-command 'dvc-generic-refresh)
              (> (buffer-size) 20000))
    (my-refine-all-diff-hunks)))

        Stefan

What I did not get in the new diff-mode implementation is why it
uses run-at-time instead of run-with-idle-timer?

run-at-time is for specifying a timer which runs at a specific time
while run-with-idle-timer runs it when emacs becomes idle. Isn't
using run-with-idle-timer is more appropriate (makes the code clearer)
in this case?

And why the code uses 0.0 instead of just 0? Is there a difference?

Let's assume I have received 5 mails:

--8<---------------cut here---------------start------------->8---
*   18 Sun 14:25   0.2k Foo  Mail #1 Hello
    18 Sun 14:25   0.4M Bar  Mail #2 How are you?
    18 Sun 17:58   6.9k Baz  Mail #3 [diff] Fix bug
    18 Sun 18:06   4.1k Baz  Mail #4 [diff] Fix doc
    18 Sun 18:27   3.1k Baz  Mail #5 I've done it
--8<---------------cut here---------------end--------------->8---

* = position of the cursor

... and I want first to read the 3rd one. I position the cursor on that one...

--8<---------------cut here---------------start------------->8---
    18 Sun 14:25   0.2k Foo  Mail #1 Hello
    18 Sun 14:25   0.4M Bar  Mail #2 How are you?
*   18 Sun 17:58   6.9k Baz  Mail #3 [diff] Fix bug
    18 Sun 18:06   4.1k Baz  Mail #4 [diff] Fix doc
    18 Sun 18:27   3.1k Baz  Mail #5 I've done it
--8<---------------cut here---------------end--------------->8---

... press RET, get a new buffer created on the mid-right pane, and read it.

--8<---------------cut here---------------start------------->8---
*   18 Sun 14:25   0.2k Foo  Mail #1 Hello           |
    18 Sun 14:25   0.4M Bar  Mail #2 How are you?    |
    18 Sun 17:58   6.9k Baz  Mail #3 [diff] Fix bug  | (contents of 3rd email)
    18 Sun 18:06   4.1k Baz  Mail #4 [diff] Fix doc  |
    18 Sun 18:27   3.1k Baz  Mail #5 I've done it    |
--8<---------------cut here---------------end--------------->8---

Then, as I always do, I simply press "n" to read the "next one" (4th)... but
Gnus displays me the 1st one!

Why?  Because the cursor has been positioned on the first line of my *summary*
buffer (on the left), after displaying the contents of the "Fix bug" email (on
the right).

Why?  Because diff-mode is called (as that email contains a diff), and some
some effect -- which I don't understand -- is that *the cursor is repositioned
onto the first line of the summary buffer* (while the code is supposed to be
executed in the other window, the one with the *contents* of the mail).

So, here, to reproduce the problem, I have to:

- have emails with diffs in them (otherwise, diff-mode won't be called)

- skip reading the first email (so that, clicking "n" does open the "next"
  unread email... from the current cursor position, which is wrong here...)

I know describing a dynamic behavior is not obvious per email. So, did you
understand the description of the problem, at least?

Do you understand why this is occurring, or have some helpful ideas?

Best regards,
  Seb

--
Sebastien Vauban

        Stefan

(run-at-time 0.0 nil (lambda () (goto-char (point-min)))))

causes the problem. It is executed after the current command is
finished and I guess Gnus puts back the cursor to the summary
buffer, so this delayed (goto-char (point-min)) is executed
there.

An obvious fix is to execute the delayed goto-char only if
the cursor is in a diff buffer after the command, that is
don't do it in the Gnus summary buffer:

(run-at-time 0.0 nil (lambda ()
                       (if (eq major-mode 'diff-mode)
                           (goto-char (point-min)))))