Re: [O] navigation broken - occur, org-babel-goto-named-src-block, etc fail

[Nicolas Richard]

Hello Charles,

"Charles C. Berry" <address@hidden> writes:
> On Sun, 15 Jun 2014, Nicolas Richard wrote:
>
>> Charles Berry <address@hidden> writes:
>>
>>> Start with emacs -q, then load today's org-mode.
>>
>> A glitch in 30220ff was quickly fixed in feca87b (three hours ago).
>> Maybe that's what you see ?
>>
>
> The ECM still fails.
>
> Did it work for you?

Unfortunately I didn't have the time to test, but it seemed very similar
to what I had experienced between these two commits, hence my suggestion
to look into them. I'll try later (and a few hours later...) M-a M-k Let me
try that now.

TL;DR: see near the end.

Ok, so here's what I think I have understood of the problem: In
org-fix-ellipsis-at-bol, we now do
(set-window-start (selected-window) (window-start))
which I thought would be a noop. But it is not, and to understand why
here's a simpler way of reproducing the funny result:

In some buffer with a few screenful of lines, when I do :
M-: (progn (forward-line 55) (set-window-start (selected-window) 
(window-start)))
(assuming point was at beginning of the buffer) I end up at line 22
(instead of 1+55 = 56) on my current frame/window configurations. (The
number 55 is chosen so that it is beyond the last line of the buffer.)

Now if I try this:
(progn (forward-line 55) (redisplay) (set-window-start (selected-window) 
(window-start)))
then it works perfectly fine.

So my conclusion is that in the first version, the command loop of emacs
doesn't redraw/recompute the window start position before the end of the
command, i.e. we have the following actions :

(progn
  ;; let's assume point is a position 1.
  (forward-line 55)
  ;; now point is at line 56, but (window-start) is still at 1.
  ;; if we used (redisplay) now, then (window-start) would be recomputed.
  (set-window-start
   (selected-window)
   (window-start)))
;; at this point, the command loop was instructed to make the window start at
;; position 1. But point is at line 56, which is out of the window,
;; hence emacs moves point to put it back in the window (namely, in its
;; center).

See (info "(elisp) Window Start and End") for more info on how
set-window-start works when point ends up out of the window.

TL;DR begins here: Hence we could either use the optional arg "NOFORCE"
or force (redisplay) before set-window-start. Both would probably the
bug you're seeing, but I'm not sure any of those is TRT in
org-fix-ellipsis-at-bol.

(BTW if you try to fix org-fix-ellipsis-at-bol in a running emacs, be
sure to re-eval the definition of its callers because it's a defsubst
and IIUC that gets inlined automatically.)

-- 
Nico.