[HarfBuzz] 'calt' in Indic shaper

[Jonathan Kew]

On 7/8/13 16:15, John Hudson wrote:
> On 07/08/13 3:45 AM, Jonathan Kew wrote:
>
>> The Indic 'calt' discrepancy we ran into that triggered this question
>> appears to actually be the features-across-cluster-boundaries issue. In
>> uniscribe, 'calt' does not apply across cluster boundaries; in harfbuzz,
>> it does. This leads to many unexpected differences: e.g. in Gujarati,
>> the "pseudo-ligation" of <U+0A9C,U+0AAF>
>
> Are you referring to the Nirmala fonts again? I think this is a font
> bug, and the context for the connecting -Ya is incorrect. This
> substitution should only be happening in conjuncts, so the full form Ja
> should not be acting as a trigger.

Yes, all this arose as a result of running comparisons (harfbuzz vs 
uniscribe) with Nirmala.

For examples of the Gujarati issue in Nirmala (when harfbuzz applies 
'calt', see

http://people.mozilla.org/~jkew/hb-test/testcase-view.html?file=out/fonts/win8/Nirmala.ttf/gu.txt.ot-vs-uniscribe#G_0_866

and many of the other testcases in the list there.

And regarding the Devanagari line-extender, again inserted by 'calt' as 
I understand it, see examples such as

http://people.mozilla.org/~jkew/hb-test/testcase-view.html?file=out/fonts/win8/Nirmala.ttf/hi.txt.ot-vs-uniscribe#L_1_805

where (to my eye) the extension of the top-line is unnecessary and looks 
poor, compared to the unextended Uniscribe rendering. There are many 
similar examples in the list; although there are also some, such as

http://people.mozilla.org/~jkew/hb-test/testcase-view.html?file=out/fonts/win8/Nirmala.ttf/hi.txt.ot-vs-uniscribe#L_1_262

where the extension does serve the intended purpose of avoiding a rather 
cramped above-line situation.

> I'm working on some Nirmala UI fixes at the moment, so will address this.

Awesome, thanks.

JK