Buggy "gj" behavior with breakindent and showbreak

[Nazri Ramliy]

Here's how to reproduce it in an 80-column terminal:

$ vi --version|head -2
VIM - Vi IMproved 7.4 (2013 Aug 10, compiled Aug 27 2014 13:10:11)
Included patches: 1-417
$ echo $COLUMNS
80
$ vi -u NONE -U NONE -c 'exec "normal 6i\<tab>\<esc>31Aa\<esc>A a"' -c
'set breakindent showbreak=\> ' -c 'normal ^'

The cursor is now placed on the first "a" character on the first line,
preceeded with 6 tabs, and we can see that the last "a" character is
breakindent-ed with ">" showbreak character on the "virtual" line due
to the breakindent. So far so good.

Here's the bug: Pressing "gj" in normal mode should move the cursor to
the virtual line below but in this case the cursors jumps to the end
of the first line instead.

Nazri.

[Christian Brabandt]

Looks like a breakindent bug, but also isn't ;)

you can also trigger the bug in a 80 column terminal with something like
this:
:let &sbr=repeat('>', 49)

Here is a patch:
diff --git a/src/normal.c b/src/normal.c
--- a/src/normal.c
+++ b/src/normal.c
@@ -4515,7 +4515,8 @@ nv_screengo(oap, dir, dist)
* screenline or move two screenlines.
*/
validate_virtcol();
- if (curwin->w_virtcol > curwin->w_curswant
+ if ((curwin->w_virtcol - (*p_sbr != NUL ? vim_strsize(p_sbr) : 0))
+ > curwin->w_curswant
&& (curwin->w_curswant < (colnr_T)width1
? (curwin->w_curswant > (colnr_T)width1 / 2)
: ((curwin->w_curswant - width1) % width2


I have to admit, I do not understand the comment right above it:

,----
| * Check for landing on a character that got split at the end of the
| * last line. We want to advance a screenline, not end up in the same
| * screenline or move two screenlines.
| */
`----

I am not sure, what kind of characters can be split at the end of a line
(screen line I guess)?

BTW: I just noticed, there is a similar bug in gj/gk and listmode

In this window:

,----------------------------------------
| 1 aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa>
| --bbb
`----------------------------------------

And the cursor on the second a of line 1, pressing gj will move the
cursor the <Tab> at the last screen column in that window.

(To reproduce, in a 75 column terminal: vim -u NONE -N -c 'call
setline(1, repeat("a",73)."\tBBB")' -c ':set list listchars=tab:>-' -c
':norm! gj'

Ah, I just noticed this happens, because when listmode is off, the
cursor is always displayed at the end of a tab char, while when list
mode is on and there is a 'tab:' lcs-setting, the cursor will always be
drawn at the start of the tab. I didn't know that, so this should at
least be documented.

Best,
Christian
--
Man errät die Menschen am besten, wenn man sie bei Erzählungen um ihre
Vermutungen der unerzählten Zukunft fragt.
-- Jean Paul

[Christian Brabandt]

Better would be, to check, that the cursor is actually on a line, where
'sbr' is set, so here is an update:

diff --git a/src/normal.c b/src/normal.c
--- a/src/normal.c
+++ b/src/normal.c

@@ -4509,13 +4509,20 @@ nv_screengo(oap, dir, dist)
#if defined(FEAT_LINEBREAK) || defined(FEAT_MBYTE)
if (curwin->w_cursor.col > 0 && curwin->w_p_wrap)
{
+ colnr_T virtcol;
/*

* Check for landing on a character that got split at the end of the
* last line. We want to advance a screenline, not end up in the same
* screenline or move two screenlines.
*/

validate_virtcol();
- if (curwin->w_virtcol > curwin->w_curswant

+ virtcol = curwin->w_virtcol;
+
+ if (virtcol > (colnr_T)width1 &&
+ *p_sbr != NUL)
+ virtcol -= vim_strsize(p_sbr);
+
+ if (virtcol > curwin->w_curswant

&& (curwin->w_curswant < (colnr_T)width1
? (curwin->w_curswant > (colnr_T)width1 / 2)
: ((curwin->w_curswant - width1) % width2

Best,
Christian
--
Bisher hat der Glaube der Massen noch immer über die Vernunft der wenigen gesiegt.