[tex-live] Ftuture state of XeTeX in TeXLive

[George N. White III]

On Sun, Oct 30, 2011 at 7:42 AM, Khaled Hosny <khaledhosny at eglug.org> wrote:

> On Sun, Oct 30, 2011 at 04:25:21PM +1100, Vafa Khalighi wrote:
>> XeTeX font support is heaps better and  stable than what luaotfload package
>> offers and I guess that is why many users still like using xetex instead
>> luatex. I personally believe that it is a bad practice that luaotfload just
>> copies ConTeXt code, it should not be deeply dependent on ConTeXt because Hans
>> may want to try experimenting with some features today and next day he gets rid
>> of them just like the recent updates of luaotfload that Khaled talked about it.
>> I think, this is awful! What should users who used those features (and need it
>> heavily in their daily typesetting tasks, do?). They wake up one day and
>> suddenly see that yes, luaotfload does not provide the features they need.
>> luaotfload needs to be written from scratch independent of any ConTeXt code.
>
> The situation is not as bad as you make it seems, what have gone is two
> minor features that IMO was a mistake to provide them in the first
> place, but since we are talking about a yet to be released version of
> luaotfload, there might be an alternate solution at the time of release.
>
> Writing an OpenType layout engine is not a simple task, and you can
> judge from the many years it toke FOSS community to have a really good
> one, HarfBuzz (the name luaotfload is misleading, font loading is about
> the easiest part of luaotfload, OpenType implementation is really what
> matters.) If it were for me, I'd plug HarfBuzz into luatex proper and
> call it a day, but this does not align well with the "design" principles
> of luatex so it is unlikely to happen.

If plugging harfbuzz into luatex does not require a huge effort, it could
serve as bridge from xetex to luatex while a more principled design
is being created.   Principles are nice, and have benefits over the long
haul, but in cases where the design is evolving it really helps to get
an implementation into the hands of users and let them point out the
areas where work is needed.

> The main goal of luaotfload, besides having an OpenType engine for
> lualatex, is too make sure we don't end up with two different OpenType
> implementations for luatex, each broken in its own way. OpenType is such
> a horribly documented "standard" that is there is no single
> implementation of it that does not have its own share of bugs (it is
> often not easy to tell if a bug is a bug or since the documentation are
> so fuzzy in certain areas).

The way to bring clarity in cases of fuzzy standards is for one implementation,
perhaps harfbuzz-ng is a candidate, as the reference and try to match
that behaviour.
The reference implementation won't be perfect, but needs to be open to change as
problems are identified.

-- 
George N. White III <aa056 at chebucto.ns.ca>
Head of St. Margarets Bay, Nova Scotia