[XeTeX] OpenType features

[Jonathan Kew]

On 2 Oct 2004, at 5:17 pm, Cyril Niklaus wrote:

>
> On 2 oct. 04, at 17:19, Yves Codet wrote:
>
>>
>> Le 1 oct. 04, à 19:18, Jonathan Kew a écrit :
>>
>>>
>>> Yes, that does look strange. It might be a font error, or it might 
>>> be an error in the OpenType implementation used in XeTeX. Are you 
>>> able to test the same text, with the same font, in a current version 
>>> of Windows to see how it renders with Uniscribe? If it behaves 
>>> properly there, then it's most likely a XeTeX/ICU layout bug.
>>>
>>
>> I'm afraid I can't do that. Here is the sample file if somebody has a 
>> PC and is willing to try (and can read Devanagari or save the text as 
>> a picture).
> The text displays fine on my win2000 box, the udatta is on the ja. I 
> just opened it in EI and Word as well (to be sure it was using the 
> Sanskrit 2003 font).
>

Is your Win2000 system actually doing positioning of the diacritics, or 
simply leaving them at default overstrike positions? (I suspect Win2000 
might not include GPOS support in Devanagari.) You could test this by 
comparing the positioning of marks on 'ba' and 'ka', for example.

I believe this may be a font problem with the attachment point on the 
'ja' glyph in the Sanskrit 2003 OpenType tables (but I'm not confident 
of this yet; more investigation is needed). Here's an experiment I 
tried, which seems to suggest the font may be at fault:

(1) In Yves' test file, I replaced the sample text with

\begin{verse}
	\dev{
	दे\\
	बे\\
	के\\
	जे\\
  	}
\end{verse}

which demonstrates that with Sanskrit 2003, the e-vowel also appears 
way to the right when used on 'ja'. (The same with other diacritics 
above, too.) Note that adjustments to the horizontal position of the 
diacritic work fine for other letters (screenshot attached):
-------------- next part --------------
A non-text attachment was scrubbed...
Name: texshop_image.pdf
Type: application/pdf
Size: 7952 bytes
Desc: not available
Url : http://tug.org/pipermail/xetex/attachments/20041002/8a148fea/texshop_image-0002.pdf
-------------- next part --------------

So this establishes that the problem is not limited to the udatta but 
occurs with other marks.

(2) In place of Sanskrit 2003, use a different Devanagari font that 
includes OpenType tables for mark positioning. One that I'm aware of is 
Raghindi, available from 
<http://rohini.ncst.ernet.in/indix/download/font/raghu.ttf>. With this 
font, I get correct positioning of the mark over all the sample base 
characters:
-------------- next part --------------
A non-text attachment was scrubbed...
Name: texshop_image.pdf
Type: application/pdf
Size: 7240 bytes
Desc: not available
Url : http://tug.org/pipermail/xetex/attachments/20041002/8a148fea/texshop_image-0003.pdf
-------------- next part --------------

So the problem with positioning marks on 'ja' seems to be unique to the 
Sanskrit 2003 font, not a general flaw in the layout engine XeTeX is 
using. (Udatta works properly in Raghindi, too.)

I'd be interested to see what current versions of Windows, and any 
other Indic OpenType layout engines around, do with this font; I'll 
test on Monday when I can get access to a Windows machine. It is, of 
course, still possible that the font is built correctly but for some 
reason XeTeX misinterprets the data for that particular glyph.

Jonathan