[HarfBuzz] Thai Tam Shaping

[Richard Wordingham]

On Wed, 2 Dec 2015 16:21:17 +0900
Simon Cozens <simon at simon-cozens.org> wrote:

> The Lanna for Lanna is "ᩋᩣᨱᩣᨧᩢᨠᩕᩃ᩶ᩣ᩠ᨶᨶᩣ"[*].

as a geographical expression. 

> I'm getting very different renderings of that string in Safari,
> Firefox and hb-shape, using the same fonts. I don't know which is
> correct.

Firefox, Lanna Alif font.

I believe Safari uses AAT plus approximate OT for favoured scripts.
Tai Tham is clearly not a favoured script; it is being rendered with no
shaping.

It would appear that the Noto Sans Tai Tham script is missing the <NA,
AA> ligature, which should simple in this phrase.

> Notice in the attached PNG that not only is HB giving me
> no-base-glyph dotty circles, the MEDIAL RA (the thing that looks like
> a ] ) is moves all over the place.

Harfbuzz has clearly abandoned support for the Lanna script; it
appears to be using the rules of the Universal Script Engine, which, as
Microsoft has been told and its management has acknowledged, do not
support the Tai Tham script with the proposed text encoding.  (Unicode
may claim to include Tai Tham script, but it only defines encoding
simple Tai Tham word fragments, not text.)  The major problem is that
it does not support post-vocalic consonants, as in the <U+1A60 SAKOT,
U+1A36 NA>.  A secondary problem is that tone marks are expected to
follow vowels, so one can't use <U+1A43 LA, U+1A76 TONE-2, U+1A36 AA>.
If one were to use <LA, AA, TONE-2>, then one would need a large table
to position TONE-2 over the LA properly while formally placing it
relative to the AA. Swapping AA and TONE-2 could well fall foul of the
undocumented semantics of GSUB and GPOS lookups.

The Harfbuzz abandonment may be an attempt to lure me to the US to
discuss matters. I couldn't justify the extravagance of a short trip to
the US.

Firefox uses an older version of Harfbuzz, which mostly supports the
Lanna script, and can be made to support nearly all strings if
workarounds are included in the font.

I'm not quite sure what has gone wrong for MEDIAL RA in hb-view and
Noto Sans Tai Tham, but it may be related to the fact that, at least
not so long ago, the font was designed to work without Indic
rearrangement, thereby making it incompatible with Harfbuzz.  That it
works with Firefox casts doubt on this explanation.  Likewise, I don't
know what has gone with <SAKOT, NA>. Perhaps the font doesn't cater for
placing it under a dotted circle.

Richard.