[XeTeX] misplaced combining diacritical marks 2

[Peter Baker]

  On 9/1/10 2:03 PM, Alexander Schultheiß wrote:
> Hey David,
>
>> This is just not true.  The example I sent you works correctly on my
>> machine; I deliberately included y-macron-acute which does not exist in
>> precomposed form in Unicode.
> I don't think they work correctly, in the sense that they are
> positioned according to the anchor points. I've attached a pdf with
> the results of the little investigation I did. For example, I've tried
> various methods to get a m+macron+acute with dotbelow (starting from
> line 2.1). It confirms my suspicion that xelatex doesn't care about
> anchor points in complex glyphs. It "fakes" them.
Something odd is going on here, but I don't think it has to do with 
anchor marks. Junicode actually has a precomposed character m + macron + 
acute + dot below. It's not a Unicode character, but rather is 
unencoded. (I can't remember why it's there: maybe a user asked for it 
at some point.) It should be getting substituted (via ccmp, the earliest 
lookup performed) whenever the renderer sees the sequence m + U0323 + 
U0304 + U0301 or m + U0304 + U0301 + U0323. I think that ICU/XeTeX is 
probably doing the right thing at this point. But Junicode's precomposed 
character is made of references instead of outlines, and sometimes 
ICU/XeTeX (not sure who's doing the rendering at this point) gets 
confused about glyphs composed of references.

Alexander, if you don't mind my sending you a big file off-list, I'd 
like to see if a change in the Junicode font will fix the problem.

Peter