[HarfBuzz] Change in HarfBuzz after version 0.90 ?

[Richard Wordingham]

On Tue, 24 Dec 2013 12:51:53 -0500
Ed Trager <ed.trager at gmail.com> wrote:

> Can someone please give me a hint about what changed in HarfBuzz? Is
> this a bug in HarfBuzz? Or is some definition in my OpenType feature
> file not correct after changes were made in HarfBuzz?

> Here is a one-word example from the images that anyone can use for
> testing:
> 
> U+1A20 U+1A62 U+1A76 U+1A60 U+1A36

As Martin Hosken points out, the problem is that U+1A76 and U+1A60 are
swapped round when the above sequence is normalised.

Now, I reported this (topic 'SEA Syllable Structure and Tai Tham
Combining Classes') as a problem with HarfBuzz 0.9.18 on 11 June 2013,
and suggested ('Solution 2') that the canonical combining class of
U+1A60 be overridden in function modified_combining_class() in
hb-unicode-private.hh.  On 13 June 2013, Behdad replied 'Sounds about
right'.  However, I can't find any evidence that this change was ever
implemented in HarfBuzz - but I use git so little that I may have
missed any brief period of incorporation.

Now, it is possible that Ed has actually fallen victim to an update in
the Unicode Character Database (UCD) version supported on his
machine.  His font will work provided the UCD information supplied to
HarfBuzz is for Unicode Version 5.1 or earlier.  When I reported the
problem, I had to link with my own UCD-provider to provoke the problem
- I was still using Ubuntu 10.04.

Richard.