[HarfBuzz] Question about HarfBuzz-ng CTL Shaper Status

[Keith Stribley]

On 21/09/10 13:21, Kenichi Handa wrote:
> For certain combinations, such as Tai Tham's "Medial R" which needs to
>> precede the base consonant, using ccmp substitutions will work.  This
>> requires creating a lot of "ligature" glyphs.  But it works.
>>      
> I think it's far from:
>    a "pure" version of the font that uses only proper OpenType features
>    

It may not be "pure" but it may be pragmatic. If you want a font to work 
on Windows and Linux today, then taking advantage of the default GPOS 
and GSUB features which are already enabled is a reasonable approach. 
The Myanmar 5.1/5.2 Unicode fonts that exist take advantage of features 
such as clig, liga, kern and in one case mark/mkmk to give correct 
rendering under Uniscribe and in some cases Harfbuzz-NG and Pango 
without having a specific shaper.

There are 2 ways to get reordering working without a script specific 
shaper. Ligatures is probably faster, but requires a lot of glyphs, 
which will probably need to be autogenerated. The other approach is 
using multiple context specific substitutions to perform the insertion 
and then remove the unwanted glyphs from their original position. The 
later requires a lot of lookups and may need guard glyphs to be inserted 
temporarily between syllables to prevent reordered glyphs matching 
unordered patterns.

I think it is a shame that Firefox 4 is not enabling Harfbuzz-NG for 
complex scripts, at least for those which don't have a specific platform 
shaper.  Uniscribe on Vista and Win 7 should support default features 
for these scripts, but many computers still run Win XP and don't have 
Complex Script support specifically enabled since it is off by default. 
Perhaps harfbuzz could be enabled at least if the web developer actively 
uses the CSS3 font-feature-settings or font-language-override. This 
might help encourage people to move away from quasi-Unicode fonts for 
these scripts, to fonts which properly conform to the standard.

> The right approach will be:
>
> (1) Write a specification document something like what
> Microsoft does for various scripts:
>      <http://www.microsoft.com/typography/SpecificationsOverview.mspx>
> This should be a base of Tai Tham font&  layout-engine
> developpers, and for that, it may be good that this document
> is officially approved by a proper Tai Tham authority
> (perhaps by the one who proposed Tai Tham script to Unicode
> or ISO10646).
>
> (2) Implement a Tai Tham layout engine as a pango module or
> a harfbuzz-ng's shaper according to (1).
>    

With some scripts it is possible that there is still insufficient 
information available to write a reliable shaper that will work with 
every language using the script. In the case of Myanmar, a shaper was 
written for the original harfbuzz for one interpretation of the 
ambiguous Unicode 4 specification for Myanmar. However, that was never 
updated for Unicode 5.1 which removed the ambiguities and actively 
prevents correct rendering of fonts which would work with the default 
shaper. Fortunately Pango doesn't use that shaper, though other apps do.

In the interim, I would suggest that if (1) and (2) are implemented for 
a script without a current shaper on all platforms, that it actively 
checks whether the font has been written for that shaper e.g. by 
checking for the presence of non-default features, which would only be 
present in a font specifically targeting that shaper. If those features 
are not present, then it should just use the default features. There is 
also the danger that even if (1) is written there is no guarantee that 
other rendering engines will implement it in the same way, but I admit 
it will give a faster rendering solution for the long term.

By the way, for OpenType font testing I use grsvg with its python GTK 
user interface. You can build it from 
http://scripts.sil.org/svn/graphite/grutils/trunk on Linux. Use CMake 
for the build and install the ICU and harfbuzz-ng libraries first. It 
will allow you to compare the output of the 2 renderers and if graphite 
/ graphiteng is available it will also show the graphite rendering. It 
is using SVGs as an intermediate format underneath (not SVG fonts), so 
it won't show hinting, but is alright for testing OT/Gr rendering.

Best regards,

Keith Stribley