Re: Fwd: OLPC arabic

Hi Tomas,
              Thanks very much for the ligature clue! I have committed
a fix to svn TRUNK. It looks like it fixes the bug :-)

We'll test it some more and see.

I'll check into the mem corruption issue next.

Cheers!

Martin

On Thu, Oct 16, 2008 at 5:26 PM, Tomas Frydrych <tf@o-hand.com> wrote:
> Martin Sevior wrote:
>> Hi Khaled,
>> Thank you very much for the detailed description of the
>> problems. Unfortunately I'm totally snowed under at my day job for the
>> next 2.5 weeks. I hope Tomas Frydrich, the main author of the
>> rendering code can fix some of these issues. Otherwise I'll have a go
>> when things ease off at work.
>
> Sorry guys, I will not have a time to look into this in the foreseeable
> future, but at least a comment.
>
>> 1) OpenType rendering: Pango does OpenType layout very well, but AbiWord
>> some times fail to apply OpenType features witch result in non-joined
>> Arabic words. I've reported this upstream a year a ago (#11307) but got
>> no response so far. This isn't Arabic specific, it can be tested in
>> fonts that has fi, ffi, fl and like ligatures (DejaVu fonts for example)
>> and you will see that the ligatures aren't always applied.
>
> My guess would be that the ligature ends up split across two runs; if
> that is the case should just be a question of fixing the run merge
> mechanism.
>
>> 2) The other issue is those random symbols instead of Arabic text, I
>> first thought that it is a problem with font fallback mechanism, but
>> this doesn't to be the case and I can't identify a regular pattern for
>> this. It occurs randomly with any text and can't tell when to expect it.
>
> That sounds like some kind of memory corruption, possibly OOB somewhere
> in the character -> byte offset conversions we do for Pango.
>
> Tomas
>
>
>
Received on Thu Oct 16 09:20:42 2008