Fallback for Sinhala Consonant Clusters

[Richard Wordingham via Unicode]

On Sun, 14 Oct 2018 17:15:26 +0900
"Martin J. Dürst via Unicode" <unicode at unicode.org> wrote:

> Hello Richard,
> 
> On 2018/10/14 09:02, Richard Wordingham via Unicode wrote:
> > Are there fallback rules for Sinhala consonant clusters?  There are
> > fallback rules for Devanagari, but I'm not sure if they read across.
> > 
> > The problem I am seeing is that the Pali syllable 'ndhe' න්‍ධෙ
> > <U+0DB1 NAYANNA, U+0DCA AL-LAKUNA, 200D ZWJ, U+0DB0 MAHAPRAANA
> > DAYANNA, U+0DD9  
> > KOMBUVA>  
> 
> Let's label this as (1)
> 
> > is being rendered identically to a hypothetical Sinhalese
> > 'nēdha' නේධ <U+0DB1, U+0DDA DIGA KOMBUVA, U+0DB0>,  
> 
> It (2) doesn't look identically to (1) here (Thunderbird on Win 8.1).
> 
> Your mail is written as if you are speaking about a general
> phenomenon, but I guess there are differences depending on the font
> and rendering stack.

The critical one is whether the font has the conjunct.  The default
Sinhala font on supported Windows, Iskoola Pota, has the conjunct. For
an example that should illustrate my points with that font (at least,
as on Windows 7) and the HarfBuzz renderer (as I believe in
Thunderbird), we have

1') Pali thve ථ්‍වෙ <U+0DAE MAHAAPRAANA TAYANNA, U+0DCA
AL-LAKUNA, 200D ZWJ, U+0DC0 VAYANNA, U+0DD9 KOMBUVA>

It's a very rare syllable - it only occurs in sandhi, and I have only
a single example.  Iskoola Pota has neither the conjunct nor the
touching form; I would actually expect it to be the touching form
that exists.

2') Misleading look-alike thēva ථේව <U+0DAE, U+0DDA DIGA KOMBUVA,
U+0DC0>

3') Preferred fallback appearance thve ථ්වෙ  <U+0DAE, U+0DCA, U+0DC0,
U+0DD9>.

My question is, 'What should a rendering stack that claims to support
the Sinhala script display when it lacks the conjunct in the font
being used?'

Now what does get displayed does depend on the rendering stack.
HarfBuzz (e.g. Firefox, Google Chrome, LibreOffice, and most Linux) and
Notepad on Windows 7 move the vowel to the left and display al-lakuna,
the display I object to. iPhone and Notepad on Windows 10 display
the vowel in the middle and display al-lakuna (possibly ligated), which
is the solution I prefer.

> Hope this helps.

Well, it has prompted me to find a 'me-too' argument for improving the
rendering.  I wanted a standards-based argument.

>> Missing touching consonants are being rendered almost as though
>> there were no ZWJ, but the combination of consonant and al-lakuna
>> is being rendered badly.

This looks like a common font problem.  Iskoola Pota does not suffer
from it.

Richard.