[HarfBuzz] harfbuzz-ng: Branch 'master' - 3 commits

[Behdad Esfahbod]

On 07/21/11 14:29, Grigori Goronzy wrote:
> On 07/21/2011 06:01 PM, Behdad Esfahbod wrote:
>> commit aa7264123a088936f2043b45d4d41ca7413fabe5
>> Author: Behdad Esfahbod <behdad at behdad.org>
>> Date:   Thu Jul 21 11:34:59 2011 -0400
>>
>>     Only form clusters if we are reversing
>>     
>>     This produces more accurate cluster mappings.  Cluster mappings are
>>     minimal now.  Combining marks get their own cluster value most of
>>     the time.
>>
> 
> Is this really desirable?!
> 
> I need to handle "logical characters" and this was easily possible with
> the old mapping that put glyph and associated mark into a single
> cluster. This change pretty much completely breaks the way I layout text
> at the moment...

Right.  I was in the same camp myself.  I remember Jonathan had a point re not
doing it.  I don't remember the details.  Jonathan, can you elaborate?

The reason I ended up doing this however is that what we were doing was not
exactly what UAX #29 [1] does anyway.  So instead of fooling users into
believing that this is useful for cursor positioning, I decided to make it do
one thing: map characters and glyphs minimally.

Here's how I fixed pango:


http://git.gnome.org/browse/pango/commit/?h=harfbuzz-ng-external&id=26475e516f2384efaefe3dd97bd99f2176291354

You can do something similar.  If it's considered useful, I can provide an
hb_buffer_*() API to do it maybe.

behdad

[1] http://www.unicode.org/reports/tr29/

> Best regards,
> Grigori