[HarfBuzz] Obscure Rendering Difference Involving Ligatures

[Richard Wordingham]

I've encountered an obscure rendering difference between
Uniscribe/DirectWrite as in IE11 on Windows 7 and HarfBuzz 0.98+.  It
shows up in my test cases in
http://homepage.ntlworld.com/richard.wordingham/lanna/renderer_test.htm.

On HarfBuzz-based browsers, the word meaning 'to lead' in the
section headed 'Ligature NAA' displays in the 'text' column as a
connected glyph with a small circle above it.  In the Firefox browser,
both on Linux and Windows 7, when I hack this to the 3-letter ASCII
string "nAM" the small circle moves up and rightwards so much that it is
severely clipped in Firefox.  However, when I display the same page in
IE11 on Windows 7, the string displays as the original Tai Tham text
does on Firefox. 

I can reproduce the bad display using the command 

hb-view --features=ss02,abvs,blws,pstf,blwf ~/.fonts/dalekh.ttf nAMn

(The extra letter 'n' of the string extends the display so as to see
the small circle.)

Is it worth investigating the difference from Windows?  There is a
fairly complicated sequence of transforms, which probably ought to be
reduced for the investigation.  I suspect it depends on the
inheritability of the base/mark difference. I could change the font to
remove the effect; it's purpose is to render the incorrect sequence
<U+1A36, U+1A63, U+1A76> differently to the correct sequence <U+1A36,
U+1A76, U+1A63> by positioning marks above on the two halves of the
ligature differently.

Richard.