bug#50646: 28.0.50; narrow-to-defun sometimes narrows to wrong defun

[Phil Sainty]

On 2021-09-18 05:06, arthur.miller@live.com wrote:
> 1. run Emacs -Q -l /path/to/attached/help-mode.el
> 2. type C-x f

C-h f

> 3. in minibuffer type: when RET
>
> The help-mode buffer that opens should show help for 'when' form, but 
> it
> shows the source code for the 'pop' macro which precedes the 'when' in
> subr.el.

> The docs for 'narrow-to-defun' says the "current-defun" is one that
> contains the point or follows the point.

I can reproduce the issue using your code, but I'm not sure whether the
reason for it is a misunderstanding of what 'point' is, some vagueness
in the 'narrow-to-defun' docstring, or an actual regression.

To eliminate the first:

   "Like other positions, point designates a place between two 
characters
(or before the first character, or after the last character), rather
than a particular character.  Usually terminals display the cursor over
the character that immediately follows point; point is actually before
the character on which the cursor sits."

-- quoted from (info "(elisp)Point")


'narrow-to-defun' says "The current defun is the one that contains point
or follows point." and if you place the cursor over the opening "(" of
"(defmacro when" then that defmacro form "follows point", and indeed
narrow-to-defun works as intended.

The issue is what happens when point is somewhere earlier than that
position, but still *after* the previous form.  In that instance
narrow-to-defun narrows to the previous form, which is surely what is
happening in your code.

If so, all you need for reproducing this is to experiment with
narrow-to-defun at the various positions between two forms.

That is a consequence of 'beginning-of-defun' jumping to the beginning
of the previous defun in that situation (which to me seems like a
reasonable behaviour for that function); but for 'narrow-to-defun'
the effect does seem contrary to its docstring.

I think we just want to change the docstring of 'narrow-to-defun' --
the last relevant commit looks like:


commit 050cc68b402f5998193a6026d0eeeecb9d2cb9c4
Author: Lennart Borgman <lennart.borgman@gmail.com>
Date:   Wed Apr 11 04:12:20 2012 +0200

    `narrow-to-defun' fixup

    * emacs-lisp/lisp.el (narrow-to-defun): `beginning-of-defun' goes
    to previous function when point is on the first character of a
    function. Take care of that in `narrow-to-defun'.

    Fixes: debbugs:6157

https://debbugs.gnu.org/cgi/bugreport.cgi?bug=6157


-Phil