This is a higher-priority shaper than default shaper ("ot"), but
only picks up fonts that have AAT "morx"/"mort" table.

Note that for this to work the font face's get_table() implementation
should know how to return the full font blob.

Based on patch from Konstantin Ritt.

Useful for Android / Chrome / etc when ICU is built in the same
library as harfbuzz itself.

Apparently some modern fonts have proper GPOS mark positioning
tables, but undesirable precomposed forms!  See thread
"Hebrew composition to presentation forms" and:

  http://tex.stackexchange.com/questions/156775/having-trouble-with-vowel-positioning-in-ezra-sil-xelatex

Test case: U+fb1d,05d9,05b4

Seems like configure.ac is already able to handle it.

See thread "[HarfBuzz] compilation error of 0.9.26 with MinGW"
started by Werner.

Fixes https://github.com/behdad/harfbuzz/pull/21

Fixes github.com/behdad/harfbuzz/pull/19

The OT shaper supersedes the fallback shaper in every case
and the latter become an extra weight for 99.9% of users.

Fixes https://github.com/behdad/harfbuzz/pull/16

Fixes https://github.com/behdad/harfbuzz/pull/17

I believe Windows 8 disables it, and spec update dated
Jan 2014 also clearly says it's disabled by default:

  http://www.microsoft.com/typography/OpenTypeDev/arabic/intro.htm#features

Tone-mark reordering and improved Old Hangul support

[unicode] Exclude the Jamo Filler characters from Default_Ignorable, as some fonts want these to be visible/spacing glyphs.

[hangul] Reorder Hangul tone mark to beginning of syllable, unless font implements it using a zero-width glyph.

[hangul] Apply the appropriate *jmo features to decomposed syllables, including Old Hangul sequences that don't have Unicode compositions. Merge clusters in decomposed syllables.

As discussed on
https://github.com/behdad/harfbuzz/pull/10#issuecomment-31442030

Not tested.  Ouch.

Seems to be working now.

Part of https://github.com/behdad/harfbuzz/pull/10

I had tried to expand to fill the blocks, but that sounds wrong in
retrospect.

Now default shaper is truly no-op.

No shaper has more than one behavior re this, so no need for a callback.

Not exhaustively tested, but I think I got the intended logic
right.

The logic can perhaps be simplified.  Maybe we should disabled
normalization with this shaper.  Then again, for now focusing on
correctness.