[MPEG-OTSPEC] Updates to specification

[Norbert Lindenberg]

> On Aug 16, 2020, at 20:44, 梁海 Liang Hai <lianghai at gmail.com> wrote:
> 
>> On Aug 16, 2020, at 20:34, James Clark <jjc at jclark.com> wrote:
>> 
>> I would urge against a "reference implementation" approach: a prospective implementer should be able to find out everything they need to know by reading the spec; …
> 
> Ideally yes. However it’s been pretty clear that in this small field of font technologies, it’s not realistic to expect many independent implementations from scratch, and thus a first-party reference implementation is vital for adoption of new technologies/features.

I’m aware of five OpenType shaping implementations at this point:

– HarfBuzz (Behdad and friends) – complete, well maintained
– CoreText (Apple) – complete, well maintained
– Uniscribe/DirectWrite (Microsoft) – complete
– Lipika (Adobe) – incomplete; Adobe products use HarfBuzz to fill in gaps
– Allsorts (YesLogic) – under development

I think the first goal for a specification would be to get these to produce the same shaping results for the same text, same fonts, and same additional parameters. The second goal would be get them to adopt new features. Enabling new independent implementations would at best be third.

How would a reference implementation make it easier to reach these goals than just a combination of a real specification and a conformance test suite?

Norbert


>> … they should not have to reverse engineer an implementation, reference or otherwise.
> 
> I don’t think there’s an intention to include certain information only in the reference implementation when people talk about it… It’s just literally meant to be an “implementation” (parallel to any implementation) and a “reference”. I don’t think anyone plans to use a reference implementation as an excuse for low quality and incompleteness of the specification.
> 
> Best,
> 梁海 Liang Hai