Re: emacs-27 30a7ee5: Fix Arabic shaping when eww/shr fill the text to be rendered

Re: emacs-27 30a7ee5: Fix Arabic shaping when eww/shr fill the text to be rendered

[Stefan Monnier]

>     Fix Arabic shaping when eww/shr fill the text to be rendered
>     * src/hbfont.c (hbfont_shape): Don't use DIRECTION if the current
>     buffer has bidi reordering disabled.  (Bug#41005)
[...]
> -  if (!NILP (direction))
> +  if (!NILP (direction)
> +      /* If they bind bidi-display-reordering to nil, the DIRECTION
> +	 they provide is meaningless, and we should let HarfBuzz guess
> +	 the real direction.  */
> +      && !NILP (BVAR (current_buffer, bidi_display_reordering)))

Hmm... so indeed I see that SHR binds `bidi-display-reordering`.
Isn't that a problem, since its docstring says:

    [...]
    Setting this to nil is intended for use in debugging the display code.
    [...]

Can we change SHR so it doesn't bind this var?


        Stefan

Re: emacs-27 30a7ee5: Fix Arabic shaping when eww/shr fill the text to be rendered

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Cc: emacs-devel@gnu.org
> Date: Fri, 05 Jun 2020 09:58:57 -0400
> 
> >     Fix Arabic shaping when eww/shr fill the text to be rendered
> >     * src/hbfont.c (hbfont_shape): Don't use DIRECTION if the current
> >     buffer has bidi reordering disabled.  (Bug#41005)
> [...]
> > -  if (!NILP (direction))
> > +  if (!NILP (direction)
> > +      /* If they bind bidi-display-reordering to nil, the DIRECTION
> > +	 they provide is meaningless, and we should let HarfBuzz guess
> > +	 the real direction.  */
> > +      && !NILP (BVAR (current_buffer, bidi_display_reordering)))
> 
> Hmm... so indeed I see that SHR binds `bidi-display-reordering`.

Yes, it does.  But the problem is more general, so the fix is not just
a kludge to fix what shr does.  As the comment explains.

> Isn't that a problem, since its docstring says:
> 
>     [...]
>     Setting this to nil is intended for use in debugging the display code.
>     [...]

It's "do like I say, not like I do" ;-)

> Can we change SHR so it doesn't bind this var?

I didn't see any better way at the time.  Shr's naïve filling
algorithm cannot cope with the complexity of RTL text in an LTR
paragraph, when a long line has been continued.  Maybe it's even
entirely impossible to DTRT in Lisp in these cases.  But if you or
someone else have ideas, please describe them.

Re: emacs-27 30a7ee5: Fix Arabic shaping when eww/shr fill the text to be rendered

[Stefan Monnier]

> Yes, it does.  But the problem is more general, so the fix is not just
> a kludge to fix what shr does.  As the comment explains.

Yes, the fix is fine.

>> Isn't that a problem, since its docstring says:
>> 
>>     [...]
>>     Setting this to nil is intended for use in debugging the display code.
>>     [...]
>
> It's "do like I say, not like I do" ;-)

;-)

>> Can we change SHR so it doesn't bind this var?
>
> I didn't see any better way at the time.  Shr's naïve filling
> algorithm cannot cope with the complexity of RTL text in an LTR
> paragraph, when a long line has been continued.  Maybe it's even
> entirely impossible to DTRT in Lisp in these cases.  But if you or
> someone else have ideas, please describe them.

Fair enough, thanks,


        Stefan