[XeTeX] New feature planned for xetex

[Jonathan Kew]

On 18/2/16 12:51, Philip Taylor wrote:
>
> Jonathan Kew wrote:
>
>> When \XeTeXinterwordspaceshaping is greater than 0, XeTeX will attempt
>> to support fonts where the width of inter-word spaces may vary
>> contextually, depending on the preceding and following text. This is
>> needed by fonts such as SIL's Awami Nastaliq (in development) where
>> words are expected to kern together across spaces.
>
>> These [other] cases are addressed with \XeTeXinterwordspaceshaping=2. With this
>> value, not only are inter-word spaces measured in context, but also each
>> run of text (words and intervening spaces) in a single font will be
>> re-shaped as a unit at \shipout time. This allows full shaping
>> (contextual swashes, ligatures, etc) to take effect across inter-word
>> spaces.
>
> I like these ideas, Jonathan, but wonder whether the two concepts should
> be orthogonal (possible implementation :  treat the value of
> \XeTeXinterwordspaceshaping as a bit mask, such that
> \XeTeXinterwordspaceshaping = 1 would be for SIL's Awami Nastaliq, etc.,
> \XeTeXinterwordspaceshaping = 2 would be for Hoefler Italic, etc., and
> \XeTeXinterwordspaceshaping = 3 would be for both). This would also
> allow future extensibility without compromising backward compatibility.


Hmm. It doesn't really make sense to re-shape text across inter-word 
spaces (as required for Hoefler Italic swashes, multi-word ligatures, 
etc) and yet NOT take account of this during line breaking (which is 
what level 1 provides).

As it happens, that exact combination IS what current xetex does with 
AAT fonts (only, not with OT or Graphite)... it doesn't do anything 
special w.r.t. line-breaking, but it does have a little-known (and not 
user-controllable) feature whereby it tries to re-shape entire runs of 
words in a single font. In practice, however, this doesn't always work 
very well, precisely because it fails to find the line-breaks that would 
be optimal for the globally-shaped text. So using Hoefler Text Italic 
with word-final swashes enabled will result in poor word spacing. The 
new implementation - when set to level 2 - gives markedly superior output.

IMO, the "two concepts" here build upon each other in a logical 
sequence: (1) do we allow smart-font shaping behavior beyond the 
boundaries of a single word to affect text measurement and line 
breaking; and (2) do we also allow it to affect the actual rendering of 
each word? Allowing it to affect the rendering WITHOUT affecting 
measurement and line-breaking would be somewhat nonsensical.

In an ideal world, \XeTeXinterwordspaceshaping=2 would be the default 
(perhaps the only) behavior. I'm not doing this, though, because there 
are compatibility implications (and in particular, there are potential 
side-effects on \special that may require workarounds at the macro 
level), as well as a significant performance cost for a feature that 
will make no visible difference with most fonts.

So... that was an interesting and thought-provoking suggestion, but at 
this point I think I'm inclined towards keeping the existing model. To 
me, it makes sense to think of these as increasing levels of support, 
rather than as independent features.

JK