Eli Zaretskii <eliz@gnu.org> writes:

> can you put a printf after this line from xdisp.c:
>
>   else if (it->bidi_it.charpos == bob
> 	   || (!string_p
> 	       && (FETCH_CHAR (it->bidi_it.bytepos - 1) == '\n'
> 		   || FETCH_CHAR (it->bidi_it.bytepos) == '\n')))
>     {
>       /* If we are at the beginning of a line/string, we can produce
> 	 the next element right away.  */
>       bidi_paragraph_init (it->paragraph_embedding, &it->bidi_it, 1); <<<<<<
>
> and tell me once in how many characters or lines of input from `ls' is
> bidi_paragraph_init called here?  E.g., to see the character count,
> print it->bidi_it.charpos just before that line.

Here are the line numbers at which the printf is triggered (and how many
times consecutively they get triggered, in parentheses):

1    (x 5)
0    (x 16)
10   (x 2)
1    (x 3)
0    (x 16)
68   (x 4)
37
1606 (x 3)
0    (x 16)
125  (x 4)
94
93   (x 3)
0    (x 16)
183  (x 4)
152
151  (x 3)
0    (x 16)
241  (x 4)
210  (x 4)
0    (x 16)
299  (x 4)
268
267  (x 3)
0    (x 16)
358  (x 4)
327
326  (x 3)
0    (x 16)
417  (x 4)
386
385  (x 3)
0    (x 16)
476  (x 4)
445
444  (x 3)
0    (x 16)
533  (x 4)
502
501  (x 3)
0    (x 16)
590  (x 4)
559
558  (x 3)
0    (x 16)
...

and so on.  This is over one single ls call.  The 0's are when
it->bidi_it.charpos == 0, the rest are charpos's converted into buffer
line numbers.

The pattern seems to be 3 calls at each window end, then 1 on the second
line of the window, then 4 on the first window line.

After many `ls' calls, when redisplay has become sluggish, the pattern is
the same.

...
6788 (x 4)
6757
6756 (x 3)
0    (x 16)
6846 (x 4)
6815
6814 (x 3)
0    (x 16)
6897 (x 4)
6866
6865 (x 3)
0  (x 16)

The shell output for one ls call is attached, in case it is useful.