[HarfBuzz] harfbuzz-ng: Branch 'master' - 86 commits

[Behdad Esfahbod]

On 05/14/2012 02:27 PM, Rajeesh K Nambiar wrote:

> More than glad to inform that Malayalam rendering is in awesome shape
> as a result. We have ran the tests and updated the test results here:
> http://silpa.org.in/pub/tests/hb/ml/ . The failure cases are only 7
> out of 41 -  test cases12, 14, 31, 32, 34, 36 and 37. The Pango
> versions can be considered as correct/expected rendering. Looks like
> test cases 14, 31 and 32 are due to 'Ra' not being treated as
> Post-Base. Cases 36 and 37 doesn't display Reph correctly (but Meera
> does with harfbuzz too). Please note that for case 36, Meera's
> rendering using harfbuzz is correct and pango rendering is incorrect.

Cool.  I actually passed my Malayalam corpus (extracted out of Wikipedia)
through our code and looks like we are failing on about 14 percent compared to
Uniscribed.  Which is expected since we don't support pre-base reordering
consonants, and as you note our Ra classification seems to be wrong.  Over the
next few weeks we get to focus on scripts other than Devanagari and hammer
them down one by one.


> Lastly, joining the party with a huge THANK YOU!

Thanks :-).  I cannot stress this enough: Jonathan Kew and I contributed
equally to this work.

behdad

>> Cheers,
>> behdad