[HarfBuzz] Regression with ZWJ in Indic scripts

[Khaled Hosny]

The issue reported below seems to be a regression in 0.9.13, basically
ZWJ seems to be completely ignored, with 0.9.12 I get the desired
output.

Regards,
Khaled

----- Forwarded message from Ian-Mathew Hornburg <imhornburg at gmail.com> -----

From: Ian-Mathew Hornburg <imhornburg at gmail.com>
To: xetex at tug.org
Subject: [XeTeX] Incorrect Bengali ZWJ behavior and v2 script spec
Reply-To: Unicode-based TeX for Mac OS X and other platforms <xetex at tug.org>

I may have identified two possible bugs in 0.9999.0: the release notes
indicate that the version-2 OpenType Indic script tags are now supported,
and I’ve been testing various Bengali-script fonts with the git version of
XeTeX and a current install of TeX Live 2012 to check them for correct
shaping behavior. I’ve posted a MWE reproducing some examples from the
Microsoft standard here: [http://pastebin.com/mgAX8c7U].

Microsoft ships two Bengali fonts (Vrinda and Shonar Bangla; both v6.80)
with Windows 8 that support both the older (beng) and newer (bng2) Bengali
script specs. (Older versions of each are shipped with Windows 7 and other
Microsoft products.) The fonts behave correctly when using the beng script
feature, with the exception of a particular ZWJ sequence: the Microsoft
spec [https://www.microsoft.com/typography/OpenTypeDev/bengali/intro.htm]
says that the sequence of consonant-hasant-ZWJ-consonant should prevent a
ligature of the two consonants, then render a half-form of the first
consonant. XeTeX currently fails to suppress the ligature.

----- End forwarded message -----