[XeTeX] Proper way to set up OT Features

[Adam Twardoch (List)]

The OpenType Layout mechanism is based on "glyph runs". First, the text
engine separates a line into glyph runs, and then applies features to
each run as if it were separate (i.e. there cannot be any feature
interaction between the runs).

The tricky part is that every layout engine has a different mechanism
for separating the line into the runs.

Of course a switch of the font, or font size, always defines a run
boundary.

Most layout engines treat the change of text directionality as a glyph
run boundary. But they all may differ in how they process characters
that do not have a strong directionality. So if you mix Arabic and
English, and put a period or a number between a portion of English and
Arabic text, different layout engines may classify the period or a
number as belonging to one or the other of the adjacent runs, or even to
a separate run. Same happens with other punctuation characters.

If you type in an Arabic letter, a slash (/) and an English letter, and
if the font has GPOS kerning pairs defined between the Arabic letter and
the slash, and between the slash and the English letter, either one of
those pairs or even both will not work.

In addition -- and here the difference is even larger -- layout engines
treat the space character very differently.

Some applications add an extra space glyph at the end of the line (for
example InDesign), so certain contextual rules won't work.

Some do and some don't render the space glyph that is used in the font.
For example, XeTeX doesn't. If the space glyph is non-blank, it will
render in Word but not in XeTeX. In other words, XeTeX puts a run
boundary at the beginning and the end of each word, so no OpenType
Layout lookups that involve the space glyph will work.

Best,
Adam