ext/System/Vector2.hpp build broken on OS X x86_64 with Clang

[jashank]

Trying to build rbSFML at revision f8297fb48d on Darwin/OS X 10.7, x86_64, with Clang/LLVM.

Compiling ./ext/System/Vector2.cpp
In file included from ./ext/System/Vector2.cpp:24:
./ext/System/Vector2.hpp:123:39: error: implicit conversion loses integer precision: 'long' to 'int' [-Wshorten-64-to-32]
        return rb_class_new_instance( RARRAY_LEN( anOther ), RARRAY_PTR( anOther ),
                                      ^~~~~~~~~~~~~~~~~~~~~
/Users/jashank/.rvm/rubies/ruby-1.9.3-p392/include/ruby-1.9.1/ruby/ruby.h:711:6: note: instantiated from:
     (long)((RBASIC(a)->flags >> RARRAY_EMBED_LEN_SHIFT) & \
     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ./ext/System/Vector2.cpp:24:
./ext/System/Vector2.hpp:123:39: error: implicit conversion loses integer precision: 'long' to 'int' [-Wshorten-64-to-32]
        return rb_class_new_instance( RARRAY_LEN( anOther ), RARRAY_PTR( anOther ),
                                      ^~~~~~~~~~~~~~~~~~~~~
/Users/jashank/.rvm/rubies/ruby-1.9.3-p392/include/ruby-1.9.1/ruby/ruby.h:713:25: note: instantiated from:
     RARRAY(a)->as.heap.len)
                        ^
2 errors generated.
rake aborted!

And no, GCC doesn't produce this error, but the resultant bundles cause Ruby to reliably segfault.

[Groogy]

I'll fix those but I don't understand why this would cause Ruby to reliably segfault.

[jashank]

Plausibly, the truncation is occurring on values used for addressing, but I don't understand enough about the way YARV's extension handling works to be certain. Also, I don't know why GCC just flags a warning when Clang raises an error. There are cases of this in

• ext/System/Vector2.hpp
• ext/System/Vector3.hpp
• ext/System/main.hpp
• ext/Window/VideoMode.hpp
• ext/Graphics/CircleShape.hpp
• ext/Graphics/Color.hpp
• ext/Graphics/Rect.hpp
• ext/Graphics/Glyph.hpp
• ext/Graphics/RenderStates.hpp
• ext/Graphics/Vertex.hpp
• ext/Graphics/RenderTarget.cpp
• ext/Graphics/Shader.cpp
• ext/Graphics/Transform.cpp
• ext/Graphics/Transformable.cpp
• ext/Graphics/View.cpp
• ext/Audio/Listener.cpp
• ext/Audio/SoundSource.cpp

As for the segfault itself, I can reliably reproduce it; I uploaded a Ruby backtrace and the matching crash log from OS X.

[Groogy]

They are not values that are used for addressing. What it is complaining about is just a simple number that describes how many elements there are in the array, nothing more and nothing less. From the crash log it looked like it crashed when sfml/all are requiring sfml/system

After I've fixed these when I get home from work you should try again and if it still fails you should file it as a new issue.

[Groogy]

Fixed the implicit casting in the Vector files. Can you give me some more warnings?

[jashank]

I ran the build using GCC and got some warnings; here's the build log.