Now the test/object test deadlocks as expected.  Fix coming.

Exposes the non-atomicity of user_data opertaions at this time because
we call finish() while still locked and modifying the object.  In fact,
I'm surprised that it doesn't deadlock.  It should.

It uses locale information to detect default language.  It's used by
hb_shape() whenever language is not set on the buffer.

Not sure how to properly test it in the test suite.  Tested by observing
that with DejaVu Sans we select the proper local glyph version for U+431
under Serbian locale.  See http://www.pango.org/ScriptGallery

Step version up to 0.5.0.

Also, fix to pass "make distcheck"

No one who builds harfbuzz static uses the autotools build system to do it.

Since we're not win32-dll clean the way libtool docs define it.

While the cluster fields of the glyph string are usually sorted, they
wouldn't be in special cases (for example for non-native direction).
Blindly using bsearch is plain wrong.  If we want to reintroduce this
optimization we have to make sure we know the buffer clusters are
monotonic and in which direction.  Not sure it's worth it though.

For now, by taking a global user_data mutex.

Note that the static variable has to be a global static, as gcc
implements local statics differently and that would require linking
to libstdc++, which we don't want.

More intuitive.

Mutex (and Windows.h by extension) are fairly isolated now.

Don't use _cached()

The Win32 definitions for LONG, ULONG, etc conflicts with
hb-open-type.h.  Avoid that by making sure hb-object-private.h
and hb-open-type.h are not included in the same compilation unit.

Needed to be able to include <Windows.h> from hb-object-private.h.

Fixes problem with uninitialized sub_blob->mutex among other things.

Reported by Bradley Grainger.

Patch from Bradley Grainger.

Based on patch by Bradley Grainger.