[HarfBuzz] Compilation Troubles on Solaris 10

[Richard Wordingham]

I had difficulty compiling HarfBuzz 0.9.14 on Solaris 10, and I would
like advice on whether it is worth developing official, robust fixes.
You might advise me that the system is obsolete and not worth
supporting.

Environment:

O/S: SunOS 5.10 Generic_118833-18
HW: sun4u SUNW,A70
Compiler: gcc Version 3.4.3

Unfortunately, upgrading this system's software is not an option.

Compilation problems:

1) libstdc++.la is an empty file!  It is therefore not recognised as a
valid library, and compilation halts.

My work-around is to manually remove "-lstdc++" from the generated
script libtool.

2) The compiler does not recognise the option
-fvisibility-inlines-hidden.

My work-around is to manually remove it from the script configure.

3) The compiler considers the that overload resolution between the 
following templates in hb-open-type-private.hh is not possible:

 /* Cast to struct T, pointer to pointer */
template<typename Type, typename TObject>
inline const Type* CastP(const TObject *X)
{ return reinterpret_cast<const Type*> (X); }
template<typename Type, typename TObject>
inline Type* CastP(TObject *X)
{ return reinterpret_cast<Type*> (X); }

My workaround is to rename the former CastCP and change CastP to CastCP
until the compiler stops complaining.  I had to make the same change
for other pairs.

Run-time problems:

The program in test.cc core suffers an assertion failure because buffer
directionality is invalid.  I believe this is the intended response to
the behaviour in test.cc.  When I correct the file by calling
hb_buffer_guess_segment_properties() before hb_shape(), I get
intelligible glyph indices on Linux.  However, on Solaris the glyph
indices are returned as 0, even for such simple strings as "abcde".
I suspect misaligned data in the font file.  If this is the problem,
would the problem be fixed, or is it now considered now worth
supporting systems that don't handle this automatically? 

Richard.