>
> The problems with C-n/C-p were unrelated, and seem to be a very old bug. 
> I've now mostly fixed that on master (and hopefully didn't introduce any 
> regressions while at that).
>

Thanks!  Indeed these problem seem fixed now, it's much better.  I did not 
see any regressions, if I see anything I'll tell you.  One case that still 
isn't quite right apparently is the arabic-small.txt.json file.  If I open 
that file with emacs -Q and type C-f, or C-n, or C-p, point moves to EOB 
(but strangly, not in 100% cases), and there C-a, or M-<, or C-p, or C-n 
do not seem to work anymore.

>
> In general, layout calculations when we have very long stretches of R2L 
> text in a left-to-right paragraph are very tricky, because many 
> assumptions become false.
>

Frankly, would programming be as fun as it is if assumptions were true? 😉