This should fix: https://github.com/larskanis/fxruby/issues/24

rb_during_gc() isn't sufficient for indication of free-phases on Ruby < 2.2.
In some sitiations it returns zero, altought being within a free-phase.
Moreover Rubinius doesn't support rb_during_gc() at all.

So we now use both, rb_during_gc() and our own GC-detection, to avoid calls
that are prohibited while GC phases.

This reverts commit acec4416.

This fixes the case that a wrong class is retrieved for FXMaterial#ambient.
This therefore fixes the TC_FXMaterial#test_bug test case.

This removes a bunch of warnings while 'rake test'.

are checked while development.

Only the address of the foxobj is checked, but this is ambiguous
for the first member of a struct or class like this:

struct FXAPI FXMaterial {
  FXVec4f    ambient;           // Ambient material color
  FXVec4f    diffuse;           // Diffuse material color
  FXVec4f    specular;          // Specular material color
  FXVec4f    emission;          // Emissive material color
  FXfloat    shininess;         // Specular shininess
};

One and the same address point to a FXMaterial and a FXVec4f object.

C++ DLLs built with i586-mingw32msvc and i686-w64-mingw32 are not compatible.

So this changes the build, so that we ship the ports directorys of both
compiler versions.

This fixes https://github.com/larskanis/fxruby/issues/22

The semantics have changed with 54a11707. Only "owned" objects can
be expected to be the same. Borrowed objects can change it's object_id
for each call.

so that all FXTRACE() and FXASSERT() calls are skiped, per default.

This should improve performance.

This fixes https://github.com/larskanis/fxruby/issues/21
It was introduced in commit 347966fc.