[Devel] Problem with ligature patch

[Owen Taylor]

Hi Werner,

Apparently there are problem with the changes in:
  
http://people.redhat.com/otaylor/opentype-patches/index.html#freetype-07-ligature-ids

See:

http://bugzilla.gnome.org/show_bug.cgi?id=118472

And certainly, I can see where this patch wouldn't always work; you
could have:

 A ' C ^

And ligatures:

 A B  => AB  (ignore marks flag set)
 ' ^  => '^ (ignore base flag set)

Then I'd expect the output:

 AB '^

Though I do understand, I think, the problem that is trying to be
solved here - the ArabicTypesetting font distributed with VOLT has
ligatures:

 uni064E uni0651=> uni0651064E  (FATHA + SHADDA)

So, if you had two base glyphs that ligatured mixed with FATHA
and SHADDA, there is potentially a problem. 

Perhaps the right solution here is simply ordering?
if you look at ArabicTypesetting, the mark+mark ligatures are
part of the 'ccmp' features which is should be the first feature
applied for Arabic, while the base+base ligatures are part of the
'rlig' feature, which is applied later.

When this ordering is maintained, the issue goes away.

So, perhaps the problem is one of the FT_GSUB_Apply_String API? 
I think the Microsoft code uses first the order of features
in the relevant spec, *then* the LookupList order, rather than
applying all lookups strictly in LookupList order, which is what
will happen with the current API (unless you make multiple passes)

Do you happen to remember what font and what input you were using
when you added this patch?

Thanks,
                                        Owen