[XeTeX] overriding \addfontfeatures

[Jonathan Kew]

On 4 Jun 2008, at 5:57 am, Will Robertson wrote:

> Okay, this is NOT a bug in fontspec.
> It might be a bug in the font, or it might be a bug in XeTeX.
>
> Take a look at the following example: (use Minion Pro if you don't  
> have Warnock Pro)
>
> \documentclass{article}
> \begin{document}
> \font\1="Warnock Pro:+onum" at 14pt \1 [123]
> \font\1="Warnock Pro:+lnum,+onum" at 14pt \1 [123?]
> \font\1="Warnock Pro:+lnum" at 14pt \1 [123]
> \font\1="Warnock Pro:+onum,+lnum" at 14pt \1 [123]
> \end{document}
>
> In the second case, I'd expect the +onum to override the +lnum, but  
> it doesn't; for some reason the +lnum has precedence. This seems to  
> happen for a number of Adobe fonts, at the very least.

This isn't really a bug, it's a manifestation of how OpenType  
features work and how the fonts are designed.

When you list several features such as "+onum,+lnum", you're not  
specifying an ordered sequence of effects to be applied, but an  
unordered collection of features to be activated "together". Within  
the font, each feature is associated with one or more "lookups" which  
actually perform the glyph substitutions or other effects. These  
lookups are applied in sequence according to their order in the font.

(Imagine a user interface where there's a check-box for each feature;  
you're just checking a collection of boxes, and the end result will  
be the same no matter which order you click them.)

So what's important is not the order in which you list the features,  
but the order in which the font designer arranged the lookups. And if  
you activate several features that potentially affect the same  
glyphs, the final result depends on how the designer built the font  
and arranged the lookups.

It seems that Adobe has chosen to arrange their lookups such that if  
'lnum' is used, the digit glyphs get replaced with alternate versions  
that are unaffected by the 'onum' lookup, and the 'lnum' lookup comes  
first.

Is this a bug in the font? I don't think so, it's a design choice  
(which you may or may not agree with). There is no single "right"  
answer for how features should interact.

Unfortunately, OpenType does not provide any mechanism for  
applications to know, in general, whether certain features interact  
or should disable each other; e.g., it really doesn't make sense to  
apply both 'sups' and 'subs' at the same time. In AAT, there are  
"exclusive" feature settings (presented as groups of radio buttons in  
the user interface), but OpenType doesn't have this, they're just  
individual on/off switches with no relation to each other.

JK