bug#25105: [Emacs-diffs] master 2c8a7e5: Improve diff-mode navigation/manipulation

bug#25105: [Emacs-diffs] master 2c8a7e5: Improve diff-mode navigation/manipulation

[Dima Kogan]

[-- Attachment #1: Type: text/plain, Size: 1722 bytes --]

Mark Oteiza <mvoteiza@udel.edu> writes:

> Fixing C-c C-a to DTRT is great, thanks, but I don't think the
> off-by-one navigation changes to "n" and "p" (diff-hunk-next,
> diff-hunk-prev) make sense. While it may have made fixing the issues
> mentioned in the commit message easier, the changes to what "n" and "p"
> do at the beginning and end of a diff are not documented, and I didn't
> see any discussion of it in the associated bug.
>
> I contend that the new behavior is inconsistent with the behavior of
> other outline/thing-with-headers type things in Emacs.  outline-mode,
> org-mode, and rst-mode are the first ones that come to mind.

Can you give a specific example of interaction in any of these modes
that is analogous to the off-by-one behavior you're referring to? The
fundamental question is what hunk diff-mode should think the point is
looking at, when it is in some non-diff message above the first hunk.
The answer I chose for the new navigation logic is "first hunk". You
could also choose "invalid hunk, not a hunk at all", which would imply
that C-c C-a and M-ret and such shouldn't work there. Better suggestions
welcome.


> It's also not clear how the introduced oddity with auto-refine is going
> to be resolved, unless a way is found to autorefine the first hunk
> without there being any user interaction.  Then opening a diff has
> inconsistent auto-refining from the start.

I don't use auto-refinement, so didn't notice the breakage. Will look at
it in a bit.

In the meantime, I'm attaching a patch to address the 2nd point in the
bug report (25105). This serves to treat the junk before the first hunk
(i.e. commit message from 'git format-patch') as a file header. Looks
reasonable?



[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: diff-mode-handle-initial-junk.patch --]
[-- Type: text/x-diff, Size: 544 bytes --]

diff --git a/lisp/vc/diff-mode.el b/lisp/vc/diff-mode.el
index 5b48c8d..41476bd 100644
--- a/lisp/vc/diff-mode.el
+++ b/lisp/vc/diff-mode.el
@@ -768,7 +768,7 @@ diff-beginning-of-file-and-junk
         (setq prevfile nextfile))
     (if (and previndex (numberp prevfile) (< previndex prevfile))
         (setq prevfile previndex))
-    (if (and (numberp prevfile) (<= prevfile start))
+    (if (numberp prevfile)
           (progn
             (goto-char prevfile)
             ;; Now skip backward over the leading junk we may have before the

bug#25105: [Emacs-diffs] master 2c8a7e5: Improve diff-mode navigation/manipulation

[Mark Oteiza]

On 06/12/16 at 11:29pm, Dima Kogan wrote:
> Mark Oteiza <mvoteiza@udel.edu> writes:
> 
> > Fixing C-c C-a to DTRT is great, thanks, but I don't think the
> > off-by-one navigation changes to "n" and "p" (diff-hunk-next,
> > diff-hunk-prev) make sense. While it may have made fixing the issues
> > mentioned in the commit message easier, the changes to what "n" and "p"
> > do at the beginning and end of a diff are not documented, and I didn't
> > see any discussion of it in the associated bug.
> >
> > I contend that the new behavior is inconsistent with the behavior of
> > other outline/thing-with-headers type things in Emacs.  outline-mode,
> > org-mode, and rst-mode are the first ones that come to mind.
> 
> Can you give a specific example of interaction in any of these modes
> that is analogous to the off-by-one behavior you're referring to?

I wrote about how your changes are make diff-mode _not_ analogous.

> The
> fundamental question is what hunk diff-mode should think the point is
> looking at, when it is in some non-diff message above the first hunk.
> The answer I chose for the new navigation logic is "first hunk". You
> could also choose "invalid hunk, not a hunk at all", which would imply
> that C-c C-a and M-ret and such shouldn't work there. Better suggestions
> welcome.

One might argue that C-c C-a and friends in a file header should apply
all hunks in a file, or perhaps that there should actually be
diff-apply-file commands, etc.

The way n and p worked was not a bug, yet you gratuitously changed them,
and broke auto-refinement.  Why do I have to now hit two keys to refine
the first hunk, and one key to refine the second?

> > It's also not clear how the introduced oddity with auto-refine is going
> > to be resolved, unless a way is found to autorefine the first hunk
> > without there being any user interaction.  Then opening a diff has
> > inconsistent auto-refining from the start.
> 
> I don't use auto-refinement, so didn't notice the breakage. Will look at
> it in a bit.

It's on by default, so this statement perplexes me.