Re: ZWJ/ZWNJ in Pango for Devanagari/Arabic

[Owen Taylor <otaylor redhat com>]

Chookij Vanatham <chookij vanatham eng sun com> writes:

> Hi Owen,
> 
> After I took a look at both Arabic and Devanagari Pango Engine,
> they are not considering the case of ZWJ and ZWNJ for their shaping.
> 
> As we know that, in unicode standard, we are able to use ZWJ and
> ZWNJ to control the shaping of scripts, particularly, Arabic and
> Devanagari. I'm wandering that how it can be done in Pango design
> because Arabic/Devanagari Pango engines themselves won't get called
> for the ZWJ/ZWNJ ?

It's simply a bug or missing feature - the fact that there are
no characters neutral with respect to shaping also means that
shaping with the Arabic engine is somewhat less efficient than
it could be, since each word is shaped individually.

(See the 3rd item on http://developer.gnome.org/status/pango-todo.html)

Of course, the ZWNJ gets handled basically OK because the fact
that the text on each side of the ZWNJ is shaped individually
means that no joining will occur.

I think the right algorithm is that there should be a standard
set of neutral characters - basically the whitespace characters -
and neutral characters with the same shaper on each side should
get handled by that shaper. Neutral characters with different
shapers on each side are more problematical. The assignment
should follow the bidi level when that differs, otherwise, its
probably pretty much arbitrary.

Regards,
                                        Owen