[HarfBuzz] 'calt' in the uniscribe arabic shaper

[Jonathan Kew]

Following up on the recent questions regarding 'calt' in Indic 
scripts... I'm also noticing some odd behavior in Arabic.

(1) According to http://support.microsoft.com/kb/2786400, contextual 
alternates (which I assume to mean the 'calt' feature) should *not* be 
applied automatically by uniscribe (by default) in Windows 8, or in a 
fully-updated Win7.

However, testing with Arabic Typesetting on Windows 8, I -am- seeing the 
'calt' feature applied by default. A simple testcase is <06C1,06D2>, 
where the 'calt' feature selects a special ligated form. While I 
actually think this is the RIGHT thing to do, it seems to conflict with 
KB2786400's description of the intended behavior.

A uniscribe bug, a documentation bug, or what...?

(2) If the same word is encoded using the Presentation Forms codepoints 
(not recommended, I know, but there's a certain amount of legacy data 
out there in the wild) as <FBA8,FBAF>, then uniscribe does NOT apply the 
'calt' feature - indeed, I suspect it's not applying any of the Arabic 
shaping features.

However, if I shape the same text using DirectWrite, the 'calt' feature 
IS applied even when the text was encoded using the presentation-form 
codepoints.

(This difference can be observed, for example, by using Firefox, 
disabling harfbuzz (gfx.font_rendering.harfbuzz.scripts = 0) and 
comparing the rendering with hardware acceleration, in which case 
DirectWrite will be used, vs with hardware acc disabled, when Uniscribe 
will be used.)

JK